{"meta":{"query_hash":"d263e7495388","filters":{"venue":"BMJ evidence-based medicine"},"cohort_total":125,"direct_labels_cover":115,"predictions_cover":125,"exported":125,"export_cap":100000,"truncated":false,"label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12"},"permalink":"https://metacan.xera.ac/q/d263e7495388","api":"https://metacan.xera.ac/api/v1/cohort?venue=BMJ+evidence-based+medicine"},"results":[{"id":"W2791660821","doi":"10.1136/bmjebm-2017-110852","title":"Assessing the validity of surrogate endpoints in the context of a controversy about the measurement of effectiveness of hepatitis C virus treatment","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Hepatitis C virus research","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"National Health and Medical Research Council","keywords":"Surrogate endpoint; Medicine; Clinical endpoint; Intensive care medicine; Context (archaeology); Endpoint Determination; Randomized controlled trial; Clinical trial; Internal medicine","score_opus":0.2545075038471375,"score_gpt":0.4540353686146979,"score_spread":0.1995278647675604,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2791660821","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9890643,0.0021558164,0.0004916464,0.0035412011,0.00018498991,0.0037028766,0.00003858948,0.000008064625,0.0008125433],"genre_scores_gemma":[0.99789345,0.0013187408,0.00005297632,0.00032416004,0.00016603293,0.00021150471,0.000006850593,0.000018927809,0.000007352857],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99229527,0.003382418,0.0012034976,0.00029637423,0.0024781846,0.000344256],"domain_scores_gemma":[0.9860871,0.010242616,0.00078798004,0.0010904525,0.0017028535,0.00008902744],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0170533,0.00025121417,0.001155924,0.00020759513,0.00008431553,0.0000082788,0.00044878433,0.000091321366,0.00014504648],"category_scores_gemma":[0.010963213,0.000109099536,0.0002179374,0.00070386264,0.0022960308,0.00013344626,0.000050263763,0.00024371086,0.0000033353144],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004599491,0.0010587106,0.6362983,0.0028143201,0.0004949382,0.00007179138,0.005905801,0.000017728424,0.30705327,0.00042024124,0.00037212652,0.04089331],"study_design_scores_gemma":[0.0065632863,0.005791358,0.5083423,0.016809782,0.00041594054,0.0000070517845,0.0012680289,0.0005123401,0.45971626,0.00008261966,0.0004031472,0.00008786864],"about_ca_topic_score_codex":0.03087585,"about_ca_topic_score_gemma":0.0051556868,"teacher_disagreement_score":0.152663,"about_ca_system_score_codex":0.00038427958,"about_ca_system_score_gemma":0.0012809908,"threshold_uncertainty_score":0.99736786},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2792407982","doi":"10.1136/bmjebm-2017-110838","title":"Procalcitonin-guided decision-making results in a significant reduction of antibiotic therapy and hospital stay in neonates with suspected early-onset sepsis","year":2018,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Procalcitonin; Medicine; Sepsis; Antibiotic therapy; Antibiotics; Intensive care medicine; Pediatrics; Randomized controlled trial; Population; Clinical trial; Internal medicine","score_opus":0.1723921777231203,"score_gpt":0.43128935713986133,"score_spread":0.25889717941674106,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2792407982","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.21930687,0.77335536,0.000013560471,0.0015027747,0.00018829637,0.0055364706,0.000025637877,0.00004204471,0.00002895041],"genre_scores_gemma":[0.27085936,0.72677064,0.0016510952,0.000071991795,0.00024015112,0.00025056678,0.00007575848,0.00006724449,0.000013201723],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9946582,0.00043427895,0.0021547184,0.0011920502,0.0010384093,0.00052236905],"domain_scores_gemma":[0.993605,0.0036221426,0.0012147938,0.0008821807,0.00045243851,0.00022342756],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0019733955,0.00076722197,0.003334462,0.0014313946,0.000053456795,0.0000244986,0.00021382904,0.0003861313,0.00006228246],"category_scores_gemma":[0.004403058,0.0004457824,0.00021635552,0.0020202277,0.0006642238,0.00014913888,0.00004292976,0.0004946121,0.000005958958],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005378747,0.0018269693,0.06251527,0.011680863,0.00064023986,0.0016906579,0.001370984,0.000006839006,0.0000135972505,0.000010587808,0.002947786,0.91191745],"study_design_scores_gemma":[0.0125202965,0.015316389,0.11654728,0.8459342,0.0016998171,0.0003097172,0.00048728965,0.000055094864,0.00027489467,0.00006535712,0.006015229,0.00077444676],"about_ca_topic_score_codex":0.00094937114,"about_ca_topic_score_gemma":0.00010307054,"teacher_disagreement_score":0.911143,"about_ca_system_score_codex":0.0005549674,"about_ca_system_score_gemma":0.0013330028,"threshold_uncertainty_score":0.9997994},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2800629548","doi":"10.1136/bmjebm-2018-110943","title":"Screening ECGs in low-risk patients are associated with increased risk of downstream cardiac testing","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Downstream (manufacturing); Internal medicine; Medicine; Cardiology; Risk assessment; Computer science; Engineering; Operations management; Computer security","score_opus":0.4956000741268031,"score_gpt":0.49399743443416266,"score_spread":0.0016026396926404618,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2800629548","genre_codex":"empirical","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7073507,0.0016554747,0.00027264105,0.27670625,0.0019926198,0.009333675,0.001656153,0.00037537716,0.00065713463],"genre_scores_gemma":[0.36149737,0.0009448087,0.003845235,0.6092371,0.01808927,0.0020392858,0.0033094627,0.00055935245,0.0004781252],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9630503,0.02645653,0.0038526186,0.0014483699,0.0033400585,0.0018520898],"domain_scores_gemma":[0.90712965,0.073758274,0.012494164,0.0018767248,0.0042015635,0.0005396396],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.025502356,0.0010629385,0.0031801823,0.0011667942,0.00086482475,0.000018493713,0.00088867365,0.0023948844,0.00043945634],"category_scores_gemma":[0.27423978,0.0008014273,0.00022042985,0.0022757775,0.00094882236,0.0005785691,0.00024035716,0.010624873,0.00009005302],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004917923,0.00010873026,0.74461246,0.0036152662,0.0002039964,0.00016359542,0.00070456986,0.000012929251,0.0000028212305,2.97975e-7,0.2479697,0.0021138596],"study_design_scores_gemma":[0.005224374,0.0018237226,0.7656139,0.21218812,0.0011531239,9.094887e-7,0.0012267003,0.00027918152,0.0000115216835,0.000033943103,0.011659198,0.000785271],"about_ca_topic_score_codex":0.09321496,"about_ca_topic_score_gemma":0.005236866,"teacher_disagreement_score":0.3458533,"about_ca_system_score_codex":0.0014461237,"about_ca_system_score_gemma":0.0047826977,"threshold_uncertainty_score":0.99944365},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2809086820","doi":"10.1136/bmjebm-2018-110979","title":"Essential medicines and the challenges in the Evidence-Based Manifesto","year":2018,"lang":"en","type":"editorial","venue":"BMJ evidence-based medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Manifesto; Data science; Computer science; Political science; Law","score_opus":0.517188658566685,"score_gpt":0.4876869785974194,"score_spread":0.02950167996926556,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2809086820","genre_codex":"commentary","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004351884,0.19992295,0.0005365014,0.5431569,0.25179514,0.003364454,0.00008110753,0.000058623802,0.00064909493],"genre_scores_gemma":[0.015685469,0.049875595,0.00036656804,0.040442936,0.8907281,0.0021358773,0.0001307209,0.00014443236,0.000490261],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98356974,0.0046653645,0.0075647766,0.0018218022,0.0014116608,0.0009666481],"domain_scores_gemma":[0.89228743,0.09828704,0.0059289383,0.0026104099,0.0005508121,0.00033536938],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.14859408,0.0008682379,0.0034451608,0.0009837885,0.00045966494,0.00016022632,0.001986637,0.0010501866,0.0006882606],"category_scores_gemma":[0.19148886,0.0005764094,0.00029355873,0.0005962496,0.0025194318,0.00073923834,0.00013516797,0.0015001247,0.0005385626],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0005763119,0.000062665655,0.0005942755,0.002986876,0.00007553056,0.00002716852,0.0026333914,0.000025747011,4.2440828e-7,0.008756612,0.98393893,0.00032205912],"study_design_scores_gemma":[0.009024735,0.0011500765,0.0023002778,0.039034322,0.00027282658,0.000019300758,0.0033108403,0.003023138,0.0000022048018,0.011165331,0.92968357,0.0010133585],"about_ca_topic_score_codex":0.0047545345,"about_ca_topic_score_gemma":0.0021288313,"teacher_disagreement_score":0.638933,"about_ca_system_score_codex":0.000685247,"about_ca_system_score_gemma":0.0019452905,"threshold_uncertainty_score":0.9996687},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2809195898","doi":"10.1136/bmjebm-2018-110968","title":"Reflections on the history of systematic reviews","year":2018,"lang":"en","type":"editorial","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":81,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Psychological intervention; Systematic review; MEDLINE; Medicine; Value (mathematics); Evidence-based medicine; Alternative medicine; Intensive care medicine; Computer science; Political science; Pathology; Psychiatry","score_opus":0.9165977472242489,"score_gpt":0.6280865975200167,"score_spread":0.2885111497042322,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2809195898","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000037831103,0.09123849,0.0039969436,0.008269978,0.86760736,0.009696753,0.00002418527,0.000014526424,0.019147974],"genre_scores_gemma":[0.00015444125,0.003945023,0.0008806289,0.0027222235,0.778227,0.0025479558,0.000039866773,0.00007143576,0.21141146],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.8589927,0.07433193,0.03590377,0.0022281245,0.028033162,0.0005103022],"domain_scores_gemma":[0.56403065,0.3390217,0.06010138,0.02475624,0.011627343,0.000462685],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.4978683,0.00096590206,0.01584924,0.0012657004,0.00019093238,0.00015902765,0.0061003338,0.0006462663,0.0314949],"category_scores_gemma":[0.88689834,0.00031529635,0.003967654,0.002288128,0.00092070864,0.00019287462,0.00013428931,0.0010407506,0.008437886],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000029204974,0.000035323905,0.000004243936,0.020034969,0.00018243324,0.0000050807143,0.00037857896,0.0000034183165,0.000019657544,0.0003243017,0.97878987,0.00019291781],"study_design_scores_gemma":[0.00016150759,0.0004953168,0.0000034050788,0.12773965,0.0020691534,0.00000266049,0.00019862725,0.00015646475,0.0000029030753,0.00053595175,0.868398,0.00023636047],"about_ca_topic_score_codex":0.00010051875,"about_ca_topic_score_gemma":0.000094144205,"teacher_disagreement_score":0.38903007,"about_ca_system_score_codex":0.00097251934,"about_ca_system_score_gemma":0.0024342444,"threshold_uncertainty_score":0.9999299},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2809788330","doi":"10.1136/bmjebm-2018-111014","title":"Don’t just blame the evidence: considering the role of medical education in the poor uptake of evidence-based medicine in clinical practice","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Blame; Critical appraisal; Evidence-based medicine; Medical education; Curriculum; Manifesto; Quality (philosophy); Psychology; Health care; Evidence-based practice; Medicine; Alternative medicine; Nursing; Pedagogy; Political science; Epistemology; Social psychology","score_opus":0.5347146743972018,"score_gpt":0.6382019501777273,"score_spread":0.10348727578052552,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2809788330","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.26770708,0.017596142,0.00025511155,0.7073939,0.0022572065,0.0039691464,0.0000019057114,0.000021502085,0.00079800666],"genre_scores_gemma":[0.93150634,0.0064223725,0.0005460199,0.055818655,0.0047903084,0.00077782286,0.000005946969,0.000021245642,0.00011127148],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9771532,0.0116124,0.0042479765,0.00070867856,0.0051737474,0.0011040067],"domain_scores_gemma":[0.85423946,0.13790429,0.0023608878,0.0020888774,0.0026931786,0.000713299],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12374071,0.00031862495,0.0008793923,0.0005280445,0.00072055066,0.000017616856,0.0023113422,0.00041690673,0.0026015076],"category_scores_gemma":[0.4055691,0.00013277256,0.00011245955,0.0029592754,0.0051521915,0.00082745077,0.00016946522,0.0024875964,0.00007077067],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004435246,0.0013246392,0.44438028,0.0032986307,0.000050945746,0.00003787352,0.053844906,0.00006791754,0.0011117555,0.004919442,0.31635645,0.17017192],"study_design_scores_gemma":[0.003405536,0.004410367,0.49306634,0.133686,0.00022992592,0.000034094417,0.24909429,0.0125891315,0.00024824176,0.001741876,0.10114942,0.00034477422],"about_ca_topic_score_codex":0.027811067,"about_ca_topic_score_gemma":0.006880348,"teacher_disagreement_score":0.6637993,"about_ca_system_score_codex":0.0003824369,"about_ca_system_score_gemma":0.038525686,"threshold_uncertainty_score":0.9998137},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2837347473","doi":"10.1136/bmjebm-2018-110981","title":"One of the proposed solutions of the EBM Manifesto Educate the public in evidence-based healthcare to make informed decisions","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Manifesto; Health care; Evidence-based medicine; Public relations; Political science; Psychology; Computer science; Medical education; MEDLINE; Medicine; Law","score_opus":0.6084637929433494,"score_gpt":0.5074133762075725,"score_spread":0.10105041673577686,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2837347473","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.15138905,0.0062017096,0.0004526689,0.82618165,0.0027936115,0.012500833,0.000053173168,0.000076488155,0.00035082502],"genre_scores_gemma":[0.964146,0.0003367681,0.0011537946,0.031655155,0.00055716536,0.001828955,0.00001213077,0.000048360704,0.00026167056],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9868884,0.0059786076,0.0031734798,0.0005691173,0.0021890274,0.0012013545],"domain_scores_gemma":[0.96841604,0.019802665,0.0021800348,0.0054513155,0.0036527764,0.0004971634],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0059021055,0.0003935672,0.0008293925,0.0005808113,0.0022140925,0.000014063064,0.0035902394,0.00033711,0.0004041695],"category_scores_gemma":[0.07550803,0.00020743716,0.0002300843,0.004197178,0.0020126333,0.00035776215,0.000985827,0.0016140207,0.00008619118],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0035157972,0.0012829859,0.7727026,0.008238551,0.00012358409,0.0000046189607,0.06527373,0.00031632037,0.0041902955,0.028751953,0.080166176,0.03543342],"study_design_scores_gemma":[0.0045459,0.0021316644,0.7879646,0.15388443,0.00019696636,0.00000708639,0.027278153,0.0012459811,0.0017346373,0.003051794,0.017397286,0.00056154747],"about_ca_topic_score_codex":0.008909661,"about_ca_topic_score_gemma":0.03619408,"teacher_disagreement_score":0.81275696,"about_ca_system_score_codex":0.0012300691,"about_ca_system_score_gemma":0.019369863,"threshold_uncertainty_score":0.9990849},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2846027767","doi":"10.1136/bmjebm-2018-110922","title":"Shared decision is the only outcome that matters when it comes to evaluating evidence-based practice","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":60,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Population; Argument (complex analysis); Psychological intervention; Medicine; Health care; Incidence (geometry); Psychology; Nursing; Political science","score_opus":0.7238961987099377,"score_gpt":0.5691387085368007,"score_spread":0.154757490173137,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2846027767","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02952231,0.005555323,0.05282738,0.90519255,0.0021702514,0.0038571365,0.00008643252,0.00012514122,0.0006634613],"genre_scores_gemma":[0.24694893,0.00014138327,0.044863936,0.7042355,0.0022594312,0.00054578367,0.000022117138,0.00009148074,0.0008914387],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98852056,0.0013929079,0.0064896643,0.0014854592,0.0011630915,0.0009483358],"domain_scores_gemma":[0.951055,0.040234912,0.004650254,0.0024787837,0.0009076117,0.00067344145],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.090679005,0.0005828833,0.0018272522,0.000835729,0.0007577181,0.00028569053,0.0015219676,0.0002545136,0.007111953],"category_scores_gemma":[0.14660013,0.0004917555,0.00028555177,0.0008185568,0.00050859683,0.0022140099,0.00018607388,0.00048708025,0.008651549],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00076811854,0.000101751895,0.13534571,0.00064528687,0.00011101746,0.000010083805,0.005650412,0.00041500235,0.000033435437,0.0019114037,0.85159653,0.0034112246],"study_design_scores_gemma":[0.010141407,0.00661206,0.18260644,0.054440837,0.0005763719,0.00009680137,0.016587844,0.09211999,0.0003331728,0.014168674,0.61907953,0.0032368435],"about_ca_topic_score_codex":0.003973341,"about_ca_topic_score_gemma":0.0005198231,"teacher_disagreement_score":0.232517,"about_ca_system_score_codex":0.001417463,"about_ca_system_score_gemma":0.0012397361,"threshold_uncertainty_score":0.9997534},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2855200104","doi":"10.1136/bmjebm-2018-110972","title":"‘Depletion of the susceptibles’ taught through a story, a table and basic arithmetic","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":42,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Phrase; Arithmetic; Table (database); Psychology; Statistics; Computer science; Natural language processing; Mathematics; Data mining","score_opus":0.22595200651833938,"score_gpt":0.4243785418092606,"score_spread":0.19842653529092122,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2855200104","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014916816,0.0036351627,0.06693821,0.9218978,0.00088837766,0.0036899184,0.00006057312,0.0003867279,0.0010115537],"genre_scores_gemma":[0.0134111615,0.0022761251,0.14332668,0.80982095,0.017700333,0.0016964495,0.0001552734,0.00055454246,0.011058461],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9956594,0.00069767394,0.0009647865,0.0007135736,0.0013984048,0.0005661416],"domain_scores_gemma":[0.99141824,0.0047840676,0.0013293632,0.0017403663,0.0006493857,0.00007857469],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0022455496,0.00062247575,0.0012502141,0.00025182473,0.00017567522,0.000022306023,0.0007671597,0.0007838411,0.0005944971],"category_scores_gemma":[0.010686773,0.00039230715,0.00015685175,0.0006573657,0.0019134827,0.0003090438,0.00018829943,0.0018290598,0.000012117687],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000045521625,0.000030220506,0.0001951014,0.0024408526,0.00004402883,0.00005014249,0.00027029437,6.9208227e-7,0.0013063778,0.0012940608,0.9938762,0.00044646108],"study_design_scores_gemma":[0.0014769841,0.0021004046,0.00018005868,0.057147834,0.0010264975,0.0001366508,0.00022322031,0.0001232521,0.011447509,0.2506336,0.6744613,0.0010427306],"about_ca_topic_score_codex":0.0002332983,"about_ca_topic_score_gemma":0.000057711084,"teacher_disagreement_score":0.31941497,"about_ca_system_score_codex":0.00027191226,"about_ca_system_score_gemma":0.0005506938,"threshold_uncertainty_score":0.9998529},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2884004244","doi":"10.1136/bmjebm-2018-110992","title":"Uncertainty surrounds anticoagulation risks and benefits in patients with chronic kidney disease with atrial fibrillation","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; Institute for Clinical Evaluative Sciences; University of Ottawa","funders":"","keywords":"Medicine; Atrial fibrillation; Stroke (engine); Kidney disease; Observational study; Population; Internal medicine; Intensive care medicine; Cardiology; Disease; Randomized controlled trial; Heart failure","score_opus":0.104221354622322,"score_gpt":0.3482517148296416,"score_spread":0.2440303602073196,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2884004244","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.29016623,0.0024894986,0.00036477123,0.7000045,0.0007358489,0.0059663896,0.00005792964,0.0001439598,0.0000708567],"genre_scores_gemma":[0.81167036,0.00071140757,0.00055179105,0.11800311,0.060093347,0.000027021084,0.0075923875,0.00022226926,0.0011283184],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9955858,0.0002523102,0.00089566124,0.0009145682,0.0018090834,0.0005425523],"domain_scores_gemma":[0.99665016,0.0008870969,0.0007584598,0.00074296026,0.0005673771,0.00039397195],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0008399314,0.0006516587,0.0011941789,0.000791372,0.00013991159,0.00005534897,0.00012668988,0.00042800084,0.0003093252],"category_scores_gemma":[0.0019116404,0.00042389892,0.00014855595,0.0007551099,0.000478633,0.00040519066,0.000051118663,0.0005747865,0.000016552603],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008347124,0.00000476785,0.91583014,0.0014158292,0.00015651475,0.00009114442,0.000035217166,0.0025585853,9.12119e-7,0.00001218748,0.06862906,0.002918545],"study_design_scores_gemma":[0.017251037,0.0033480725,0.8682531,0.018056016,0.0017812484,0.000003521925,0.000006578674,0.0025846784,0.0000020967886,0.000029050881,0.08813268,0.00055193214],"about_ca_topic_score_codex":0.0005216017,"about_ca_topic_score_gemma":0.00009695029,"teacher_disagreement_score":0.5820014,"about_ca_system_score_codex":0.0005633128,"about_ca_system_score_gemma":0.0011557476,"threshold_uncertainty_score":0.9998213},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2885383088","doi":"10.1136/bmjebm-2018-111009","title":"In critically ill children, fluid overload is consistently associated with worse outcomes","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Sepsis Diagnosis and Treatment","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Sinai Health System; Toronto General Hospital; University Health Network","funders":"","keywords":"Critically ill; Intensive care medicine; Medicine; Critical illness","score_opus":0.12533883239647847,"score_gpt":0.38304514254165956,"score_spread":0.2577063101451811,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2885383088","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.04729031,0.004213786,0.000019592342,0.94325507,0.0007161176,0.003468597,0.000120108096,0.0001569781,0.000759458],"genre_scores_gemma":[0.031128597,0.0005467444,0.0004328823,0.95877755,0.0039672335,0.0006686771,0.0008610225,0.00017320485,0.0034440698],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9927762,0.00041024093,0.0015711859,0.001462761,0.0025686268,0.0012109571],"domain_scores_gemma":[0.9910526,0.004161662,0.000593257,0.0016347442,0.0020961445,0.00046153896],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0011981374,0.0011947373,0.0029634463,0.00086456473,0.00011922172,0.00004573731,0.0003878606,0.0011968251,0.0050382707],"category_scores_gemma":[0.013951209,0.0007531195,0.00046117467,0.00083114026,0.0012107054,0.00020791823,0.00006829841,0.0017423442,0.0003161462],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00030491158,0.000442873,0.32495573,0.00024159811,0.00079678086,0.0051108766,0.00009242084,3.9349254e-7,0.000010961435,0.0000063148186,0.6678393,0.00019786396],"study_design_scores_gemma":[0.019960048,0.007812493,0.7654463,0.07036005,0.0056886007,0.00035586758,0.00009022442,0.000038218626,0.0004367314,0.00009472678,0.12835482,0.0013619052],"about_ca_topic_score_codex":0.0011124683,"about_ca_topic_score_gemma":0.0001775674,"teacher_disagreement_score":0.53948444,"about_ca_system_score_codex":0.0009891834,"about_ca_system_score_gemma":0.0012474974,"threshold_uncertainty_score":0.999492},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2887808077","doi":"10.1136/bmjebm-2018-110975","title":"Meta-analysis suggests that metformin may reduce pre-eclampsia compared with insulin use during pregnancy","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Pregnancy and preeclampsia studies","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Metformin; Medicine; Polycystic ovary; Eclampsia; Gestational diabetes; Pregnancy; Obstetrics; Preeclampsia; Type 2 diabetes; Diabetes mellitus; Meta-analysis; Insulin; Insulin resistance; Internal medicine; Endocrinology; Gestation; Biology","score_opus":0.2691340093160504,"score_gpt":0.3723894350958936,"score_spread":0.1032554257798432,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2887808077","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00651748,0.22385427,0.0011366435,0.7572064,0.0007334123,0.008346728,0.00024792668,0.00071169244,0.0012454629],"genre_scores_gemma":[0.34139124,0.022632979,0.019646395,0.5055823,0.01768611,0.009057626,0.004650946,0.00106967,0.07828275],"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","domain_scores_codex":[0.988401,0.0011747811,0.0022450841,0.0025673758,0.0038957421,0.0017160444],"domain_scores_gemma":[0.98794717,0.0038768165,0.0020957643,0.0043476783,0.0011589299,0.000573647],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow","research_integrity"],"category_scores_codex":[0.0016296469,0.0021281098,0.008221743,0.0019361533,0.0005131148,0.00011758566,0.0010984113,0.0013861748,0.0026040575],"category_scores_gemma":[0.0021656232,0.0013183443,0.0023869327,0.0027590229,0.0019187558,0.00088826835,0.00028481593,0.003283826,0.0000979762],"study_design_candidate":"meta_analysis","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.005160904,0.00029815634,0.021616518,0.009857421,0.30638802,0.014546105,0.002043835,0.000099489276,0.00012232354,0.000017737108,0.6394845,0.0003649559],"study_design_scores_gemma":[0.007993829,0.0027209236,0.029092131,0.048711795,0.8229536,0.0010293439,0.0001775799,0.00014098814,0.002861258,0.000093258735,0.08183259,0.002392727],"about_ca_topic_score_codex":0.0009734381,"about_ca_topic_score_gemma":0.00035805572,"teacher_disagreement_score":0.55765194,"about_ca_system_score_codex":0.00046765167,"about_ca_system_score_gemma":0.0010413285,"threshold_uncertainty_score":0.99991024},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2887866346","doi":"10.1136/bmjebm-2018-111040","title":"More than 2 billion pairs of eyeballs: Why aren’t you sharing medical knowledge on Wikipedia?","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Wikis in Education and Collaboration","field":"Social Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"","keywords":"Reading (process); Knowledge translation; Internet privacy; World Wide Web; Public relations; Knowledge sharing; Encyclopedia; Resource (disambiguation); Information Dissemination; Computer science; Medical education; Political science; Library science; Knowledge management; Medicine; Law","score_opus":0.11614870997935471,"score_gpt":0.45045097746982166,"score_spread":0.33430226749046693,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2887866346","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4803763,0.0037095414,0.0030652455,0.43174964,0.009758262,0.0022328957,0.000009810873,0.00040671244,0.06869159],"genre_scores_gemma":[0.98646975,0.00044701286,0.00019787405,0.004389865,0.0069115995,0.00008565325,0.000014773565,0.000018082623,0.0014653964],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9965275,0.0003873924,0.00059201807,0.00042649294,0.001709188,0.00035741724],"domain_scores_gemma":[0.9968806,0.001008123,0.0003086268,0.00048233045,0.0009118203,0.00040849583],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.004630015,0.00018302073,0.00032548016,0.00031108392,0.00035434662,0.000028605584,0.0005918871,0.0002250205,0.0033609867],"category_scores_gemma":[0.014056142,0.00014239362,0.00006670058,0.0013826257,0.0015786991,0.00030065668,0.000043516113,0.00020774707,0.00010811899],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009796361,0.0013426183,0.11781403,0.00053986325,0.00009314148,0.000031571406,0.12203845,0.0000374425,0.0010069945,0.06455932,0.5705924,0.12096451],"study_design_scores_gemma":[0.0048393225,0.005114344,0.058967076,0.028205127,0.00031563028,0.0000057914567,0.09055906,0.0041609583,0.0067026992,0.0052085924,0.7945506,0.0013708025],"about_ca_topic_score_codex":0.0016440307,"about_ca_topic_score_gemma":0.0032966787,"teacher_disagreement_score":0.50609344,"about_ca_system_score_codex":0.0002458562,"about_ca_system_score_gemma":0.001977165,"threshold_uncertainty_score":0.9975501},"labels":[{"model":"gemma","categories":["scholarly_communication","open_science"],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["scholarly_communication"],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2899136888","doi":"10.1136/bmjebm-2018-111089","title":"Understanding Risk for Better Stroke Prevention","year":2018,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Stroke (engine); Medicine; Physical medicine and rehabilitation; Engineering","score_opus":0.24907372318812274,"score_gpt":0.3807631036811311,"score_spread":0.13168938049300835,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2899136888","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00031783246,0.0010584758,0.2391611,0.74860466,0.0017018253,0.0054429006,0.00010018489,0.00021381647,0.003399204],"genre_scores_gemma":[0.0019545865,0.00051184685,0.016551195,0.88202244,0.057346087,0.0013016078,0.0016422671,0.0002807305,0.038389213],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9941997,0.00026148415,0.0013014595,0.0012704771,0.0019663312,0.0010005377],"domain_scores_gemma":[0.9941885,0.0022546272,0.0012586629,0.001600188,0.0004528234,0.0002452103],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.002646302,0.00084198813,0.0015311948,0.00086214,0.00022218912,0.000039410716,0.00048526793,0.0010705278,0.001718638],"category_scores_gemma":[0.004049511,0.000659476,0.00059914077,0.00036126864,0.0006626258,0.00019480966,0.00012121623,0.0018676448,0.00013630473],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00069661223,0.00003912683,0.0015757757,0.002716813,0.0006590728,0.00034594096,0.000059488535,0.000001698015,0.00029042063,0.000011194152,0.99186206,0.0017418106],"study_design_scores_gemma":[0.0062260167,0.0035832154,0.00024097713,0.022098282,0.0056296075,0.000057701505,0.00015648697,0.0006480424,0.0006324371,0.00049165427,0.9596087,0.00062692096],"about_ca_topic_score_codex":0.000067836256,"about_ca_topic_score_gemma":0.000007690693,"teacher_disagreement_score":0.22260992,"about_ca_system_score_codex":0.001500406,"about_ca_system_score_gemma":0.00057811994,"threshold_uncertainty_score":0.9995856},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2901928126","doi":"10.1136/bmjebm-2018-111019","title":"Evidence-based practice education for healthcare professions: an expert view","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":317,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Curriculum; Health care; Stakeholder; Medical education; Professional development; Stakeholder engagement; Best practice; Sociology; Pedagogy; Medicine; Psychology; Political science; Nursing; Public relations","score_opus":0.5722276379167234,"score_gpt":0.6578693978875588,"score_spread":0.08564175997083545,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2901928126","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.012852766,0.029962191,0.01721189,0.9129424,0.010951513,0.014949667,0.000014741721,0.00029964797,0.00081519387],"genre_scores_gemma":[0.4999389,0.007722046,0.048782572,0.38407892,0.032507367,0.023480838,0.0002554788,0.00012956298,0.0031042837],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98877037,0.0049901656,0.0017260514,0.0010714704,0.0017888985,0.0016530659],"domain_scores_gemma":[0.970828,0.018647298,0.0010522805,0.0014864613,0.0061271284,0.0018588095],"candidate_categories":["metaresearch","metaepi_narrow","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.020166157,0.00038751363,0.00064907776,0.0006079188,0.0031797702,0.00003706049,0.0008622458,0.0003962928,0.0021121772],"category_scores_gemma":[0.093521275,0.0002787618,0.00009256976,0.001684871,0.0007924275,0.0026070199,0.000067775014,0.00093106274,0.0005133643],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0033744841,0.0007681067,0.009323052,0.0060680765,0.000015182853,0.0000043641585,0.0119618485,0.000004334921,0.0014292848,0.0022514428,0.78127915,0.18352066],"study_design_scores_gemma":[0.0029064089,0.01141549,0.013464501,0.1343673,0.00010173124,0.000009280474,0.053215206,0.0051308814,0.00060753693,0.0018141724,0.7761946,0.0007728825],"about_ca_topic_score_codex":0.0069787423,"about_ca_topic_score_gemma":0.0009931956,"teacher_disagreement_score":0.52886343,"about_ca_system_score_codex":0.0012100963,"about_ca_system_score_gemma":0.043853972,"threshold_uncertainty_score":0.99996644},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2905106223","doi":"10.1136/bmjebm-2018-111121","title":"Drug discovery today: no molecules required","year":2018,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Chemical and Physical Studies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Drug discovery; Drug; Computer science; Medicine; Computational biology; Data science; Pharmacology; Bioinformatics; Biology","score_opus":0.03348223847339627,"score_gpt":0.318468529405376,"score_spread":0.28498629093197975,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2905106223","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.94922,0.0069435867,0.0050755483,0.029135313,0.00090733607,0.00065851,0.000022728504,0.00006681391,0.007970167],"genre_scores_gemma":[0.9862801,0.00035255688,0.00042063888,0.0047525223,0.005036925,0.000054115135,0.000059737697,0.000019763585,0.003023659],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.99864095,0.00006438916,0.0002765103,0.00042646838,0.00029883918,0.00029285974],"domain_scores_gemma":[0.9989853,0.00013213854,0.000094832205,0.00042532693,0.00022469567,0.0001377107],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00027919988,0.00020691236,0.0002650062,0.000022205742,0.00010487899,0.000014108923,0.00023384688,0.000075418946,0.000111641064],"category_scores_gemma":[0.0015046393,0.00014160067,0.00010086536,0.00014200198,0.0006839795,0.000011250219,0.00013272627,0.00008333229,0.000111683104],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004808766,0.0000568271,0.0009655011,0.00006514994,0.000044067525,0.000007643751,0.00003669691,0.000001993602,0.86009705,0.00018752628,0.13623631,0.0018203296],"study_design_scores_gemma":[0.0011241194,0.0012293869,0.0015442093,0.00107805,0.0000695518,0.0000047969397,0.000051203904,0.00008922312,0.8604413,0.0007800916,0.13321792,0.00037014965],"about_ca_topic_score_codex":0.00010265917,"about_ca_topic_score_gemma":0.00002130692,"teacher_disagreement_score":0.037060086,"about_ca_system_score_codex":0.000019025783,"about_ca_system_score_gemma":0.00008847146,"threshold_uncertainty_score":0.5774308},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2911416111","doi":"10.1136/bmjebm-2018-111069","title":"Causal inference for clinicians","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":34,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Jewish General Hospital","funders":"","keywords":"Inference; Causal inference; Psychology; Computer science; Artificial intelligence; Econometrics; Mathematics","score_opus":0.38310932339326553,"score_gpt":0.5368590741435255,"score_spread":0.15374975075025998,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2911416111","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.06182809,0.0006471072,0.907122,0.018764209,0.001098521,0.006100208,0.000025563117,0.0011780029,0.0032362845],"genre_scores_gemma":[0.8122418,0.00017710114,0.17842782,0.0046340786,0.0009998265,0.00090931804,0.00002463204,0.00009867409,0.0024867489],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99726003,0.00013580067,0.0009326602,0.000532308,0.0006118934,0.0005272892],"domain_scores_gemma":[0.9847475,0.013070645,0.00047348117,0.001016879,0.00050100713,0.00019044743],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0025014584,0.00034193037,0.0007692828,0.0002030403,0.000073682604,0.000017086964,0.00048044682,0.00020654273,0.00092239474],"category_scores_gemma":[0.022172503,0.00026197097,0.00013067221,0.0003309873,0.0002545741,0.00038117904,0.00006178846,0.0003216556,0.00012596506],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017396545,0.00051348284,0.034875836,0.0043877535,0.00016023061,0.000081535996,0.0011179311,0.00017833029,0.045344085,0.7144293,0.14765911,0.04951271],"study_design_scores_gemma":[0.0042881025,0.006458261,0.0015504012,0.01726535,0.00033256348,0.000020932426,0.00061019225,0.0044600773,0.031403635,0.9084217,0.023936585,0.0012521981],"about_ca_topic_score_codex":0.000050150524,"about_ca_topic_score_gemma":0.00003784749,"teacher_disagreement_score":0.7504137,"about_ca_system_score_codex":0.00014736774,"about_ca_system_score_gemma":0.00040754306,"threshold_uncertainty_score":0.9999909},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2911968494","doi":"10.1136/bmjebm-2018-111057","title":"Use of modified intention-to-treat analysis in studies of direct oral anticoagulants and risk of selection bias: a systematic review","year":2019,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Atrial Fibrillation Management and Outcomes","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre Hospitalier de l’Université de Montréal","funders":"Fonds de Recherche du Québec - Santé","keywords":"Selection (genetic algorithm); Risk analysis (engineering); Selection bias; Medicine; Intensive care medicine; Psychology; Actuarial science; Computer science; Business; Artificial intelligence; Pathology","score_opus":0.5686784527537067,"score_gpt":0.5045638600117248,"score_spread":0.06411459274198184,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2911968494","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0017711039,0.9899677,0.0002542753,0.0002827995,0.000103988976,0.0075679505,0.000023436567,0.00001712889,0.000011585111],"genre_scores_gemma":[0.00520394,0.99383104,0.00031394305,0.000055806286,0.00006044392,0.000059235594,0.00005837406,0.00002461369,0.00039261614],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.993974,0.0012847815,0.0032992042,0.0004456423,0.0008087247,0.0001876828],"domain_scores_gemma":[0.9904341,0.0049069775,0.0032188185,0.00065528613,0.0006900176,0.00009481291],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0044198935,0.00043263214,0.011644305,0.0022614459,0.000019716592,0.000004655714,0.00012474672,0.0001436813,0.000033395067],"category_scores_gemma":[0.027442748,0.00025424123,0.0008559121,0.003703985,0.0001633839,0.00012616735,0.00006659931,0.00015940139,0.0000033171343],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00015011287,0.000020657058,0.056460988,0.92883676,0.0045141103,0.0000055881364,0.0000613212,0.000055253004,0.0000049935106,0.000010958071,0.00016947444,0.009709801],"study_design_scores_gemma":[0.00065635185,0.00054553867,0.0026127666,0.8563666,0.1388536,0.000004195192,0.00006500849,0.00032466894,0.000008871146,0.0000023052323,0.0003968511,0.0001632725],"about_ca_topic_score_codex":0.0008178712,"about_ca_topic_score_gemma":0.000105957544,"teacher_disagreement_score":0.13433948,"about_ca_system_score_codex":0.00012984367,"about_ca_system_score_gemma":0.00019800136,"threshold_uncertainty_score":0.999991},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaepi_narrow","metaepi_broad","metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2925846695","doi":"10.1136/bmjebm-2019-111174","title":"Assessing assumptions for statistical analyses in randomised clinical trials","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":53,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Novo Nordisk Fonden","keywords":"Statistical power; Statistical hypothesis testing; Clinical trial; Statistical analysis; Computer science; Statistical model; Econometrics; Medical physics; Management science; Data science; Medicine; Statistics; Machine learning; Mathematics; Engineering","score_opus":0.9827549661353883,"score_gpt":0.784218247525338,"score_spread":0.1985367186100503,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2925846695","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07567447,0.0033540193,0.88545984,0.020596383,0.0019819136,0.011734404,0.00033873395,0.00001682122,0.0008434198],"genre_scores_gemma":[0.9425531,0.00014093137,0.04856948,0.0038535146,0.0014924651,0.00092221994,0.00039697776,0.000029181972,0.002042103],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.8520341,0.09455587,0.042358875,0.0020644057,0.008381313,0.0006054457],"domain_scores_gemma":[0.4141476,0.56547076,0.013002444,0.0046570282,0.0021777337,0.00054442824],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.65141046,0.00037273802,0.010696663,0.0009201188,0.000097782,0.00057066215,0.0012730458,0.00018543354,0.030496651],"category_scores_gemma":[0.83608127,0.00015932374,0.0026041502,0.0016909953,0.00022733954,0.00057373976,0.0000488186,0.00030267198,0.0012670431],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021344938,0.00027358893,0.12379715,0.0004479366,0.00051742356,0.000033091652,0.00010700229,0.0013078196,0.0013966179,0.0015962704,0.79740274,0.07098584],"study_design_scores_gemma":[0.08785263,0.0017689711,0.35320953,0.0095987,0.007833742,0.000021978296,0.0036728256,0.339555,0.00013912705,0.023018833,0.1719331,0.0013955523],"about_ca_topic_score_codex":0.00005494031,"about_ca_topic_score_gemma":0.00004675775,"teacher_disagreement_score":0.8668786,"about_ca_system_score_codex":0.00006789946,"about_ca_system_score_gemma":0.0006219662,"threshold_uncertainty_score":0.9995106},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2935997213","doi":"10.1136/bmjebm-2019-111193","title":"Guidelines do not self-implement: time for a research paradigm shift from massive creation to effective implementation in evidence-based medicine research in China","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"European Society of Cardiology","keywords":"Guideline; Context (archaeology); Quality (philosophy); China; Paradigm shift; Engineering ethics; Chen; Clinical Practice; Quality management; Computer science; Psychology; Medical education; Management science; Medicine; Political science; Nursing; Engineering; Epistemology; Operations management","score_opus":0.5299034806332391,"score_gpt":0.6399346695960708,"score_spread":0.1100311889628317,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2935997213","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4792701,0.0015111476,0.0039596036,0.4893835,0.0004913461,0.025066102,0.00008296782,0.0001137261,0.00012151095],"genre_scores_gemma":[0.97196406,0.0004670974,0.0076523423,0.009648205,0.0029701116,0.0060378984,0.0010039282,0.000135279,0.00012109252],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9807048,0.0047962586,0.005111491,0.0022120199,0.005344219,0.0018312269],"domain_scores_gemma":[0.9273381,0.06650263,0.00091921375,0.0016622858,0.0028010942,0.0007766383],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06511396,0.00069771707,0.0018488003,0.003946222,0.0002768219,0.00009037941,0.00073888106,0.00038239214,0.0042182957],"category_scores_gemma":[0.089584336,0.0005697071,0.00020905475,0.004745918,0.0004016961,0.001238383,0.00023292612,0.0015693932,0.0004633099],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.078556314,0.0019090527,0.5156892,0.004332963,0.0004999754,0.00058918534,0.011795785,0.0031615715,0.061267175,0.0029139004,0.1554365,0.16384839],"study_design_scores_gemma":[0.07640855,0.033680398,0.7258802,0.07670777,0.0008304742,0.000015929523,0.00816484,0.036336813,0.009953589,0.012297655,0.018399298,0.0013244807],"about_ca_topic_score_codex":0.029054498,"about_ca_topic_score_gemma":0.0035199148,"teacher_disagreement_score":0.49269393,"about_ca_system_score_codex":0.0032521738,"about_ca_system_score_gemma":0.0031589475,"threshold_uncertainty_score":0.99967545},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2942870102","doi":"10.1136/bmjebm-2019-111186","title":"Training curriculum to help patient representatives participate meaningfully in the development of clinical practice guidelines","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary","funders":"","keywords":"Guideline; Curriculum; Medical education; Grading (engineering); Likert scale; Medicine; Curriculum development; Face validity; Psychology; Scale (ratio); Readability; Pedagogy; Engineering; Computer science","score_opus":0.787443666098379,"score_gpt":0.6554227771389366,"score_spread":0.13202088895944242,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2942870102","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9417729,0.000715053,0.0009222085,0.04506941,0.0020619181,0.0060951766,0.0000047701583,0.00003053351,0.003328028],"genre_scores_gemma":[0.8961114,0.0000962078,0.04237852,0.05920048,0.000639733,0.0013320162,0.000045535537,0.00002317483,0.00017293135],"study_design_codex":"observational","study_design_gemma":"qualitative","domain_scores_codex":[0.98928684,0.00327385,0.0046937596,0.00051942427,0.0015087742,0.00071736594],"domain_scores_gemma":[0.98745054,0.009032088,0.0014323373,0.00063538126,0.0010824403,0.00036722812],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.022334272,0.00023700246,0.0007406704,0.00018699738,0.0002548312,0.000003940413,0.00037480603,0.00011935599,0.0004961793],"category_scores_gemma":[0.022463007,0.00013953212,0.00007161838,0.00062856573,0.00009186492,0.00021471754,0.00011861756,0.00060280995,0.0002549595],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002169866,0.000757034,0.6152365,0.002847183,0.000053748307,0.00004013034,0.23982377,0.00037536238,0.0002022389,0.00059684535,0.025325727,0.11257158],"study_design_scores_gemma":[0.00660765,0.007430283,0.090731576,0.07412777,0.0001346584,0.0000038217386,0.6105456,0.0019696797,0.00044783627,0.00023108878,0.20718114,0.0005889016],"about_ca_topic_score_codex":0.00082294416,"about_ca_topic_score_gemma":0.0003376539,"teacher_disagreement_score":0.52450496,"about_ca_system_score_codex":0.0001484333,"about_ca_system_score_gemma":0.0018719787,"threshold_uncertainty_score":0.9857712},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"nonrandomized_trial","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2951259638","doi":"10.1136/bmjebm-2019-111212","title":"Evidence from large Danish cohort does not support an association between the MMR vaccine and autism: facts in a post-truth world","year":2019,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Vaccine Coverage and Hesitancy","field":"Social Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Danish; Autism; Association (psychology); MMR vaccine; Cohort; Cohort study; Medicine; Psychology; Virology; Psychiatry; Vaccination; Philosophy; Rubella; Internal medicine; Linguistics; Psychotherapist","score_opus":0.05852519767125543,"score_gpt":0.34461354752757245,"score_spread":0.28608834985631704,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2951259638","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12964967,0.0023360287,0.000073683,0.86274856,0.0016377897,0.0027670288,0.0002664937,0.00012735315,0.0003933962],"genre_scores_gemma":[0.5372131,0.003137595,0.00008539347,0.43538183,0.015051153,0.00019449515,0.0007854036,0.00008897348,0.008062062],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9909528,0.0021765286,0.0013719024,0.0013389058,0.0028695608,0.0012902938],"domain_scores_gemma":[0.97964936,0.017103491,0.0013089228,0.0011691807,0.0004915826,0.00027746067],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0098966425,0.00068731816,0.0015910246,0.000541561,0.00056653825,0.000270837,0.0012984243,0.0010519773,0.0027714137],"category_scores_gemma":[0.014372725,0.00044544894,0.0001814885,0.00090972014,0.0000885137,0.0015995253,0.00016961677,0.0024558112,0.000118154145],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008741195,0.000021100663,0.6518509,0.00012363598,0.00010435959,0.00024053041,0.0024873752,8.304691e-7,0.00003697704,0.000031863867,0.34425247,0.0007625473],"study_design_scores_gemma":[0.0014037577,0.0004237657,0.84042835,0.006935035,0.0008320549,0.0000014969492,0.00057047105,0.000019101773,0.00005170181,0.00047653032,0.14819747,0.0006602518],"about_ca_topic_score_codex":0.023882799,"about_ca_topic_score_gemma":0.032529548,"teacher_disagreement_score":0.42736673,"about_ca_system_score_codex":0.0010935338,"about_ca_system_score_gemma":0.0020288385,"threshold_uncertainty_score":0.99984556},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W2966910029","doi":"10.1136/bmjebm-2019-111236","title":"Treating postoperative pain? Avoid tramadol, long-acting opioid analgesics and long-term use","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Opioid Use Disorder Treatment","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"National Institute for Health and Care Research","keywords":"Tramadol; Medicine; Opioid; Anesthesia; Term (time); Pain ladder; Analgesic; Internal medicine","score_opus":0.058941608157173385,"score_gpt":0.3453892618272062,"score_spread":0.28644765367003283,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2966910029","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9728967,0.007295283,0.002111668,0.0136024365,0.0003348034,0.0033013243,0.000009794239,0.0001863892,0.0002616305],"genre_scores_gemma":[0.9894546,0.0011155771,0.0016483286,0.0055564316,0.00046917368,0.00015477881,0.00016510775,0.00009379068,0.0013421562],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99614793,0.0004485963,0.00087245455,0.0009295439,0.0009262774,0.0006752191],"domain_scores_gemma":[0.99491185,0.0028800613,0.00037252987,0.0009085751,0.000399722,0.000527264],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.001667435,0.0006166208,0.0011185186,0.0003535314,0.00020433248,0.00007960126,0.00016562911,0.00020529509,0.00072365487],"category_scores_gemma":[0.0039332,0.00045600312,0.00015981946,0.0005434999,0.00028818974,0.0006044596,0.00006262353,0.00051011646,0.000077128694],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00096026313,0.0003074648,0.9520652,0.0013982387,0.00020807462,0.00063679356,0.0026681663,0.00003345214,0.004148213,0.000039505652,0.0005365755,0.036998034],"study_design_scores_gemma":[0.009676652,0.0065886583,0.95024776,0.023665093,0.0009203509,0.00019319751,0.0023236915,0.0021155297,0.003364244,0.000019311694,0.00022732298,0.00065815856],"about_ca_topic_score_codex":0.0005155424,"about_ca_topic_score_gemma":0.00008804827,"teacher_disagreement_score":0.036339875,"about_ca_system_score_codex":0.00035346116,"about_ca_system_score_gemma":0.000610561,"threshold_uncertainty_score":0.9997892},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2969601771","doi":"10.1136/bmjebm-2019-111225","title":"Challenges in interpreting results from ‘multiple regression’ when there is interaction between covariates","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; McGill University; Jewish General Hospital","funders":"","keywords":"Covariate; Observational study; Outcome (game theory); Popularity; Meta-regression; Regression; Population; Econometrics; Regression analysis; Statistics; Regression toward the mean; Pace; Baseline (sea); Medicine; Psychology; Meta-analysis; Mathematics; Geography; Social psychology; Internal medicine","score_opus":0.4058680440234082,"score_gpt":0.4765542601293872,"score_spread":0.07068621610597897,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2969601771","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.82493377,0.007345703,0.041313242,0.109360084,0.00122254,0.004709174,0.000117770745,0.0013971833,0.00960053],"genre_scores_gemma":[0.9612014,0.0007079788,0.036707304,0.0005507673,0.0004363594,0.000105191495,0.00004208812,0.000051536146,0.00019735683],"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","domain_scores_codex":[0.99681157,0.0004028552,0.0011316132,0.0007002651,0.000589856,0.00036381546],"domain_scores_gemma":[0.98108524,0.016841898,0.00074097834,0.001008542,0.00020410243,0.00011924918],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0024196282,0.0003600949,0.00078357203,0.0002774358,0.000043256183,0.000020987338,0.00044445574,0.00024454188,0.00063194335],"category_scores_gemma":[0.018292228,0.00025745152,0.00007915543,0.00019858583,0.00010725446,0.00059465045,0.00012956376,0.00060536474,0.000086890366],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010511243,0.0007638501,0.334231,0.0031617326,0.0004765046,0.00034080422,0.06757751,0.00013622353,0.08146612,0.012424841,0.062000524,0.42690966],"study_design_scores_gemma":[0.00886815,0.002479412,0.019818567,0.20964207,0.00024806484,0.000012609002,0.012941478,0.0078865485,0.13955995,0.5905202,0.00642175,0.0016012044],"about_ca_topic_score_codex":0.00077532785,"about_ca_topic_score_gemma":0.00022037746,"teacher_disagreement_score":0.5780954,"about_ca_system_score_codex":0.00026441735,"about_ca_system_score_gemma":0.00009915672,"threshold_uncertainty_score":0.9999878},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2981608944","doi":"10.1136/bmjebm-2019-111273","title":"Challenges facing early-career and mid-career researchers: potential solutions to safeguard the future of evidence-based medicine","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":32,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"National Institute for Health and Care Research","keywords":"Globe; Attendance; Career development; Career Pathways; General partnership; Public relations; Medical education; Psychology; Management; Political science; Medicine","score_opus":0.6353307993045753,"score_gpt":0.534381934536827,"score_spread":0.10094886476774823,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2981608944","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.2265831,0.050422776,0.0012196746,0.71286803,0.0026334298,0.0052944445,0.000014263952,0.00007064615,0.0008936542],"genre_scores_gemma":[0.97921467,0.00742354,0.0005035803,0.006093552,0.004683883,0.00064230454,0.000015410506,0.00003730603,0.0013857794],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.989946,0.002960129,0.001329352,0.00087226793,0.0031743655,0.0017179037],"domain_scores_gemma":[0.98525053,0.009722828,0.0004639832,0.0013520386,0.0018709551,0.0013396835],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02214429,0.0003572167,0.0007873347,0.00076710025,0.0012015641,0.000021510032,0.0010011867,0.0003324717,0.002619265],"category_scores_gemma":[0.015686834,0.00019188471,0.00009713304,0.0015329302,0.0011787419,0.0006477709,0.00024207949,0.0014088275,0.00029934052],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0066375355,0.00034931747,0.2015357,0.01800258,0.00021992018,0.00009685228,0.074066974,0.0012452212,0.032956358,0.0032016952,0.42527863,0.23640922],"study_design_scores_gemma":[0.0058303303,0.008597729,0.7054395,0.08417825,0.00021429521,0.000018196073,0.11326703,0.0034475212,0.00047286804,0.0009583603,0.07676794,0.00080796145],"about_ca_topic_score_codex":0.0045812135,"about_ca_topic_score_gemma":0.0008243399,"teacher_disagreement_score":0.75263155,"about_ca_system_score_codex":0.000453102,"about_ca_system_score_gemma":0.0043872297,"threshold_uncertainty_score":0.99829245},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"incentives","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2990740094","doi":"10.1136/bmjebm-2019-111272","title":"Building capacity in evidence-based medicine in low-income and middle-income countries: problems and potential solutions","year":2019,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"National Institute for Health and Care Research","keywords":"Low and middle income countries; Income distribution; Middle income; Development economics; Economics; Demographic economics; Developing country; Economic growth; Mathematics; Inequality","score_opus":0.2712587169033339,"score_gpt":0.46077319020027663,"score_spread":0.1895144732969427,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2990740094","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8848535,0.0068479083,0.0011882345,0.10114313,0.0014401474,0.0043452173,0.000008450108,0.00005943325,0.000113997354],"genre_scores_gemma":[0.99176574,0.00233849,0.00088224077,0.0031957822,0.0009295808,0.0006422365,0.000008832681,0.000026471638,0.00021062807],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9922779,0.001760873,0.0018066571,0.0009151775,0.0015668975,0.001672447],"domain_scores_gemma":[0.988663,0.008694844,0.00051667274,0.0006497645,0.00053723296,0.0009384628],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.018365102,0.00037229888,0.0009287625,0.0013640401,0.000702323,0.000026142132,0.00043636785,0.00033706255,0.0011655091],"category_scores_gemma":[0.013599607,0.000277599,0.000038797112,0.0018134285,0.0012949769,0.0010743667,0.00013810294,0.001348432,0.000082611936],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065756415,0.000081401995,0.97629035,0.014071746,0.000008660743,0.000049039652,0.0047971457,0.0006657092,0.0012823145,0.0007544332,0.0008716058,0.00047000594],"study_design_scores_gemma":[0.006697354,0.0014758715,0.74170375,0.21603234,0.00003147928,0.000016675047,0.002787903,0.02875029,0.000038542876,0.0013358984,0.0007219985,0.0004079154],"about_ca_topic_score_codex":0.018491555,"about_ca_topic_score_gemma":0.0025014516,"teacher_disagreement_score":0.23458664,"about_ca_system_score_codex":0.0010152842,"about_ca_system_score_gemma":0.0035249407,"threshold_uncertainty_score":0.99996763},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"incentives","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2993142824","doi":"10.1136/bmjebm-2019-111300","title":"The need for clinical judgement in the application of evidence-based medicine","year":2019,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Robarts Clinical Trials; Western University","funders":"","keywords":"Judgement; Evidence-based medicine; Medicine; Intensive care medicine; MEDLINE; Alternative medicine; Psychotherapist; Psychology; Epistemology; Pathology","score_opus":0.9715138239990201,"score_gpt":0.7161295285443602,"score_spread":0.2553842954546599,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W2993142824","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000038894896,0.8704239,0.05103543,0.057535358,0.0013008765,0.019436492,0.000017531785,0.000008619724,0.00023790574],"genre_scores_gemma":[0.0002950188,0.98580056,0.00095027394,0.004218432,0.0020460822,0.004878618,0.000075552394,0.00005402694,0.0016814367],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.86598647,0.059894253,0.050942693,0.0033213831,0.018878452,0.0009767798],"domain_scores_gemma":[0.5949908,0.35760707,0.029929755,0.0136392815,0.003528343,0.00030475378],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.55339503,0.0011591556,0.018084874,0.0011779751,0.0002792509,0.00023744059,0.00953487,0.00052596605,0.0013950111],"category_scores_gemma":[0.43586642,0.00032131193,0.005459218,0.0053387284,0.001253553,0.0002861917,0.00017631351,0.0009120569,0.0007813229],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008126811,0.000072641764,0.00051506876,0.011048854,0.00020977708,0.0000031069467,0.00008087335,0.000026867305,5.47101e-7,0.0011371328,0.21122016,0.7756037],"study_design_scores_gemma":[0.0008746986,0.0008419699,0.00022406885,0.0852736,0.005338903,0.0000058597334,0.00042780372,0.0035097864,3.0674119e-7,0.0007357417,0.9024619,0.0003053897],"about_ca_topic_score_codex":0.00012523362,"about_ca_topic_score_gemma":0.000090157526,"teacher_disagreement_score":0.7752983,"about_ca_system_score_codex":0.00021678525,"about_ca_system_score_gemma":0.0021654845,"threshold_uncertainty_score":0.99999666},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3003086189","doi":"10.1136/bmjebm-2019-111268","title":"Assessment of assumptions of statistical analysis methods in randomised clinical trials: the what and how","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":68,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Novo Nordisk Fonden","keywords":"Statistical analysis; Clinical trial; Statistics; Medical physics; Medicine; Econometrics; Mathematics; Internal medicine","score_opus":0.9498731355574201,"score_gpt":0.7462174734380085,"score_spread":0.20365566211941166,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3003086189","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.011281768,0.0075599225,0.8924442,0.08633724,0.00017659135,0.002086048,0.000013457289,0.0000022876152,0.00009846483],"genre_scores_gemma":[0.84452814,0.0016861376,0.15147743,0.0017854845,0.00022719246,0.00014299898,0.000009373009,0.000007835277,0.0001354296],"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.6728549,0.28036153,0.034818433,0.0017560484,0.0098326905,0.00037642615],"domain_scores_gemma":[0.4428292,0.5352082,0.016156463,0.0034726008,0.00182777,0.00050578616],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6434263,0.00024537783,0.01022185,0.0005542534,0.000048365033,0.00019910415,0.0010332296,0.00010219034,0.005867246],"category_scores_gemma":[0.71231323,0.00008922105,0.001955482,0.0037189256,0.0006130015,0.00029242114,0.000085216685,0.000292823,0.000008487899],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0016351249,0.00029714682,0.5479127,0.0008495338,0.0054983436,0.00002875154,0.0013945859,0.0020406235,0.0017481401,0.006453751,0.057937756,0.37420353],"study_design_scores_gemma":[0.012297173,0.0007945096,0.470051,0.0015202931,0.017384961,0.000002113488,0.0063650194,0.483827,0.00007667261,0.0021542103,0.005281995,0.00024505108],"about_ca_topic_score_codex":0.000025971725,"about_ca_topic_score_gemma":0.00003270548,"teacher_disagreement_score":0.83324635,"about_ca_system_score_codex":0.000018691479,"about_ca_system_score_gemma":0.0003366075,"threshold_uncertainty_score":0.99504155},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3009190329","doi":"10.1136/bmjebm-2019-111312","title":"Ibuprofen for acute pain relief in the emergency department","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Anesthesia and Pain Management","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Ibuprofen; Medicine; Emergency department; Analgesic; Antipyretic; Acute pain; Adverse effect; Anesthesia; Medical prescription; Analgesic agents; Acetaminophen; Emergency medicine; Intensive care medicine; Internal medicine; Pharmacology","score_opus":0.10219635562718302,"score_gpt":0.37857842951306253,"score_spread":0.27638207388587954,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3009190329","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023364948,0.0033979265,0.039144732,0.92679775,0.00012620557,0.006523736,0.0000018731447,0.000100007994,0.0005428041],"genre_scores_gemma":[0.65960497,0.0016104396,0.0037293392,0.3296362,0.0020463502,0.002190241,0.00016550257,0.000054102427,0.00096284517],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99741614,0.00045872285,0.0006434268,0.00040493318,0.0006904562,0.00038633638],"domain_scores_gemma":[0.99789304,0.0011208042,0.00016040057,0.0005015563,0.00010677464,0.00021741814],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004899734,0.00024010769,0.00048271762,0.00010043141,0.000087303626,0.000006751286,0.00030779437,0.0000781495,0.00047125228],"category_scores_gemma":[0.002506641,0.00014012719,0.000142398,0.0005252091,0.00009477516,0.00007410805,0.000018340881,0.00020995144,0.000048355047],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0025448678,0.0002287378,0.0987883,0.0012517278,0.00016547614,0.00069275853,0.0020993191,0.000048759342,0.00057923805,0.0014620665,0.88055676,0.011581969],"study_design_scores_gemma":[0.009239517,0.013710007,0.112074316,0.0052174693,0.0020139744,0.000042140517,0.0033723442,0.01479386,0.00034314027,0.0010438522,0.8375379,0.0006114751],"about_ca_topic_score_codex":0.000047215985,"about_ca_topic_score_gemma":0.0000123276495,"teacher_disagreement_score":0.63624,"about_ca_system_score_codex":0.00006747544,"about_ca_system_score_gemma":0.00016317322,"threshold_uncertainty_score":0.5714221},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3011157861","doi":"10.1136/bmjebm-2019-111331","title":"Adjuvant-containing control arms in pivotal quadrivalent human papillomavirus vaccine trials: restoration of previously unpublished methodology","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Cervical Cancer and HPV Research","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"U.S. Food and Drug Administration; National Institute for Health and Care Research; Laura and John Arnold Foundation; European Respiratory Society; American Association of Colleges of Pharmacy","keywords":"Clinical trial; Medicine; Blinding; Family medicine; Informed consent; Placebo; Alternative medicine; Internal medicine; Pathology","score_opus":0.5405509864717235,"score_gpt":0.5166450541185875,"score_spread":0.023905932353135984,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3011157861","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.56625146,0.041249335,0.03967319,0.33745068,0.0009081879,0.012120694,0.00003603736,0.0003232374,0.001987143],"genre_scores_gemma":[0.9892677,0.00050520216,0.0011484092,0.0067533557,0.0018664235,0.00029301914,0.00007321462,0.000044909913,0.000047781275],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99168825,0.00300873,0.0025392286,0.00074135297,0.0013863298,0.0006361154],"domain_scores_gemma":[0.990784,0.0062958077,0.0007933067,0.0006364882,0.00078769797,0.0007027062],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.015478769,0.00037330645,0.0029029136,0.0005383274,0.00006992899,0.000019413881,0.0003505746,0.00025051605,0.0042243153],"category_scores_gemma":[0.05733848,0.00026636486,0.00028287328,0.0010651809,0.00025377455,0.0002975286,0.000067087,0.0008142403,0.00001533989],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.1497984,0.0010217649,0.36812747,0.008237397,0.00069133047,0.0014632972,0.008456782,0.0007677126,0.23246169,0.0020103042,0.02038618,0.20657769],"study_design_scores_gemma":[0.14093357,0.058890324,0.7440165,0.0109835835,0.0024332707,0.00008638735,0.004345104,0.01078572,0.020266348,0.0012863552,0.0048063514,0.0011664852],"about_ca_topic_score_codex":0.0016351249,"about_ca_topic_score_gemma":0.0004572315,"teacher_disagreement_score":0.4230162,"about_ca_system_score_codex":0.0003153449,"about_ca_system_score_gemma":0.0010055259,"threshold_uncertainty_score":0.99997884},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3027820216","doi":"10.1136/bmjebm-2020-111371","title":"Problem with patient decision aids","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval","funders":"","keywords":"Decision aids; Certainty; Presentation (obstetrics); Health care; Decision support system; Medicine; Decision quality; Decision analysis; Value (mathematics); Quality (philosophy); Medical emergency; Patient satisfaction; Nursing; Computer science; Alternative medicine; Artificial intelligence; Surgery","score_opus":0.38599478060132236,"score_gpt":0.47030998431235405,"score_spread":0.08431520371103168,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3027820216","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.17790078,0.024318047,0.02882199,0.7401028,0.0017312532,0.0185566,0.000042517368,0.0012675184,0.007258448],"genre_scores_gemma":[0.90263116,0.00024766757,0.03980089,0.0548984,0.0006767155,0.001552653,0.000045445806,0.00006234642,0.00008472855],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9945966,0.0014090048,0.0014644051,0.0005919122,0.0012932399,0.00064486556],"domain_scores_gemma":[0.9906505,0.0054996517,0.0007872405,0.0012889684,0.0009895368,0.0007841325],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0007805927,0.0003124363,0.0005946803,0.0001704879,0.0007077033,0.0000069386533,0.0006362555,0.00020207571,0.0013286261],"category_scores_gemma":[0.006859427,0.00020706617,0.00005790198,0.0009849756,0.00025015723,0.00031845935,0.00022251588,0.001067466,0.00049510214],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008716326,0.00023941888,0.46626168,0.0061283517,0.000079383855,0.0001341551,0.063945584,0.0009885655,0.0009021353,0.002993146,0.23958516,0.21002609],"study_design_scores_gemma":[0.020173771,0.030006787,0.03109853,0.13150811,0.00037786007,0.000019811549,0.066301174,0.007841729,0.00068530726,0.0027324182,0.7074426,0.0018119419],"about_ca_topic_score_codex":0.0006745636,"about_ca_topic_score_gemma":0.00012004087,"teacher_disagreement_score":0.7247304,"about_ca_system_score_codex":0.00033315804,"about_ca_system_score_gemma":0.0017622014,"threshold_uncertainty_score":0.9995843},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W3028006966","doi":"10.1136/bmjebm-2020-111386","title":"Thrombolysis with alteplase 3–4.5 hours after acute ischaemic stroke: trial reanalysis adjusted for baseline imbalances","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":41,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Impact","funders":"","keywords":"Medicine; Thrombolysis; Stroke (engine); Modified Rankin Scale; Ischaemic stroke; Clinical trial; Placebo; Internal medicine; Physical therapy; Randomized controlled trial; Tissue plasminogen activator; Ischemic stroke; Ischemia; Myocardial infarction","score_opus":0.06341808429054929,"score_gpt":0.33722726806590375,"score_spread":0.27380918377535446,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3028006966","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.48389578,0.0063512553,0.16163754,0.33234414,0.0008662692,0.012759794,0.00038779835,0.00070932944,0.0010481062],"genre_scores_gemma":[0.9359354,0.00088146544,0.020164035,0.03276777,0.0056154826,0.0022765186,0.0005951154,0.00015729856,0.0016069129],"study_design_codex":"not_applicable","study_design_gemma":"randomized_trial","domain_scores_codex":[0.9945742,0.00020848795,0.0012956731,0.0012842953,0.0018252132,0.00081211614],"domain_scores_gemma":[0.99525917,0.0013796154,0.0006732777,0.0012036876,0.0006052031,0.00087905716],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0017093101,0.0007210145,0.0019636133,0.00043574578,0.000118285825,0.000031320436,0.0004434291,0.00020349775,0.0013956326],"category_scores_gemma":[0.0037937597,0.000482986,0.0004926598,0.0013150957,0.00047544972,0.0003194563,0.00010382319,0.00047883825,0.00005300624],"study_design_candidate":"randomized_trial","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.46445882,0.0005015442,0.023180928,0.0017566242,0.0063392804,0.0019536964,0.00075307104,0.00035503646,0.016118038,0.00001853504,0.4743387,0.010225722],"study_design_scores_gemma":[0.5757156,0.06431678,0.010456008,0.02357329,0.10865515,0.00024378215,0.0059856484,0.061123326,0.05237218,0.000022391583,0.0937123,0.003823573],"about_ca_topic_score_codex":0.00012917265,"about_ca_topic_score_gemma":0.00003554003,"teacher_disagreement_score":0.45203963,"about_ca_system_score_codex":0.00022209673,"about_ca_system_score_gemma":0.0005975784,"threshold_uncertainty_score":0.9997622},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3032356396","doi":"10.1136/bmjebm-2019-111326","title":"Evaluating the quality of systematic reviews and meta-analyses published on direct oral anticoagulants in the past 5 years","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Venous Thromboembolism Diagnosis and Management","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Public Health Ontario; University of Toronto; McMaster University","funders":"","keywords":"Systematic review; Checklist; Medicine; Data extraction; MEDLINE; Guideline; Meta-analysis; Quality (philosophy); Quality Score; Medical physics; Family medicine; Internal medicine; Pathology; Psychology","score_opus":0.7738296929297955,"score_gpt":0.5664963431092327,"score_spread":0.20733334982056284,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3032356396","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.61653084,0.05772997,0.00021419176,0.30790246,0.00019832604,0.016413432,0.000009340586,0.00005708318,0.0009443781],"genre_scores_gemma":[0.97485113,0.001135211,0.00030301593,0.022897495,0.00017723527,0.00058533635,0.0000072585763,0.0000144327305,0.000028881124],"study_design_codex":"systematic_review","study_design_gemma":"observational","domain_scores_codex":[0.9931137,0.00322114,0.0015995917,0.00040528073,0.0014214278,0.00023884006],"domain_scores_gemma":[0.99335444,0.0046548215,0.0008295079,0.0008859073,0.00015395811,0.000121375146],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.020697394,0.00025094303,0.0031179087,0.00013373271,0.00006291398,0.000036538848,0.0003473901,0.00004750232,0.00022294735],"category_scores_gemma":[0.03269328,0.000098824974,0.00028223745,0.0007438897,0.0002075056,0.000111942645,0.000056619898,0.00023878763,0.000011467311],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0040655737,0.0042948476,0.089374095,0.51219475,0.022689614,0.0007307793,0.038859762,0.0022624293,0.016987747,0.004286528,0.28979072,0.014463133],"study_design_scores_gemma":[0.008190617,0.011543699,0.76569396,0.07198083,0.12355578,0.0000291994,0.00892744,0.007117992,0.00064016576,0.0002488603,0.0012392254,0.0008322516],"about_ca_topic_score_codex":0.00067530497,"about_ca_topic_score_gemma":0.00004276442,"teacher_disagreement_score":0.67631984,"about_ca_system_score_codex":0.0000340397,"about_ca_system_score_gemma":0.00008440357,"threshold_uncertainty_score":0.97545475},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch","metaepi_broad"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3039340826","doi":"10.1136/bmjebm-2020-111412","title":"Dietary Recommendations for Familial Hypercholesterolaemia: an Evidence-Free Zone","year":2020,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Diet and metabolism studies","field":"Medicine","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian VIGOUR Centre","funders":"","keywords":"Medicine; Insulin resistance; Saturated fat; Internal medicine; Cholesterol; Obesity; Endocrinology; Coronary heart disease; Low carbohydrate; Cardiology; Weight loss","score_opus":0.24919084720613657,"score_gpt":0.4015866636004412,"score_spread":0.15239581639430463,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3039340826","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004310044,0.019121295,0.0017976516,0.9671151,0.004889767,0.005539749,0.0002813802,0.0004458301,0.00037823795],"genre_scores_gemma":[0.0008476671,0.009163685,0.011743234,0.91182554,0.056970492,0.0021333336,0.0023658257,0.00025932645,0.004690901],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99262947,0.0005317016,0.0018794152,0.0017908689,0.0019480375,0.0012204796],"domain_scores_gemma":[0.99014664,0.004649153,0.0007749545,0.0025130464,0.0011709046,0.00074532384],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0018448303,0.0012389295,0.003056668,0.00066849863,0.00041801351,0.00006459336,0.001178535,0.0010584744,0.00050734787],"category_scores_gemma":[0.01740431,0.00095652754,0.0006888177,0.00075093773,0.00088661804,0.00080718246,0.0002814446,0.0031486151,0.000109974804],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017938439,0.00012490526,0.00048996304,0.003607178,0.00037354726,0.00077818613,0.0006493289,0.0000015381108,0.0010907487,0.000010082873,0.96265477,0.02842593],"study_design_scores_gemma":[0.0059113544,0.0053927344,0.0020221807,0.026779227,0.0048971213,0.000056155033,0.00046793965,0.0001491346,0.0001250045,0.0003320266,0.95296067,0.0009064409],"about_ca_topic_score_codex":0.00039533657,"about_ca_topic_score_gemma":0.0000729819,"teacher_disagreement_score":0.055289548,"about_ca_system_score_codex":0.00032101918,"about_ca_system_score_gemma":0.0014723703,"threshold_uncertainty_score":0.9992885},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3045810088","doi":"10.1136/bmjebm-2020-111339","title":"Partnering with patients in the production of evidence","year":2020,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Clinical Evaluative Sciences; SickKids Foundation; Hospital for Sick Children; University of Toronto","funders":"","keywords":"General partnership; Manifesto; Evidence-based medicine; Healthcare delivery; Evidence-based practice; Health care; Production (economics); Trustworthiness; Public relations; Medicine; Knowledge management; Engineering ethics; Psychology; Business; Political science; Alternative medicine; Computer science; Engineering; Social psychology","score_opus":0.6388868732396912,"score_gpt":0.5055099970302084,"score_spread":0.1333768762094828,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3045810088","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.016516794,0.0013315384,0.00007099136,0.9706844,0.0022897436,0.0088130385,0.000014398793,0.000041363255,0.00023769763],"genre_scores_gemma":[0.07259881,0.0008022878,0.00021324972,0.9166606,0.006612882,0.0020509472,0.00047397966,0.000051907475,0.000535329],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9923328,0.0024624711,0.0016987293,0.00065442524,0.0021206934,0.0007308462],"domain_scores_gemma":[0.9918653,0.005261173,0.0014216199,0.0008653654,0.00044074073,0.00014580209],"candidate_categories":["research_integrity"],"consensus_categories":[],"category_scores_codex":[0.003240946,0.00040334638,0.000813957,0.0002563285,0.00031464754,0.0000036818296,0.0005669963,0.00033838156,0.00026047658],"category_scores_gemma":[0.003443828,0.00019614628,0.000062395724,0.0007748735,0.00027563988,0.00021791327,0.00006689959,0.0028680158,0.00007296112],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0006871966,0.00005791125,0.17258906,0.021056717,0.000012526783,0.00008427488,0.003489661,0.000009558398,0.00000853221,0.000005688938,0.8009338,0.001065118],"study_design_scores_gemma":[0.0023635803,0.0068246135,0.027720278,0.4738932,0.00026140653,0.000002544347,0.002243724,0.000089701505,0.00007562196,0.000085580745,0.4858964,0.000543359],"about_ca_topic_score_codex":0.0016416254,"about_ca_topic_score_gemma":0.00029686306,"teacher_disagreement_score":0.45283648,"about_ca_system_score_codex":0.0003949875,"about_ca_system_score_gemma":0.0012590749,"threshold_uncertainty_score":0.9994324},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3045872399","doi":"10.1136/bmjebm-2020-111399","title":"Development of five online modules for teaching evidence-informed healthcare: the West coast Interprofessional Clinical Knowledge Evidence Disseminator (WICKED) Project","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"University of British Columbia","keywords":"Psychological intervention; Interactivity; Evidence-based practice; Critical appraisal; Medical education; Health care; Psychology; Evidence-based medicine; Nursing; Knowledge management; Medicine; Computer science; Alternative medicine; Multimedia","score_opus":0.6310685343411546,"score_gpt":0.6791755430567533,"score_spread":0.0481070087155987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3045872399","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006623392,0.919763,0.0014272986,0.053268764,0.005366164,0.019226704,0.00014620702,0.000108830965,0.000030681782],"genre_scores_gemma":[0.000863861,0.96764076,0.012007828,0.0026079067,0.006675119,0.008567009,0.00042844916,0.0001027221,0.0011063279],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9777459,0.008470414,0.007953927,0.0015708929,0.0023647295,0.0018941375],"domain_scores_gemma":[0.87743706,0.11346231,0.0043886355,0.0014227169,0.0018570637,0.0014322428],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03711147,0.000993509,0.004173033,0.00076003297,0.0025682,0.00002821168,0.0028153493,0.00093820033,0.00022981122],"category_scores_gemma":[0.17655163,0.00053632015,0.0006864837,0.0018106267,0.0018837568,0.0007893369,0.0007703314,0.0050786147,0.00017098177],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00068385404,0.00021937395,0.00046249354,0.13847291,0.00008313813,0.0000058927076,0.016484395,2.901734e-7,0.0000015851797,0.0001227181,0.12121262,0.72225076],"study_design_scores_gemma":[0.0004065932,0.00086905196,0.00047183988,0.45657498,0.00022658655,0.0000051434527,0.0137640815,0.00034903552,6.293148e-7,0.000018963447,0.5269978,0.00031528284],"about_ca_topic_score_codex":0.0014482801,"about_ca_topic_score_gemma":0.0037365104,"teacher_disagreement_score":0.72193545,"about_ca_system_score_codex":0.0021613797,"about_ca_system_score_gemma":0.13181469,"threshold_uncertainty_score":0.99970883},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W3048641009","doi":"10.1136/bmjebm-2020-111517","title":"Thrombolysis with alteplase 3–4.5 hours after acute ischaemic stroke: trial reanalysis adjusted for baseline imbalances","year":2020,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Acute Ischemic Stroke Management","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Western University","funders":"","keywords":"Thrombolysis; Medicine; Ischaemic stroke; Stroke (engine); Baseline (sea); Acute stroke; Clinical trial; Internal medicine; Cardiology; Emergency medicine; Tissue plasminogen activator; Myocardial infarction; Ischemia","score_opus":0.05651902862013785,"score_gpt":0.3302200308703915,"score_spread":0.2737010022502536,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3048641009","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0029466122,0.0046793274,0.025735555,0.9523519,0.0013718541,0.011062036,0.00089851033,0.0004815011,0.00047270072],"genre_scores_gemma":[0.008459379,0.0018522593,0.014463143,0.91099554,0.040536366,0.0063198335,0.007230055,0.00049225974,0.009651166],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98903763,0.0005038807,0.0023964148,0.002623754,0.0039024262,0.0015359095],"domain_scores_gemma":[0.9901908,0.0033653313,0.0018546853,0.002792963,0.0010523387,0.0007438927],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.0019929593,0.0018267158,0.0047930838,0.0012993428,0.00018055207,0.0000698953,0.0009832459,0.0010391269,0.0020516096],"category_scores_gemma":[0.0042010914,0.0012487036,0.0012543122,0.0016318106,0.0009504528,0.00033194866,0.0002013747,0.0028610325,0.00010117447],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.09770994,0.00012252037,0.0008017017,0.002359814,0.006125421,0.008109339,0.00010252152,0.000020684842,0.00046454242,7.084075e-7,0.8826047,0.001578112],"study_design_scores_gemma":[0.123888865,0.015053002,0.00020589691,0.023409622,0.0865444,0.00020365806,0.00041327998,0.0028141828,0.0018441994,0.000009643154,0.7435217,0.00209153],"about_ca_topic_score_codex":0.0002622096,"about_ca_topic_score_gemma":0.000040946357,"teacher_disagreement_score":0.13908298,"about_ca_system_score_codex":0.0006692754,"about_ca_system_score_gemma":0.001551725,"threshold_uncertainty_score":0.99944776},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3048654576","doi":"10.1136/bmjebm-2020-111405","title":"Preventing hospital readmission through better medication continuity after hospital discharge","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Pharmaceutical Practices and Patient Outcomes","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of British Columbia Hospital","funders":"","keywords":"Medicine; Discontinuation; Transitional care; Psychological intervention; Confusion; Hospital discharge; Health care; Hospital readmission; Hospital care; Medication Reconciliation; Patient discharge; Emergency medicine; Intensive care medicine; Medical emergency; MEDLINE; Family medicine; Nursing; Psychiatry; Pharmacy; Psychology","score_opus":0.1439962306500625,"score_gpt":0.4227465836670851,"score_spread":0.2787503530170226,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3048654576","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4161291,0.0030620887,0.0043579224,0.57149935,0.00088751427,0.0029517773,0.000009721836,0.00018389813,0.00091859157],"genre_scores_gemma":[0.953539,0.0003219967,0.002102284,0.04103283,0.002411755,0.0002439376,0.00007611365,0.000042596534,0.0002294939],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9964587,0.00022762896,0.0008815169,0.0006625078,0.0012738426,0.00049582147],"domain_scores_gemma":[0.99706984,0.0009584127,0.00048635234,0.00046545803,0.00021616896,0.0008037587],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.00097645936,0.00035861798,0.0006928375,0.00006613919,0.00010620143,0.0000253073,0.00021937388,0.00016733381,0.0026335977],"category_scores_gemma":[0.008335643,0.00024512038,0.00020309666,0.00036270986,0.00038238964,0.00085615425,0.000102324564,0.0006681297,0.00019793592],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0034118933,0.0008607753,0.910707,0.0023099286,0.0002358714,0.00054389617,0.0059072524,0.0000023017067,0.0042181327,0.00020722958,0.046404023,0.025191719],"study_design_scores_gemma":[0.024148013,0.015844202,0.462989,0.027193975,0.00386011,0.00006267495,0.0022242665,0.008281915,0.018071763,0.00066496857,0.43464902,0.002010086],"about_ca_topic_score_codex":0.000068588524,"about_ca_topic_score_gemma":0.000001647361,"teacher_disagreement_score":0.5374099,"about_ca_system_score_codex":0.00008928063,"about_ca_system_score_gemma":0.00020258191,"threshold_uncertainty_score":0.99957186},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3082235742","doi":"10.1136/bmjebm-2020-111362","title":"Conventional compared to network meta-analysis to evaluate antibiotic prophylaxis in patients with cancer and haematopoietic stem cell transplantation recipients","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Neutropenia and Cancer Infections","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"SickKids Foundation; Hospital for Sick Children; Pediatric Oncology Group; Toronto General Hospital; University of Toronto","funders":"National Institute for Health and Care Research; Pediatric Oncology Group of Ontario","keywords":"Meta-analysis; Medicine; Internal medicine; Placebo; Transplantation; Trimethoprim; Relative risk; Hematopoietic stem cell transplantation; Antibiotics; Oncology; Surgery; Confidence interval; Biology; Microbiology; Pathology","score_opus":0.14006914249045047,"score_gpt":0.3592804178242128,"score_spread":0.21921127533376236,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3082235742","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96739095,0.0013115716,0.0030124064,0.02458723,0.00015335681,0.0033931674,0.00004565901,0.000040550985,0.00006509534],"genre_scores_gemma":[0.9918417,0.000059640428,0.0006294308,0.006801752,0.00015419413,0.0002984819,0.00007230908,0.0000235541,0.00011897869],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99733263,0.00019961606,0.000614228,0.0006001045,0.00093981455,0.00031359642],"domain_scores_gemma":[0.99843204,0.00019918583,0.00017447816,0.00027646037,0.00037684408,0.0005409733],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0005617865,0.00026663815,0.0012182189,0.0003176376,0.00008269422,0.000018022809,0.00010159877,0.0000551863,0.00066519494],"category_scores_gemma":[0.000057536763,0.00018020139,0.00017569878,0.002243472,0.000083678206,0.0001139098,0.00001882759,0.00018552234,0.000026377755],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.006555292,0.00023439848,0.94628775,0.0008954798,0.0065166582,0.00008015792,0.0010654781,0.034482606,0.00046850755,0.000013561443,0.002327529,0.0010725609],"study_design_scores_gemma":[0.006962511,0.004337783,0.909039,0.0014029946,0.0761214,0.000004320884,0.0000872205,0.00063181185,0.0007926087,0.000004051636,0.0002987497,0.00031754887],"about_ca_topic_score_codex":0.00036597028,"about_ca_topic_score_gemma":0.00033639892,"teacher_disagreement_score":0.06960474,"about_ca_system_score_codex":0.00013450946,"about_ca_system_score_gemma":0.00024162645,"threshold_uncertainty_score":0.7348399},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"meta_analysis","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3090996960","doi":"10.1136/bmjebm-2020-111499","title":"Quality of early evidence on the pathogenesis, diagnosis, prognosis and treatment of COVID-19","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"COVID-19 Clinical Research Studies","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of Ottawa; McMaster University","funders":"","keywords":"Checklist; MEDLINE; Critical appraisal; Medicine; Rigour; Systematic review; Consolidated Standards of Reporting Trials; Declaration; Family medicine; Publication bias; Clinical study design; Data extraction; Clinical trial; Alternative medicine; Psychology; Meta-analysis; Pathology; Political science","score_opus":0.5759781474648025,"score_gpt":0.5831047699438358,"score_spread":0.007126622479033284,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3090996960","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00033447111,0.8836675,0.00010627154,0.106363006,0.00011397191,0.009200175,0.00014135666,0.000057843332,0.000015381906],"genre_scores_gemma":[0.005272352,0.9850683,0.00015362242,0.00398845,0.0007058112,0.0046352544,0.000028391867,0.00007020654,0.00007761468],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9887285,0.0031762328,0.0033379404,0.0013601374,0.002802623,0.00059455266],"domain_scores_gemma":[0.79159653,0.20278504,0.0021592856,0.001767313,0.00067480846,0.0010169864],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0075322916,0.00094783196,0.006883209,0.00044914076,0.00016537607,0.000015461146,0.0006558876,0.00039134658,0.00031223078],"category_scores_gemma":[0.41404334,0.0004683181,0.0011692956,0.0015310892,0.002361379,0.00009167012,0.00025576676,0.00060463,0.000021817896],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017442615,0.0007627913,0.023541007,0.15127037,0.0009768867,0.0002596352,0.00040273587,0.0000013948127,0.000026530694,0.00019414356,0.003118599,0.81770164],"study_design_scores_gemma":[0.0021187577,0.018473707,0.00896362,0.26421022,0.004894983,0.0000056764407,0.00008383263,0.0000081035605,0.000058656537,0.000050135845,0.70071024,0.00042205182],"about_ca_topic_score_codex":0.0018872094,"about_ca_topic_score_gemma":0.000043900844,"teacher_disagreement_score":0.8172796,"about_ca_system_score_codex":0.0009378795,"about_ca_system_score_gemma":0.007517273,"threshold_uncertainty_score":0.99977684},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3094203153","doi":"10.1136/bmjebm-2020-111452","title":"Optimising the process for conducting scoping reviews","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":28,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Izaak Walton Killam Health Centre; BC Children's Hospital; University of Toronto; Dalhousie University; York University; University of British Columbia","funders":"Canadian Institutes of Health Research","keywords":"Process (computing); Systematic review; Resource (disambiguation); Knowledge management; Process management; Best practice; Key (lock); Computer science; Management science; Data science; Engineering; MEDLINE; Political science","score_opus":0.895103126102174,"score_gpt":0.7197301643337385,"score_spread":0.17537296176843553,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3094203153","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000011113873,0.92003435,0.0029960063,0.04882947,0.0026302726,0.025135271,0.000008587311,0.00007246426,0.00029245633],"genre_scores_gemma":[0.0000025631641,0.9695638,0.0011702315,0.007254298,0.009902827,0.0115453685,0.000073008305,0.000057902384,0.0004300384],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.99002653,0.0036058337,0.00292172,0.00084798486,0.001111593,0.0014863388],"domain_scores_gemma":[0.9722173,0.022975633,0.0023490656,0.0008300898,0.0008015365,0.0008263656],"candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.028991865,0.00048712303,0.0027275467,0.00031921262,0.0026406052,0.000033815602,0.0011819132,0.0003581536,0.0006569694],"category_scores_gemma":[0.10179639,0.00024343682,0.0003006394,0.002010405,0.00041852015,0.00034383018,0.000109746376,0.0019354305,0.00032475888],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019739588,0.000008051316,0.000008112787,0.31934646,0.000011202377,0.0000026624548,0.0012937605,0.000002146368,4.862236e-7,0.00002994522,0.070177466,0.6091],"study_design_scores_gemma":[0.0001622236,0.00014322474,3.8042288e-7,0.43489125,0.00015909651,0.0000031380905,0.0010923144,0.00012103917,2.8720538e-7,0.000024321107,0.56326693,0.0001357563],"about_ca_topic_score_codex":0.0001640911,"about_ca_topic_score_gemma":0.00003315778,"teacher_disagreement_score":0.6089642,"about_ca_system_score_codex":0.000624968,"about_ca_system_score_gemma":0.023733709,"threshold_uncertainty_score":0.9998572},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3098080120","doi":"10.1136/bmjebm-2020-111503","title":"Ten years later: a review of the US 2009 institute of medicine report on conflicts of interest and solutions for further reform","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":33,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Conflict of interest; Guideline; Alternative medicine; Political science; Health care; Medical research; Public relations; Medical education; Engineering ethics; Medicine; Engineering; Law","score_opus":0.7738178979150093,"score_gpt":0.6174324461110864,"score_spread":0.15638545180392294,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3098080120","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00005984295,0.93055004,0.00001394017,0.063426875,0.0011388744,0.0039249654,0.0002932059,0.00001909156,0.0005731732],"genre_scores_gemma":[0.0011426569,0.98734635,0.000032845892,0.01000314,0.0007883641,0.0002698097,0.00010069685,0.000033207485,0.0002829444],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9953872,0.00086859125,0.0024619848,0.00047077276,0.00038988452,0.0004215945],"domain_scores_gemma":[0.9945471,0.0017107378,0.0022155042,0.0006390502,0.00046187462,0.0004256877],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004985844,0.00044130758,0.0028828974,0.00018671097,0.00009693319,0.0000014921307,0.0005755045,0.0007329551,0.00054083375],"category_scores_gemma":[0.005762634,0.00025125174,0.00036580756,0.0006216591,0.0022103472,0.00007586263,0.00014862447,0.0016913952,0.0000045598235],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00066757115,0.00020304335,0.00013147178,0.3786568,0.0007300445,0.00019752099,0.00025566723,0.0000037649402,0.00011104526,0.0021700184,0.039665442,0.57720757],"study_design_scores_gemma":[0.0010079598,0.00074973115,0.000055037744,0.28774452,0.002853624,0.00012133751,0.00001764541,0.000022762071,0.000084237225,0.0000545719,0.7071473,0.00014126005],"about_ca_topic_score_codex":0.00013090712,"about_ca_topic_score_gemma":0.000026358759,"teacher_disagreement_score":0.66748184,"about_ca_system_score_codex":0.00015862672,"about_ca_system_score_gemma":0.0011872284,"threshold_uncertainty_score":0.999994},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"incentives","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3101218852","doi":"10.1136/bmjebm-2020-111577","title":"Blinding practices during acute point-of-care ultrasound research: the BLIND-US meta-research study","year":2020,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; University of Ottawa","funders":"","keywords":"Blinding; Medicine; Subgroup analysis; Point of care ultrasound; Medical physics; Meta-analysis; Research design; Physical therapy; Clinical trial; Radiology; Ultrasound; Internal medicine; Statistics","score_opus":0.717634259424709,"score_gpt":0.5996602644583762,"score_spread":0.11797399496633287,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3101218852","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.77517885,0.00559626,0.00019518519,0.20999558,0.0001557406,0.007554821,0.00002081906,0.00013080818,0.0011719547],"genre_scores_gemma":[0.99092764,0.0010589877,0.0014724792,0.0023542324,0.0021987413,0.0015396209,0.00003627242,0.00007546234,0.00033659372],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98815304,0.0027535802,0.0017695342,0.0011973123,0.00515218,0.00097437075],"domain_scores_gemma":[0.9165762,0.07452274,0.0011621891,0.0024405043,0.004295807,0.0010025473],"candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019739853,0.0004010341,0.0013625036,0.0005822474,0.00088510214,0.00008927068,0.0013194773,0.00021558651,0.0019824083],"category_scores_gemma":[0.16528776,0.00022807367,0.00036904705,0.0044570914,0.002175352,0.00045931645,0.00031228404,0.00315741,0.0002507689],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.030695086,0.005557392,0.57128394,0.006315762,0.014258268,0.0013132388,0.061158415,0.00022343724,0.26273298,0.001761222,0.042216606,0.0024836678],"study_design_scores_gemma":[0.04658049,0.06672771,0.3735665,0.013009029,0.042840295,0.00063692295,0.32328472,0.00059321563,0.08762902,0.0019919125,0.041109387,0.0020308031],"about_ca_topic_score_codex":0.0005116296,"about_ca_topic_score_gemma":0.00008839469,"teacher_disagreement_score":0.2621263,"about_ca_system_score_codex":0.00024562725,"about_ca_system_score_gemma":0.001278241,"threshold_uncertainty_score":0.99914235},"labels":[],"label_agreement":null},{"id":"W3103117573","doi":"10.1136/bmjebm-2020-111479","title":"Ten papers for teachers of evidence-based medicine and health care: Sicily workshop 2019","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"National Institute for Health and Care Research","keywords":"Health care; Medical education; Library science; Psychology; Medicine; Political science; Computer science","score_opus":0.7018070483281779,"score_gpt":0.6353810479535997,"score_spread":0.06642600037457813,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3103117573","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000044414865,0.725331,0.00061785366,0.2604767,0.0017571425,0.011485293,0.000063533815,0.000088338194,0.00017567968],"genre_scores_gemma":[0.00008844558,0.967553,0.001689505,0.020551354,0.0052489983,0.0032616942,0.0003990856,0.00011414685,0.001093765],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.98578423,0.0039264355,0.004229362,0.001589539,0.0024716605,0.0019987766],"domain_scores_gemma":[0.9570081,0.034936067,0.0031234294,0.0013051553,0.0011748008,0.002452434],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.019619536,0.00095664535,0.0050616344,0.0011439064,0.0011162717,0.00001710489,0.0012378219,0.0008222198,0.00074519747],"category_scores_gemma":[0.051381666,0.0005598017,0.00040241968,0.0025605545,0.0016502903,0.0003817916,0.000148663,0.0023141026,0.00006986311],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017127096,0.000026021251,0.00020195536,0.1036194,0.00004229153,0.0000058804776,0.003973732,0.0000025751099,0.0000020228158,0.000046241737,0.34885597,0.5430527],"study_design_scores_gemma":[0.001213166,0.002285121,0.00015627066,0.33423516,0.0003791687,0.0000022570887,0.0056937593,0.000054418037,3.5187625e-7,0.000015921718,0.6556509,0.00031345309],"about_ca_topic_score_codex":0.004525818,"about_ca_topic_score_gemma":0.00038598606,"teacher_disagreement_score":0.5427392,"about_ca_system_score_codex":0.0021429635,"about_ca_system_score_gemma":0.045417387,"threshold_uncertainty_score":0.9999876},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W3112825017","doi":"10.1136/bmjebm-2020-111493","title":"Framework for the synthesis of non-randomised studies and randomised controlled trials: a guidance on conducting a systematic review and meta-analysis for healthcare decision making","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":103,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Agency for Drugs and Technologies in Health","funders":"Sanofi Pasteur; Sanofi","keywords":"Health care; Randomized controlled trial; Evidence-based medicine; Systematic review; Delphi method; Management science; Medicine; Quality (philosophy); Transparency (behavior); Best practice; MEDLINE; Computer science; Alternative medicine; Engineering; Artificial intelligence; Political science","score_opus":0.9173359485862664,"score_gpt":0.6565785159814549,"score_spread":0.26075743260481155,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3112825017","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":"methods","domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[1.9418988e-7,0.7137744,0.19176729,0.014413551,0.00017294196,0.07974252,0.000118143296,0.000007243581,0.000003700114],"genre_scores_gemma":[0.0001700166,0.91320604,0.024859734,0.0030235522,0.00028885674,0.058274433,0.000006144409,0.00005968208,0.00011155757],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.85137033,0.07963595,0.058483843,0.0029408834,0.0069740755,0.00059494283],"domain_scores_gemma":[0.055432267,0.86469644,0.06914026,0.0054791854,0.0049097575,0.00034206966],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"category_scores_codex":[0.5970172,0.0017180387,0.17388646,0.0013545112,0.0005178866,0.00039129352,0.0023662203,0.00039237732,0.0004501731],"category_scores_gemma":[0.95799196,0.0005147084,0.03550149,0.003912121,0.0004268749,0.00017199728,0.00015780436,0.00048235964,0.000011467777],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0031477525,0.00001506959,5.6288775e-7,0.72092074,0.22796775,0.0000036872887,0.0001774767,0.000013467756,1.068129e-7,0.00021524107,0.002917705,0.04462042],"study_design_scores_gemma":[0.008844359,0.00019622748,1.08585404e-7,0.3787683,0.60592985,0.0000066495545,0.0004453892,0.0010993016,2.8183416e-7,0.00077751034,0.003667225,0.00026477047],"about_ca_topic_score_codex":0.0000082324605,"about_ca_topic_score_gemma":0.0000117290365,"teacher_disagreement_score":0.7959381,"about_ca_system_score_codex":0.0001194312,"about_ca_system_score_gemma":0.0006336566,"threshold_uncertainty_score":0.99973047},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W3117477626","doi":"10.1136/bmjebm-2020-111393","title":"Successes, lessons and opportunities: 15-year follow-up of an integrated evidence-based medicine curriculum","year":2020,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"University of British Columbia; College of Family Physicians of Canada; University of Alberta","funders":"University of Alberta","keywords":"Curriculum; Medical education; Evidence-based medicine; Critical appraisal; Curriculum development; Evidence-based practice; Medicine; Family medicine; Psychology; Alternative medicine; Pedagogy; Pathology","score_opus":0.5774288423614425,"score_gpt":0.5805625078589207,"score_spread":0.0031336654974781775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3117477626","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001480768,0.8781462,0.0006736825,0.11097,0.003114992,0.0063564936,0.00012347671,0.00017564572,0.00029148202],"genre_scores_gemma":[0.00080411334,0.9861412,0.0005626841,0.00499331,0.0038684562,0.0018251917,0.00078327453,0.00011737251,0.00090438506],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9826136,0.0065843207,0.004368913,0.0016334105,0.0030372047,0.0017625304],"domain_scores_gemma":[0.97234625,0.017427836,0.002981413,0.001656737,0.0023743776,0.0032133996],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.016428793,0.0011203394,0.004989987,0.0017164176,0.0010387014,0.000033314576,0.0017760636,0.0009633475,0.005146588],"category_scores_gemma":[0.05140471,0.0007056519,0.00030630955,0.0035209837,0.002391601,0.0011259667,0.0002175587,0.0029591557,0.00013902171],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003937227,0.00015966041,0.00064270827,0.14841187,0.00009702932,0.00015535184,0.0017682358,0.0000019759782,0.000007442208,0.00062016083,0.15742901,0.6903128],"study_design_scores_gemma":[0.0014226235,0.0021441747,0.000065530636,0.3502629,0.0005958972,0.000009126443,0.0061531877,0.00033321403,0.0000014140314,0.000046999987,0.63855255,0.00041234863],"about_ca_topic_score_codex":0.0139365755,"about_ca_topic_score_gemma":0.0010044725,"teacher_disagreement_score":0.68990046,"about_ca_system_score_codex":0.00097182527,"about_ca_system_score_gemma":0.04168568,"threshold_uncertainty_score":0.99953943},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"qualitative","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W3121563112","doi":"10.1136/bmjebm-2020-111604","title":"Barriers to reporting guideline adherence in point-of-care ultrasound research: a cross-sectional survey of authors and journal editors","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Likert scale; Guideline; Medicine; Family medicine; Cross-sectional study; Alternative medicine; Survey research; MEDLINE; Primary care; Medical education; Medical physics; Psychology; Pathology; Applied psychology","score_opus":0.37060179582142727,"score_gpt":0.5493210041737503,"score_spread":0.17871920835232302,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3121563112","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98539764,0.0020530543,0.0018325173,0.008327161,0.0011939859,0.0009268759,0.000040537518,0.000020853338,0.00020739643],"genre_scores_gemma":[0.9852462,0.00025771733,0.011374915,0.0007231087,0.0020487267,0.000072189934,0.000077833116,0.000028082002,0.00017121606],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99157006,0.00081411796,0.0046402137,0.00064357,0.00187946,0.00045256875],"domain_scores_gemma":[0.9688174,0.018492382,0.0018542346,0.0008758767,0.008746776,0.0012133225],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.028540729,0.0002083276,0.00090607046,0.00050515804,0.00017763786,0.00003158236,0.00025785452,0.00023720028,0.0013212481],"category_scores_gemma":[0.5496205,0.00017002004,0.00012193859,0.0021972798,0.0010690127,0.00017997979,0.00008949903,0.0010997334,0.000005162568],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00075742503,0.00007710012,0.9670929,0.00033980282,0.00003524978,0.00009415442,0.0005076342,0.0001093993,0.02033971,0.000034721037,0.009921089,0.00069081987],"study_design_scores_gemma":[0.0017297871,0.0016735499,0.9826974,0.004789034,0.000042909127,0.00038196522,0.0018666358,0.000059425023,0.0049456414,0.00042767686,0.0012398043,0.00014612361],"about_ca_topic_score_codex":0.0012138258,"about_ca_topic_score_gemma":0.00044482324,"teacher_disagreement_score":0.52107984,"about_ca_system_score_codex":0.00025631182,"about_ca_system_score_gemma":0.004447122,"threshold_uncertainty_score":0.9995917},"labels":[],"label_agreement":null},{"id":"W3136262608","doi":"10.1136/bmjebm-2020-111542","title":"Understanding of research results, evidence summaries and their applicability—not critical appraisal—are core skills of medical curriculum","year":2021,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":51,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Terveyden Tutkimuksen Toimikunta; Sigrid Juséliuksen Säätiö; Helsingin ja Uudenmaan Sairaanhoitopiiri","keywords":"Critical appraisal; Rigour; Evidence-based medicine; Observational study; Systematic review; Medical education; Psychology; Medicine; Critical reading; Evidence-based practice; Psychological intervention; Medical literature; Curriculum; Reading (process); MEDLINE; Alternative medicine; Nursing; Pedagogy; Epistemology; Pathology","score_opus":0.8221500894503884,"score_gpt":0.6922985889787482,"score_spread":0.12985150047164018,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3136262608","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0001708024,0.93728966,0.0009864394,0.055391345,0.0016816213,0.004144358,0.00016845654,0.0000429409,0.00012437401],"genre_scores_gemma":[0.004303672,0.99155086,0.00032110434,0.00034764924,0.00212771,0.001098503,0.00011369804,0.00004427284,0.0000925021],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9768684,0.009646107,0.00438275,0.0014368773,0.005947994,0.0017179039],"domain_scores_gemma":[0.64150274,0.34751946,0.0020269032,0.0022963795,0.0041827573,0.0024717434],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07394298,0.0005499406,0.0037806751,0.0009332692,0.0010665506,0.000027746628,0.0015138674,0.0012394119,0.0009825596],"category_scores_gemma":[0.52697676,0.0003388712,0.0002744907,0.0038867004,0.0076546795,0.0003541876,0.0007544247,0.0041482644,0.000023239309],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0004158938,0.0006127253,0.0049966597,0.5607635,0.00010149719,0.0001054019,0.0054415483,7.242642e-7,0.0000054851735,0.011141533,0.2673598,0.14905523],"study_design_scores_gemma":[0.0005643659,0.0007082992,0.0002905758,0.7038736,0.0001355456,0.000020526038,0.024563346,0.00015440135,0.0000031705242,0.00052289694,0.26888198,0.00028133407],"about_ca_topic_score_codex":0.0024963198,"about_ca_topic_score_gemma":0.00044811112,"teacher_disagreement_score":0.4530338,"about_ca_system_score_codex":0.0010556623,"about_ca_system_score_gemma":0.028700365,"threshold_uncertainty_score":0.9999307},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3150860698","doi":"10.1136/bmjebm-2020-111652","title":"Completeness of reporting for systematic reviews of point-of-care ultrasound: a meta-research study","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Ultrasound in Clinical Applications","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Vancouver General Hospital; McMaster University; Queen's University; University of British Columbia; University of Ottawa","funders":"","keywords":"Checklist; Systematic review; Medicine; Cochrane Library; MEDLINE; Meta-analysis; Subgroup analysis; Publication bias; Medical physics; Pathology; Psychology","score_opus":0.6497725731400538,"score_gpt":0.5765663789344181,"score_spread":0.07320619420563568,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3150860698","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.76860124,0.108206704,0.034655873,0.017165337,0.00052979734,0.06967634,0.00013002518,0.000117376534,0.0009173271],"genre_scores_gemma":[0.9801033,0.00055249606,0.0148760965,0.00030787286,0.00022884912,0.0036941366,0.000071903574,0.000037592163,0.00012774227],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98805267,0.0014767876,0.0077271713,0.0006192931,0.0017723229,0.00035175798],"domain_scores_gemma":[0.9266808,0.057329882,0.0059974715,0.0023622343,0.007375202,0.00025443258],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.026478492,0.0002759189,0.00463785,0.00028884283,0.0000901348,0.000006117134,0.00036841125,0.00011855287,0.0005494096],"category_scores_gemma":[0.35135746,0.00017249335,0.00077610527,0.0015133058,0.00064560043,0.00008771301,0.00006730744,0.00035400514,0.0000056255303],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011500167,0.0035296204,0.114042774,0.6500482,0.003245007,0.00007449416,0.0049190237,0.000032220833,0.21791074,0.0013655782,0.0029340722,0.0007482423],"study_design_scores_gemma":[0.016933318,0.026186168,0.0565001,0.6088585,0.050681423,0.0005200237,0.08684662,0.00032506557,0.14750585,0.002475793,0.0021574073,0.0010097139],"about_ca_topic_score_codex":0.00020635335,"about_ca_topic_score_gemma":0.00006701312,"teacher_disagreement_score":0.32487896,"about_ca_system_score_codex":0.000102193364,"about_ca_system_score_gemma":0.0010142679,"threshold_uncertainty_score":0.91769695},"labels":[],"label_agreement":null},{"id":"W3152785930","doi":"10.1136/bmjebm-2020-111633","title":"Does my patient have SARS-CoV-2 infection? A reminder of clinical probability formulas","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Jewish General Hospital","funders":"","keywords":"Bayes' theorem; Conditional probability; Law of total probability; Table (database); Probability theory; Contingency table; Medicine; Mathematics; Computer science; Posterior probability; Calculus (dental); Bayesian probability; Artificial intelligence; Statistics; Data mining","score_opus":0.1913117687147357,"score_gpt":0.47436889575860586,"score_spread":0.28305712704387015,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3152785930","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96181273,0.0008523674,0.0013362578,0.031716183,0.001643763,0.0013000893,0.000009497173,0.00009346643,0.0012356656],"genre_scores_gemma":[0.98647124,0.0004932723,0.003546288,0.007888218,0.0011114527,0.00011533307,0.000048551312,0.000029451767,0.0002962125],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9948089,0.0004699541,0.0023966187,0.0007862552,0.0011334099,0.00040483952],"domain_scores_gemma":[0.96293515,0.033021394,0.0007500449,0.001319625,0.00154906,0.00042475035],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003713264,0.000300676,0.001436571,0.000120350065,0.00006687285,0.000010085887,0.00011258422,0.0003408685,0.0005234384],"category_scores_gemma":[0.412501,0.00016628749,0.00047966142,0.00047170944,0.00074828265,0.00011051204,0.00008970777,0.0006133913,0.000031619085],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0017880611,0.0019641037,0.9112291,0.0006804256,0.0001677899,0.00043372545,0.0002710364,0.000019674813,0.002270046,0.00017410812,0.020851284,0.060150664],"study_design_scores_gemma":[0.013310791,0.008046188,0.7696353,0.07243218,0.0015000924,0.0001774972,0.0003107197,0.001096119,0.11464987,0.0041978667,0.014065153,0.00057822425],"about_ca_topic_score_codex":0.0006257737,"about_ca_topic_score_gemma":0.00023072063,"teacher_disagreement_score":0.40878776,"about_ca_system_score_codex":0.00011718012,"about_ca_system_score_gemma":0.0018659276,"threshold_uncertainty_score":0.6781007},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W3165462729","doi":"10.1136/bmjebm-2021-111675","title":"Quality of systematic reviews supporting the 2017 ACC/AHA and 2018 ESC/ESH guidelines for the management of hypertension","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"","keywords":"Checklist; Systematic review; Medicine; Protocol (science); MEDLINE; Family medicine; Canadian Cardiovascular Society; Alternative medicine; Internal medicine; Psychology; Pathology; Myocardial infarction","score_opus":0.9622276600800553,"score_gpt":0.6563818915020886,"score_spread":0.30584576857796675,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3165462729","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021401769,0.52110434,0.23947175,0.1914007,0.0016687597,0.023974309,0.000028481269,0.000012238574,0.0009376593],"genre_scores_gemma":[0.8309267,0.05094185,0.07286264,0.01879501,0.00144008,0.0032660197,0.000031931337,0.00007252248,0.021663211],"study_design_codex":"not_applicable","study_design_gemma":"systematic_review","domain_scores_codex":[0.95670444,0.012758158,0.023462707,0.00090853183,0.0058658887,0.00030030523],"domain_scores_gemma":[0.8684294,0.08080724,0.029348519,0.01060261,0.010600289,0.00021191295],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.36063233,0.0003102004,0.005633274,0.00013604881,0.00020940666,0.000109048786,0.0015656158,0.000061301515,0.00091329217],"category_scores_gemma":[0.36208808,0.00009273541,0.0012947301,0.0010703654,0.0002856916,0.00016497844,0.00020018776,0.00009764965,0.00004560479],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00008630961,0.00009014839,0.0037565501,0.13123442,0.0010669613,0.000022788156,0.0007274277,0.00015281691,0.0023001367,0.0035377652,0.81699467,0.040030025],"study_design_scores_gemma":[0.0057688025,0.0010181813,0.04518874,0.51392174,0.02773887,0.00030233085,0.06783527,0.109889105,0.0022479962,0.013509359,0.21105693,0.0015226776],"about_ca_topic_score_codex":0.0000579887,"about_ca_topic_score_gemma":0.000037108966,"teacher_disagreement_score":0.80952495,"about_ca_system_score_codex":0.000019650795,"about_ca_system_score_gemma":0.00013599644,"threshold_uncertainty_score":0.9999909},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W3201255490","doi":"10.1136/bmjebm-2021-111772","title":"Appraising qualitative health research—towards a differentiated approach","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":3,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Nipissing University","funders":"","keywords":"Critical appraisal; Qualitative research; Management science; Health care; Research design; Neglect; Qualitative property; Engineering ethics; Psychology; Data science; Computer science; Knowledge management; Sociology; Medicine; Social science; Engineering; Political science; Nursing; Alternative medicine; Pathology","score_opus":0.745284703788777,"score_gpt":0.6760284951873954,"score_spread":0.0692562086013816,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W3201255490","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.023181848,0.032833796,0.04213146,0.85452574,0.002955099,0.0052845045,0.00006362829,0.00056265324,0.03846126],"genre_scores_gemma":[0.65787524,0.0051650004,0.021755138,0.28736687,0.006350747,0.0025058803,0.0011109102,0.00023200226,0.017638214],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9758587,0.017199624,0.0019035559,0.00092193583,0.0022584451,0.0018577514],"domain_scores_gemma":[0.98089767,0.014118472,0.0006475522,0.0011620449,0.0022041623,0.0009700791],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.024872858,0.00037379222,0.0013540138,0.0005743355,0.0019446039,0.000019464542,0.00045832415,0.00033070735,0.0021910327],"category_scores_gemma":[0.023153855,0.00028069882,0.00015297889,0.0020470566,0.0005510876,0.000304844,0.00028065633,0.0025148767,0.00026124675],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007832834,0.0003721088,0.0055738767,0.009920449,0.00012281684,0.00016217378,0.07385322,0.000007323427,0.0006857734,0.009868869,0.8728244,0.025825765],"study_design_scores_gemma":[0.01972923,0.003963572,0.122471526,0.05301718,0.0002807869,0.000059344376,0.3830852,0.0013756941,0.0005091067,0.013483814,0.40015578,0.0018687708],"about_ca_topic_score_codex":0.0038861507,"about_ca_topic_score_gemma":0.0004533999,"teacher_disagreement_score":0.6346934,"about_ca_system_score_codex":0.0019755564,"about_ca_system_score_gemma":0.024283148,"threshold_uncertainty_score":0.99996454},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4200239034","doi":"10.1136/bmjebm-2021-111850","title":"Users' Guides to the Medical Literature series on social media (part 2): how to appraise studies using data from platforms","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Social Media in Health Education","field":"Social Sciences","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; Université de Montréal","funders":"","keywords":"Series (stratigraphy); Social media; Data science; Computer science; World Wide Web; Information retrieval; Geology","score_opus":0.64309403782733,"score_gpt":0.5438279267303098,"score_spread":0.09926611109702022,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200239034","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19577633,0.009285205,0.00019949023,0.77453595,0.017982585,0.0018945229,0.00011391474,0.00011645941,0.00009556179],"genre_scores_gemma":[0.5461551,0.014684841,0.018727792,0.15081418,0.26332706,0.0037376382,0.0014984885,0.0002319345,0.00082297256],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99349797,0.0008360172,0.00069496554,0.0008393913,0.0034064504,0.0007252171],"domain_scores_gemma":[0.9697146,0.026350176,0.00033921492,0.0012619628,0.0013062585,0.0010278291],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.0072135204,0.0003153873,0.00068514084,0.0001785272,0.0014031644,0.00017059606,0.0014108842,0.0003324189,0.0005821879],"category_scores_gemma":[0.34629515,0.0002122637,0.00007206599,0.0021283946,0.0009159776,0.00078822597,0.00033230416,0.0005587315,0.000066245026],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00041758336,0.00010545672,0.0029051173,0.0001981872,0.00012092105,0.00021389392,0.2661308,0.000019320416,0.00011835399,0.0022611748,0.69760805,0.029901117],"study_design_scores_gemma":[0.000980814,0.0002683267,0.009163463,0.01623545,0.00026720288,0.000008951142,0.2094912,0.00008136846,0.00034766537,0.0014649465,0.7610567,0.0006338917],"about_ca_topic_score_codex":0.0017542042,"about_ca_topic_score_gemma":0.020005945,"teacher_disagreement_score":0.6237218,"about_ca_system_score_codex":0.0005187217,"about_ca_system_score_gemma":0.0043059248,"threshold_uncertainty_score":0.9998969},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4200294250","doi":"10.1136/bmjebm-2021-111817","title":"Users' Guides to the Medical Literature series on social media (part 1): how to interpret healthcare information available on platforms","year":2021,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Social Media in Health Education","field":"Social Sciences","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; Université de Montréal","funders":"","keywords":"Social media; Health care; Medical information; Data science; Internet privacy; Computer science; Public relations; World Wide Web; Knowledge management; Political science","score_opus":0.20942782448776642,"score_gpt":0.434394690256741,"score_spread":0.22496686576897457,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4200294250","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.041521695,0.0010085145,0.00010561343,0.9397186,0.013668647,0.0025570828,0.0000371548,0.00015074085,0.001231948],"genre_scores_gemma":[0.62822,0.0032872004,0.0012877611,0.3153249,0.044871252,0.00520304,0.00047085606,0.00007776519,0.0012571787],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9941326,0.0006749054,0.0007532372,0.00042656908,0.0032684589,0.0007442747],"domain_scores_gemma":[0.98737687,0.0095317215,0.00023821738,0.0005307341,0.0012660399,0.0010564328],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0057498286,0.00028468051,0.00047875015,0.00030608755,0.0011335739,0.00021983703,0.00067723554,0.000417699,0.0014528998],"category_scores_gemma":[0.18680772,0.0001973295,0.00009247343,0.0021122273,0.00039277683,0.0010062504,0.000074701726,0.0007138345,0.00057884003],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00049973646,0.000047550315,0.00096145575,0.00023154187,0.000015103918,0.000030359322,0.20439842,0.000008353196,0.0000075026164,0.01079488,0.7264316,0.056573465],"study_design_scores_gemma":[0.0005718278,0.00078968174,0.0043672444,0.011472818,0.000030483976,0.00000702715,0.048037045,0.000013948716,0.00035643877,0.0005849106,0.93343323,0.00033535252],"about_ca_topic_score_codex":0.0011727628,"about_ca_topic_score_gemma":0.008668172,"teacher_disagreement_score":0.6243937,"about_ca_system_score_codex":0.0008248908,"about_ca_system_score_gemma":0.004542502,"threshold_uncertainty_score":0.9994599},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4220907181","doi":"10.1136/bmjebm-2022-111953","title":"Advocating for evidence-informed decisions to make healthcare fit for each person","year":2022,"lang":"en","type":"editorial","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital; University of British Columbia","funders":"","keywords":"Health care; Psychology; Public relations; Internet privacy; Social psychology; Political science; Computer science; Law","score_opus":0.555224960111113,"score_gpt":0.6254457142454214,"score_spread":0.0702207541343084,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4220907181","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0004257946,0.0061603454,0.0073663895,0.25414523,0.7043544,0.026369724,0.00082346913,0.00019170385,0.00016294843],"genre_scores_gemma":[0.0009009738,0.0028317054,0.013783083,0.015821114,0.89001226,0.061757647,0.0020968101,0.00018092278,0.012615448],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9853586,0.0017237692,0.002910013,0.0017692099,0.0050428817,0.0031955072],"domain_scores_gemma":[0.78161544,0.20661779,0.0017846813,0.0018316568,0.0052876775,0.0028627736],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.029675202,0.00074446923,0.0017046009,0.0014289198,0.0054904926,0.0000591198,0.0020310844,0.0011993273,0.0019263176],"category_scores_gemma":[0.50214326,0.00059844716,0.00036474149,0.002046595,0.00028542083,0.0004992202,0.00027331023,0.0032851016,0.00020565414],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.002288916,0.000049730523,0.0004293634,0.010037075,0.000020980522,0.000003952886,0.006616272,0.000043718694,0.000015919532,0.000038562805,0.95751584,0.022939662],"study_design_scores_gemma":[0.0020627966,0.005344846,0.00036690748,0.052508343,0.00010055628,9.271526e-7,0.021447066,0.0010869014,0.0000033008764,0.00014564286,0.9164104,0.0005223271],"about_ca_topic_score_codex":0.007920371,"about_ca_topic_score_gemma":0.00519969,"teacher_disagreement_score":0.47246805,"about_ca_system_score_codex":0.0051576816,"about_ca_system_score_gemma":0.07396291,"threshold_uncertainty_score":0.99964666},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4223598199","doi":"10.1136/bmjebm-2022-111978","title":"In memory of Richard Saitz (1963–2022): former editor of BMJ Evidence-Based Medicine","year":2022,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Psychoanalysis; Medicine; History; Classics; Psychology; Cognitive science","score_opus":0.2879360818303817,"score_gpt":0.5279963544396121,"score_spread":0.2400602726092304,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4223598199","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0028497337,0.0137442555,0.0003234667,0.93428224,0.03872155,0.009018748,0.000111603076,0.00007577142,0.0008726304],"genre_scores_gemma":[0.0068548787,0.005268487,0.0007326017,0.82278955,0.14700887,0.0067050997,0.00067959493,0.00018548076,0.009775442],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97548777,0.0065283217,0.0057990435,0.0015724882,0.0080225095,0.0025898728],"domain_scores_gemma":[0.9544165,0.03601903,0.004043544,0.0023812526,0.0022159496,0.00092369784],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.033300344,0.0009169384,0.0031489339,0.0036441123,0.0007350226,0.000008155048,0.0023440141,0.0014395246,0.040440798],"category_scores_gemma":[0.062178425,0.00067762134,0.00032403713,0.0045375596,0.002266144,0.0008965498,0.00032382697,0.0077121034,0.00018795532],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012613611,0.00012230624,0.0058120554,0.018821077,0.000037737987,0.00019902397,0.0024293528,0.000050489693,0.00055436156,0.000012053482,0.96948165,0.0012185287],"study_design_scores_gemma":[0.0039701727,0.005075105,0.00427347,0.092014775,0.00021458906,0.000007822087,0.0045192493,0.00068737287,0.00016865713,0.00025919703,0.88812983,0.0006797813],"about_ca_topic_score_codex":0.018116906,"about_ca_topic_score_gemma":0.00051433884,"teacher_disagreement_score":0.11149269,"about_ca_system_score_codex":0.0023365705,"about_ca_system_score_gemma":0.02601984,"threshold_uncertainty_score":0.9998568},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4224057184","doi":"10.1136/bmjebm-2021-111866","title":"Which actionable statements qualify as good practice statements In Covid-19 guidelines? A systematic appraisal","year":2022,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":17,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Ottawa Hospital; Centre hospitalier universitaire de Québec; McMaster University; Impact; Cochrane; Bruyère; Hôpital de l'Enfant-Jésus; University of Ottawa","funders":"Canadian Institutes of Health Research; World Health Organization","keywords":"Guideline; Global Positioning System; Critical appraisal; Systematic review; Grading (engineering); MEDLINE; Quality (philosophy); Medicine; Computer science; Alternative medicine; Engineering; Political science","score_opus":0.7104620991194237,"score_gpt":0.7006955760346055,"score_spread":0.00976652308481818,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4224057184","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000008126289,0.9019367,0.00422244,0.06966313,0.002733854,0.020266403,0.00013100375,0.00026767998,0.0007706812],"genre_scores_gemma":[0.00004165208,0.9543876,0.004118689,0.027284844,0.0014956995,0.007922586,0.0034023495,0.00027112232,0.0010754423],"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","domain_scores_codex":[0.9668465,0.008379072,0.014961173,0.0022040915,0.006317417,0.0012917235],"domain_scores_gemma":[0.8865637,0.09363968,0.012198361,0.0031625524,0.0025536604,0.0018820332],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow"],"category_scores_codex":[0.048434243,0.0015001143,0.007043426,0.0019563648,0.00036512507,0.00013240616,0.0009906235,0.0005960188,0.010611615],"category_scores_gemma":[0.6167063,0.0011811084,0.00065900333,0.0043221074,0.00020967562,0.0028129818,0.00044728784,0.0026434243,0.0007073942],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.0027591772,0.0017591082,0.00014446031,0.78280663,0.001994514,0.0021344933,0.00058772776,0.00032904514,0.0000034889533,0.0002716536,0.090923816,0.116285875],"study_design_scores_gemma":[0.0064916206,0.0019410149,0.0000032742707,0.25181672,0.009215738,0.0008145485,0.004847692,0.00026334808,3.219398e-7,0.0000567266,0.7238622,0.00068685383],"about_ca_topic_score_codex":0.008832488,"about_ca_topic_score_gemma":0.000627561,"teacher_disagreement_score":0.6329383,"about_ca_system_score_codex":0.0057391115,"about_ca_system_score_gemma":0.020521624,"threshold_uncertainty_score":0.9997748},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4224095215","doi":"10.1136/bmjebm-2022-111962","title":"Good or best practice statements: proposal for the operationalisation and implementation of GRADE guidance","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":164,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université Laval; Ottawa Hospital; Centre hospitalier universitaire de Québec; McMaster University; Impact; Cochrane; Bruyère; Hôpital de l'Enfant-Jésus; University of Ottawa","funders":"Canadian Institutes of Health Research; World Health Organization","keywords":"Grading (engineering); Guideline; Global Positioning System; Computer science; Fidelity; Process management; Best practice; Medical education; Knowledge management; Medicine; Business; Engineering; Political science","score_opus":0.5414977190713219,"score_gpt":0.5727772921882497,"score_spread":0.03127957311692775,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4224095215","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.02243711,0.010883607,0.06795789,0.8902171,0.00089559413,0.0068405257,0.0006155295,0.000025345182,0.00012728658],"genre_scores_gemma":[0.89310473,0.0012379232,0.037238713,0.057319194,0.0014362432,0.008106401,0.00037613328,0.00006406169,0.0011165744],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","domain_scores_codex":[0.99574035,0.0005720393,0.0028600635,0.00035611924,0.00026856843,0.00020287631],"domain_scores_gemma":[0.9899577,0.006935209,0.002501218,0.0003223797,0.00021285911,0.00007062926],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03090917,0.0001258853,0.00044566704,0.00016516255,0.00044991897,0.000028838927,0.00020398162,0.00002886714,0.0010548419],"category_scores_gemma":[0.01242347,0.00010640409,0.000039180526,0.00023227265,0.00011730555,0.00089539395,0.000043917273,0.000102197075,0.0000143331945],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011276512,0.0005499142,0.05465704,0.0028857936,0.00038934377,0.000003499483,0.013637711,0.0072267703,0.00020441988,0.7482204,0.16653872,0.004558735],"study_design_scores_gemma":[0.017789537,0.010122981,0.035240673,0.0017601955,0.00039472297,0.00006452195,0.13077813,0.063297525,0.00036690084,0.030227939,0.7088756,0.001081271],"about_ca_topic_score_codex":0.0026094485,"about_ca_topic_score_gemma":0.00035656392,"teacher_disagreement_score":0.87066764,"about_ca_system_score_codex":0.0004153823,"about_ca_system_score_gemma":0.0008746596,"threshold_uncertainty_score":0.9998583},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4283587010","doi":"10.1136/bmjebm-2022-111928","title":"Theory and practice of Bayesian and frequentist frameworks for network meta-analysis","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":114,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Joseph’s Healthcare Hamilton; McMaster University; Ted Rogers Centre for Heart Research; University Health Network; Impact","funders":"","keywords":"Frequentist inference; Bayesian probability; Computer science; Bayesian network; Bayesian statistics; Frequentist probability; Machine learning; Bayesian inference; Artificial intelligence; Data science","score_opus":0.7536134177763099,"score_gpt":0.5727628547311456,"score_spread":0.18085056304516423,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4283587010","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0008163686,0.1052002,0.8445493,0.045907747,0.0002561488,0.0023268159,0.000044576493,0.000007041049,0.00089177897],"genre_scores_gemma":[0.9014411,0.00034560135,0.07948804,0.012901153,0.00038538422,0.0010417858,0.000025989184,0.000027640046,0.004343335],"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9439968,0.03885422,0.008892325,0.0014199998,0.006467069,0.00036962263],"domain_scores_gemma":[0.8230204,0.16428775,0.0077852267,0.0032179875,0.0014185298,0.00027009443],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3995094,0.00033699517,0.006762949,0.00064535986,0.00038694934,0.00018612397,0.0010292049,0.00011089217,0.021722738],"category_scores_gemma":[0.30296808,0.00016150103,0.0024389538,0.0034472737,0.00036399468,0.00033262512,0.000197115,0.0003707439,0.000008795756],"study_design_candidate":"meta_analysis","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021018388,0.00033478858,0.02375129,0.0010863045,0.2366084,0.00011181547,0.0031976833,0.025432736,0.00018109167,0.20250556,0.46782258,0.036865894],"study_design_scores_gemma":[0.00095237355,0.0010436525,0.0024801556,0.00018318415,0.6646189,0.00008877558,0.0061720014,0.03261759,0.000012982391,0.11617012,0.17513403,0.000526203],"about_ca_topic_score_codex":0.00005985098,"about_ca_topic_score_gemma":0.000027413042,"teacher_disagreement_score":0.9006247,"about_ca_system_score_codex":0.000028801802,"about_ca_system_score_gemma":0.00012680996,"threshold_uncertainty_score":0.9791715},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4285794520","doi":"10.1136/bmjebm-2022-111952","title":"Adapt or die: how the pandemic made the shift from EBM to EBM+ more urgent","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":123,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"CARE Canada; Association of Professional Engineers and Geoscientists of Manitoba; Public Health Ontario; University of Toronto","funders":"National Institute for Health and Care Research; Canadian Institutes of Health Research; NIHR Oxford Biomedical Research Centre; National Medical Research Council; National Health and Medical Research Council; Wellcome Trust","keywords":"Context (archaeology); Checklist; Evidence-based medicine; Psychological intervention; Quality (philosophy); Yesterday; Evidence-based practice; Health care; Scientific evidence; Intervention (counseling); Psychology; Engineering ethics; MEDLINE; Management science; Medicine; Political science; Engineering; Alternative medicine; Nursing; Epistemology; History","score_opus":0.7945608969354264,"score_gpt":0.5892763489863074,"score_spread":0.205284547949119,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4285794520","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.16381684,0.0074916547,0.0008456003,0.8189709,0.0031062986,0.005269233,0.00011916333,0.00021308666,0.00016721874],"genre_scores_gemma":[0.70357436,0.0007164622,0.00021176618,0.2841518,0.0034809508,0.004016275,0.000061920065,0.000069931535,0.0037165645],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98208344,0.011985897,0.0013892185,0.00083805126,0.002528933,0.0011744546],"domain_scores_gemma":[0.95688134,0.038550705,0.0010505127,0.0022173221,0.00040686072,0.000893241],"candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.019468496,0.0004345082,0.00077418727,0.0001800305,0.003261043,0.00002909983,0.0016437665,0.00020485374,0.011239315],"category_scores_gemma":[0.049547777,0.00021213018,0.00012294402,0.0011293545,0.00041940107,0.00030331445,0.00075539143,0.0032818133,0.0003089907],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004640232,0.00019493976,0.1436288,0.000857466,0.00017085,0.0003430047,0.05556172,0.00018234906,0.0006965786,0.00066296064,0.76615983,0.026901279],"study_design_scores_gemma":[0.0016850911,0.0009553066,0.15025257,0.003417216,0.0002557151,0.000010733322,0.0682379,0.00037281492,0.000025318595,0.0005634782,0.7738438,0.00038005065],"about_ca_topic_score_codex":0.01650484,"about_ca_topic_score_gemma":0.006588056,"teacher_disagreement_score":0.5397575,"about_ca_system_score_codex":0.00096375716,"about_ca_system_score_gemma":0.0054677487,"threshold_uncertainty_score":0.99901766},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4291021075","doi":"10.1136/bmjebm-2022-111944","title":"Knowledge user survey and Delphi process to inform development of a new risk of bias tool to assess systematic reviews with network meta-analysis (RoB NMA tool)","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":19,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University; Simon Fraser University; Ottawa Hospital; University of Ottawa; University of Toronto; University of British Columbia; Cochrane; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Medical Research Council; National Institute for Health and Care Research; World Health Organization","keywords":"Delphi method; Knowledge translation; Systematic review; Delphi; Psychology; Process (computing); Quality (philosophy); Knowledge management; Computer science; MEDLINE; Political science; Artificial intelligence","score_opus":0.9019285677618601,"score_gpt":0.5686405846106061,"score_spread":0.3332879831512541,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4291021075","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.39569026,0.07893616,0.49201468,0.0035459676,0.00037169084,0.029011313,0.000096740805,0.000025328563,0.000307838],"genre_scores_gemma":[0.8768491,0.00023092265,0.10542043,0.0023840622,0.0001867432,0.005727638,0.000050843584,0.000063815794,0.009086456],"study_design_codex":"observational","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9250226,0.033307076,0.026928054,0.0018123849,0.012289418,0.0006404861],"domain_scores_gemma":[0.9333867,0.036996238,0.01813498,0.0057702586,0.005008185,0.0007036795],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.41700166,0.0007063499,0.017458824,0.0016341564,0.00033092106,0.000178627,0.002442498,0.00007214879,0.006381604],"category_scores_gemma":[0.23142092,0.00029640362,0.0021928845,0.017646827,0.00009410682,0.0003419352,0.0004016459,0.00025408066,0.00014554152],"study_design_candidate":"meta_analysis","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011949221,0.00054329203,0.39313582,0.063972086,0.14239858,0.000029185345,0.030016797,0.11507956,0.000043856802,0.0004340581,0.23107377,0.022078084],"study_design_scores_gemma":[0.0031367803,0.004124143,0.18855603,0.028176079,0.6358028,0.000060301336,0.012967232,0.016566005,0.0002083226,0.0004978776,0.106665514,0.0032389134],"about_ca_topic_score_codex":0.00025685708,"about_ca_topic_score_gemma":0.0031645352,"teacher_disagreement_score":0.49340424,"about_ca_system_score_codex":0.00012609802,"about_ca_system_score_gemma":0.0013637104,"threshold_uncertainty_score":0.9999488},"labels":[],"label_agreement":null},{"id":"W4297996183","doi":"10.1136/bmjebm-2022-112002","title":"Proof of concept: all-virtual guideline development workshops using GRADE during the COVID-19 pandemic","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Grading (engineering); Guideline; Medical education; Pandemic; Psychological intervention; Coronavirus disease 2019 (COVID-19); Psychology; Medicine; Nursing; Engineering","score_opus":0.548356336968319,"score_gpt":0.5763784541983684,"score_spread":0.028022117230049437,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4297996183","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8239501,0.00356341,0.012507189,0.15317193,0.001999903,0.0046311193,0.000015323878,0.000087833665,0.000073173156],"genre_scores_gemma":[0.97042346,0.00009055777,0.001461893,0.024490463,0.0011782486,0.0010923436,0.000050330036,0.000021737615,0.0011909904],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99275446,0.0023049237,0.0016744721,0.00042353195,0.0019198029,0.00092281087],"domain_scores_gemma":[0.99302405,0.0046262285,0.0007207884,0.00053316256,0.00035239328,0.00074338866],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.01595448,0.0001853378,0.00039762392,0.0002970219,0.002660115,0.0000056325516,0.0007034781,0.00008586417,0.004067687],"category_scores_gemma":[0.013070211,0.000118944256,0.00004763863,0.0011883291,0.00055001077,0.00016554931,0.00027935172,0.0010461735,0.000009228425],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0018967241,0.0004293027,0.3176646,0.0034032133,0.00010141043,0.00007499131,0.09630002,0.07755966,0.0077946186,0.00030756692,0.4768119,0.017656013],"study_design_scores_gemma":[0.008641548,0.0018823099,0.028703464,0.006128522,0.0001225461,0.00011628582,0.13292973,0.022195479,0.0016760586,0.0004491841,0.79627603,0.0008788172],"about_ca_topic_score_codex":0.0017921604,"about_ca_topic_score_gemma":0.00020395026,"teacher_disagreement_score":0.31946415,"about_ca_system_score_codex":0.002027466,"about_ca_system_score_gemma":0.017191956,"threshold_uncertainty_score":0.9986383},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4306181334","doi":"10.1136/bmjebm-2022-112065","title":"The Pandora’s Box of Evidence Synthesis and the case for a living Evidence Synthesis Taxonomy","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":24,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; St. Michael's Hospital; University of Toronto; McMaster University; Queen's University; Impact","funders":"National Health and Medical Research Council; University of Adelaide","keywords":"Context (archaeology); Pandemic; Nursing; Scope (computer science); Public relations; Psychology; Medical education; Business; Medicine; Coronavirus disease 2019 (COVID-19); Political science; Geography; Computer science","score_opus":0.7695973013225446,"score_gpt":0.5618726478439796,"score_spread":0.20772465347856495,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4306181334","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09481427,0.104466006,0.0052043786,0.7740754,0.0020537488,0.01891594,0.00009224699,0.00019519759,0.00018283162],"genre_scores_gemma":[0.9444512,0.009312398,0.0011103249,0.011390498,0.0009802452,0.032217458,9.415832e-7,0.00006326784,0.00047363457],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9701009,0.023907956,0.0025257324,0.0008852295,0.0014791568,0.0011010353],"domain_scores_gemma":[0.44358376,0.5511241,0.0023200666,0.0016695482,0.00092445815,0.00037806714],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.088950366,0.00045166595,0.0012449747,0.00025210582,0.0055225524,0.000044504093,0.0012246927,0.00018235904,0.0010280912],"category_scores_gemma":[0.5281563,0.00023593091,0.00023017792,0.0008683498,0.0017491578,0.0008889062,0.0007103605,0.0014949922,0.00001814769],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.04064043,0.0006093096,0.16404559,0.041859098,0.0010609297,0.0016416807,0.0379085,0.0002609977,0.001410557,0.015557555,0.19047672,0.50452864],"study_design_scores_gemma":[0.0075249486,0.005039957,0.021874987,0.32830215,0.007608755,0.0017656337,0.3118049,0.015905079,0.0012891331,0.0026930396,0.29346487,0.0027265495],"about_ca_topic_score_codex":0.014886424,"about_ca_topic_score_gemma":0.002445777,"teacher_disagreement_score":0.849637,"about_ca_system_score_codex":0.0006320087,"about_ca_system_score_gemma":0.0037190241,"threshold_uncertainty_score":0.9998851},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W4309247470","doi":"10.1136/bmjebm-2022-112064","title":"What is the place in therapy for nirmatrelvir/ritonavir?","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; McGill University","funders":"","keywords":"Ritonavir; Medicine; Antiretroviral therapy; Virology; Human immunodeficiency virus (HIV); Viral load","score_opus":0.7470211808498358,"score_gpt":0.6191185209135557,"score_spread":0.12790265993628014,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4309247470","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.11882874,0.032600217,0.00027352944,0.8352851,0.004554646,0.008080705,0.000035567875,0.00011249931,0.00022896717],"genre_scores_gemma":[0.43138474,0.01281079,0.00046940419,0.5328875,0.0030686164,0.014535058,0.00007607216,0.00011457732,0.004653213],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98716784,0.008336506,0.0014946267,0.00061946164,0.0013664521,0.0010150891],"domain_scores_gemma":[0.9608843,0.03652219,0.0008532123,0.001138675,0.00034494395,0.00025668018],"candidate_categories":["sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.028681092,0.00030240076,0.0006205857,0.00026422896,0.0019185839,0.000023701421,0.00084382494,0.00017630395,0.0043300916],"category_scores_gemma":[0.0076623587,0.00020054022,0.00011598624,0.0008995591,0.000263299,0.0009267602,0.00021659488,0.0020116954,0.00011517332],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007974757,0.00047567434,0.094718575,0.0024766747,0.00008322581,0.000097490636,0.05815914,0.00020086348,0.000493013,0.0032235591,0.7388972,0.0931998],"study_design_scores_gemma":[0.0057437057,0.0015537271,0.013608418,0.0054162527,0.000044460623,0.000009006788,0.07480723,0.0010929476,0.0000706343,0.0017291651,0.89558643,0.000338016],"about_ca_topic_score_codex":0.0042703734,"about_ca_topic_score_gemma":0.0009067096,"teacher_disagreement_score":0.312556,"about_ca_system_score_codex":0.0008170431,"about_ca_system_score_gemma":0.002977645,"threshold_uncertainty_score":0.99938077},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4312061410","doi":"10.1136/bmjebm-2022-112024","title":"Retrieval barriers in individual participant data reviews with network meta-analysis","year":2022,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Randomized controlled trial; MEDLINE; Medicine; Psychological intervention; Systematic review; Unavailability; Meta-analysis; Data sharing; Computer science; Alternative medicine; Nursing; Political science; Surgery; Engineering","score_opus":0.9577443300487736,"score_gpt":0.6060650932160673,"score_spread":0.3516792368327063,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4312061410","genre_codex":"review","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.021422202,0.71632934,0.06826182,0.17107318,0.0016144994,0.018758798,0.00066939543,0.00007244538,0.0017983387],"genre_scores_gemma":[0.95490634,0.00077552046,0.013396424,0.021303542,0.0008912636,0.0016626634,0.00051349227,0.00006480582,0.0064859604],"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","domain_scores_codex":[0.8788678,0.07041906,0.022745205,0.0037911034,0.023037111,0.001139741],"domain_scores_gemma":[0.9367366,0.027516259,0.013379351,0.020393385,0.00097644434,0.0009979856],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5393995,0.00068469456,0.015017841,0.0014656595,0.00048126286,0.00036140514,0.0077279327,0.00008111708,0.14160162],"category_scores_gemma":[0.1982478,0.00028676708,0.003181848,0.019505166,0.00033518983,0.0006582847,0.0010160622,0.0007112744,0.000390477],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00044755504,0.00011141769,0.0770483,0.00015663115,0.077507235,0.00035558923,0.0013473362,0.06867061,0.000010261065,0.00043356325,0.77155066,0.0023608315],"study_design_scores_gemma":[0.0008580069,0.0006384046,0.006616291,0.00021836982,0.5391726,0.000028719182,0.0023007526,0.031664297,0.0000029896696,0.00068893464,0.41720983,0.0006007861],"about_ca_topic_score_codex":0.00018076428,"about_ca_topic_score_gemma":0.0006441035,"teacher_disagreement_score":0.93348414,"about_ca_system_score_codex":0.0000882886,"about_ca_system_score_gemma":0.00079793285,"threshold_uncertainty_score":0.99995846},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4315475599","doi":"10.1136/bmjebm-2022-112117","title":"Expanding the measurement of overdiagnosis in the context of disease precursors and risk factors","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Agency of Canada; Université de Montréal; Université Laval; University of Alberta; Public Health Ontario; University of Toronto; McGill University","funders":"","keywords":"Overdiagnosis; Context (archaeology); Disease; Medicine; Intensive care medicine; Risk analysis (engineering); Internal medicine; History","score_opus":0.11546235201575854,"score_gpt":0.35582039208908084,"score_spread":0.2403580400733223,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4315475599","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.98700637,0.0035251244,0.000093086644,0.008091069,0.00015936133,0.0010193188,0.000008019583,0.000027768274,0.00006987535],"genre_scores_gemma":[0.9978577,0.0015925458,0.000009011265,0.0002517543,0.00012609891,0.00013982161,0.0000036967904,0.000009098454,0.000010260317],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.998041,0.00025300172,0.00039624877,0.00019577061,0.0009526886,0.00016132872],"domain_scores_gemma":[0.9969453,0.002243579,0.00023073796,0.00033077324,0.00015403696,0.00009553107],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0031951468,0.000127501,0.00033627497,0.00020908988,0.000069165704,0.000004400582,0.00010627399,0.00003829017,0.0000302906],"category_scores_gemma":[0.0065972507,0.00006206514,0.00008100956,0.0007677095,0.000274144,0.000057553305,0.000023570085,0.00018332605,0.0000010233359],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.008207529,0.00008569396,0.9571326,0.00064208603,0.000089772955,0.000019282743,0.0072225216,0.00018502715,0.0014387934,0.000030009565,0.004127109,0.02081962],"study_design_scores_gemma":[0.0018923692,0.0028003806,0.96938115,0.006283843,0.0004362695,0.000002018532,0.008976921,0.0009648236,0.008850016,0.00008324794,0.00024692516,0.000082059036],"about_ca_topic_score_codex":0.0026348492,"about_ca_topic_score_gemma":0.00031667587,"teacher_disagreement_score":0.02073756,"about_ca_system_score_codex":0.00007925486,"about_ca_system_score_gemma":0.00017354194,"threshold_uncertainty_score":0.78980017},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4318973081","doi":"10.1136/bmjebm-2022-112053","title":"Different meta-analysis methods can change judgements about imprecision of effect estimates: a meta-epidemiological study","year":2023,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Meta-analysis; Statistics; Random effects model; Mathematics; Medicine; Restricted maximum likelihood; Maximum likelihood; Econometrics; Internal medicine","score_opus":0.9820049695252123,"score_gpt":0.7362051535903333,"score_spread":0.24579981593487898,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4318973081","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000026479369,0.941686,0.035551738,0.0012982978,0.00088342046,0.020254027,0.00019375369,0.000051734714,0.00005459706],"genre_scores_gemma":[0.0002460588,0.96465504,0.011140316,0.0007026578,0.000682681,0.019788336,0.00023435002,0.00014404584,0.002406525],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.6700647,0.24845043,0.05377088,0.006069376,0.020360285,0.001284344],"domain_scores_gemma":[0.6266546,0.3138774,0.040614326,0.015240023,0.0026586838,0.00095495436],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.61449736,0.0036427567,0.15125279,0.006392405,0.00028243288,0.00038423185,0.008078281,0.00077824167,0.035556857],"category_scores_gemma":[0.536687,0.001118133,0.07427885,0.016860403,0.00049318926,0.00029865943,0.0011139681,0.0011487169,0.0009962902],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000035257755,0.00030827217,0.0009351403,0.013056746,0.61072904,0.000087566026,0.00019709743,0.000060264258,5.269549e-7,0.000029277599,0.011220056,0.36334074],"study_design_scores_gemma":[0.00038038517,0.0015729293,0.00061702466,0.005560864,0.8966656,0.000006883195,0.00011040448,0.0014388227,0.000001173906,0.00035358602,0.09257792,0.0007144275],"about_ca_topic_score_codex":0.00064650824,"about_ca_topic_score_gemma":0.00025075715,"teacher_disagreement_score":0.3626263,"about_ca_system_score_codex":0.0002750828,"about_ca_system_score_gemma":0.00035088262,"threshold_uncertainty_score":0.99978155},"labels":[{"model":"gemma","categories":["metaresearch","metaepi_broad","metaepi_narrow"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4322758031","doi":"10.1136/bmjebm-2022-112098","title":"Decision Analysis in SHared decision making for Thromboprophylaxis during Pregnancy (DASH-TOP): a sequential explanatory mixed-methods pilot study","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":12,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Mount Sinai Hospital; McMaster University; Impact","funders":"","keywords":"Decision aids; Intervention (counseling); Decision analysis; Medicine; Psychology; Nursing; Alternative medicine","score_opus":0.4146435220118736,"score_gpt":0.5623066713106735,"score_spread":0.14766314929879987,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4322758031","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.919865,0.008456768,0.049900454,0.003546375,0.0032980472,0.01423529,0.000074276046,0.00056844205,0.000055313474],"genre_scores_gemma":[0.95645905,0.00036320882,0.035397053,0.00044034474,0.0003924824,0.0066521517,0.00012028982,0.00009110058,0.000084318366],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.98814434,0.0047582122,0.0031126912,0.0012609995,0.0015448086,0.0011789507],"domain_scores_gemma":[0.9686132,0.026298666,0.0011639999,0.002627309,0.00094329385,0.00035353116],"candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.0058496026,0.00050077913,0.001422924,0.0030856119,0.0013323467,0.000025207788,0.0012238644,0.0002695943,0.0005926988],"category_scores_gemma":[0.026294772,0.00042621052,0.00022015061,0.006153266,0.00015879198,0.00053204386,0.000644527,0.0010671298,0.00010505332],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0062555736,0.0007843478,0.8581892,0.003973675,0.00038309986,0.00021844986,0.022740996,0.002228941,0.0021233023,0.000090380854,0.003468878,0.099543154],"study_design_scores_gemma":[0.009018635,0.003937401,0.8891975,0.053204007,0.0005485541,0.00000239935,0.02688409,0.014254937,0.00011138366,0.0020288178,0.00019806143,0.0006141958],"about_ca_topic_score_codex":0.0012745921,"about_ca_topic_score_gemma":0.0040605473,"teacher_disagreement_score":0.09892896,"about_ca_system_score_codex":0.0012007048,"about_ca_system_score_gemma":0.0010980915,"threshold_uncertainty_score":0.99996775},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4323542543","doi":"10.1136/bmjebm-2022-112100","title":"Proposed triggers for retiring a living systematic review","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Construct (python library); Computer science; Certainty; Actuarial science; Risk analysis (engineering); Business","score_opus":0.8168270424800806,"score_gpt":0.6492158980082676,"score_spread":0.16761114447181302,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4323542543","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0007034876,0.031929586,0.81172496,0.096880935,0.0036847922,0.051973984,0.0000602215,0.0022379218,0.00080413313],"genre_scores_gemma":[0.0064296466,0.009068269,0.95867336,0.01251451,0.0028120603,0.007239824,0.000009766056,0.0002659906,0.002986588],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.9923989,0.0023178216,0.002875955,0.00061576185,0.0011948231,0.0005967488],"domain_scores_gemma":[0.67636704,0.3209464,0.0009504853,0.0010399802,0.0004232047,0.00027289696],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.045198187,0.00035163906,0.0025772073,0.00023442434,0.00013670388,0.000023669767,0.00054823887,0.00016092758,0.00038924426],"category_scores_gemma":[0.8755799,0.00023113948,0.00034098767,0.0012014502,0.00027240068,0.00010769152,0.00007889716,0.0002872524,0.00012413082],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00013007801,0.000055978795,0.000025719963,0.93642116,0.00011050589,0.00006770914,0.00010326137,0.0000023516395,0.0002719453,0.010199597,0.051352415,0.0012592617],"study_design_scores_gemma":[0.0009989426,0.0006070661,0.000021055195,0.86160856,0.0017509517,0.000009526775,0.00010986224,0.0018875307,0.00021926625,0.13241112,0.0000876646,0.000288415],"about_ca_topic_score_codex":0.0000070948495,"about_ca_topic_score_gemma":0.0000015834838,"teacher_disagreement_score":0.8303817,"about_ca_system_score_codex":0.0001427996,"about_ca_system_score_gemma":0.00029231503,"threshold_uncertainty_score":0.9831694},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4328055071","doi":"10.1136/bmjebm-2022-112005","title":"US Food and Drug Administration regulatory reviewer disagreements and postmarket safety actions among new therapeutics","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Pharmaceutical Economics and Policy","field":"Economics, Econometrics and Finance","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Dalhousie University","funders":"Canadian Institutes of Health Research","keywords":"Medicine; Food and drug administration; Drug; Pharmacology; Patient safety; Health care","score_opus":0.18732288818352935,"score_gpt":0.37053351688652847,"score_spread":0.18321062870299912,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4328055071","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8485906,0.024121726,0.00047450932,0.12014555,0.0010704914,0.0012633255,0.00015090278,0.000105817664,0.0040770946],"genre_scores_gemma":[0.9686149,0.021708995,0.000120822355,0.0074289744,0.00063491607,0.000038257556,0.00003851426,0.000026173215,0.0013884492],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99842966,0.000048186743,0.00075354485,0.00041342914,0.0000695785,0.00028560907],"domain_scores_gemma":[0.9985431,0.00043509208,0.00032039738,0.00034501374,0.000040475596,0.00031591882],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016810895,0.00019274598,0.00043335874,0.00021947871,0.00014831254,0.000045724955,0.00012639133,0.00008246075,0.00038132234],"category_scores_gemma":[0.0006690101,0.00018725758,0.000057484882,0.0003061307,0.00022858707,0.0003301934,0.000050775907,0.00016930506,0.00007946143],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0007603352,0.00016268932,0.6631355,0.0011288218,0.00051755784,0.000022745913,0.0012568497,0.000078303514,0.00019253927,0.08983959,0.16411734,0.078787684],"study_design_scores_gemma":[0.0021503882,0.000454872,0.6230991,0.00094010896,0.00010096082,0.0000073656456,0.00008674299,0.0050995057,0.00012607718,0.012969443,0.35452014,0.00044529908],"about_ca_topic_score_codex":0.00014172471,"about_ca_topic_score_gemma":0.000120115365,"teacher_disagreement_score":0.1904028,"about_ca_system_score_codex":0.00008133269,"about_ca_system_score_gemma":0.000105547,"threshold_uncertainty_score":0.76361424},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4366386724","doi":"10.1136/bmjebm-2022-112070","title":"Rapid Reviews Methods Series: Involving patient and public partners, healthcare providers and policymakers as knowledge users","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Mental Health and Patient Involvement","field":"Health Professions","cited_by":48,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Canadian Agency for Drugs and Technologies in Health; Public Health Ontario; University of Toronto; St. Michael's Hospital; Cochrane; Public Health Agency of Canada; University of Ottawa","funders":"","keywords":"Timeline; Health care; Commission; Knowledge translation; Psychology; Knowledge management; Public relations; Management science; Computer science; Business; Political science; Engineering; Mathematics","score_opus":0.6031864510411815,"score_gpt":0.5799916421850461,"score_spread":0.02319480885613545,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366386724","genre_codex":"commentary","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.258769,0.3104999,0.0003951119,0.39451313,0.005428823,0.027100762,0.000061310886,0.0008907566,0.0023411582],"genre_scores_gemma":[0.2743151,0.40422675,0.022631912,0.26988566,0.004321454,0.016724505,0.0013448485,0.00039518726,0.006154581],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9904715,0.0049237465,0.0018518257,0.000790769,0.0005938545,0.0013683083],"domain_scores_gemma":[0.99269044,0.003935236,0.00079247373,0.0006636499,0.00032404665,0.0015941488],"candidate_categories":["metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.009362301,0.0004527502,0.00096644915,0.0005871428,0.0013038561,0.000021228967,0.00020074565,0.000259814,0.0003374251],"category_scores_gemma":[0.0057132333,0.00033660102,0.000074375144,0.0012644324,0.00045176988,0.0005382662,0.00023503136,0.00071676815,0.00018857689],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00095243956,0.00013601535,0.14637071,0.05403343,0.000068334135,0.00004701844,0.055924833,0.0000015672067,0.0003208958,0.004351052,0.24700503,0.49078867],"study_design_scores_gemma":[0.0028489768,0.004454576,0.004048455,0.043863464,0.00009513939,0.000009804124,0.059806712,0.0004965396,0.000076521355,0.00098081,0.88275915,0.00055987755],"about_ca_topic_score_codex":0.0033075605,"about_ca_topic_score_gemma":0.0011844067,"teacher_disagreement_score":0.6357541,"about_ca_system_score_codex":0.00048545274,"about_ca_system_score_gemma":0.0018406057,"threshold_uncertainty_score":0.9999963},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4366386735","doi":"10.1136/bmjebm-2022-112185","title":"Rapid reviews methods series: Guidance on team considerations, study selection, data extraction and risk of bias assessment","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":78,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Data extraction; Correctness; Computer science; Systematic review; Completeness (order theory); Selection (genetic algorithm); Data mining; Information retrieval; MEDLINE; Artificial intelligence; Mathematics; Algorithm","score_opus":0.9188994641379558,"score_gpt":0.6765426507253051,"score_spread":0.2423568134126507,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366386735","genre_codex":"methods","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":"methods","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09856418,0.07306467,0.7290888,0.05918693,0.0041638683,0.031031149,0.00018275403,0.00018778881,0.0045298627],"genre_scores_gemma":[0.33114946,0.06655069,0.58402014,0.0032504715,0.0018980416,0.0020144144,0.00014789091,0.00009240501,0.010876506],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9202019,0.060054325,0.012331255,0.0017326644,0.005403225,0.00027664605],"domain_scores_gemma":[0.9172035,0.061807558,0.011173549,0.007394707,0.0021665823,0.00025414175],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.46440104,0.00037769933,0.003930617,0.00091476477,0.0003361525,0.00023076619,0.0009927761,0.00007931896,0.0049177464],"category_scores_gemma":[0.49952415,0.00017926413,0.00034612912,0.0036646624,0.00018739027,0.0007935393,0.00015516707,0.0003018303,0.00028057705],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000050336075,0.00017092816,0.13952224,0.00029740293,0.00030537075,0.000008573134,0.0005091123,0.00031420926,0.0005449297,0.00027231695,0.72088546,0.13711913],"study_design_scores_gemma":[0.0015505345,0.0030490572,0.26205298,0.0044794083,0.0032455735,0.00006005912,0.006860294,0.060303167,0.0003596329,0.0034273562,0.6540655,0.0005464271],"about_ca_topic_score_codex":0.00017543929,"about_ca_topic_score_gemma":0.00021057934,"teacher_disagreement_score":0.23258528,"about_ca_system_score_codex":0.00006013089,"about_ca_system_score_gemma":0.0003792458,"threshold_uncertainty_score":0.9959919},"labels":[],"label_agreement":null},{"id":"W4366386740","doi":"10.1136/bmjebm-2022-112136","title":"Implementing hierarchical network meta-analysis incorporating exchangeable dose effects compared to standard hierarchical network meta-analysis","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; University of Calgary; St. Michael's Hospital","funders":"Medical Research Council","keywords":"Meta-analysis; Network analysis; Computer science; Multilevel model; Medicine; Internal medicine; Engineering; Machine learning","score_opus":0.8066750011977708,"score_gpt":0.5551599690728524,"score_spread":0.25151503212491844,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366386740","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009967649,0.037485387,0.8376179,0.10069394,0.0010030415,0.0108622275,0.00023040948,0.00030900704,0.0018304841],"genre_scores_gemma":[0.86120826,0.00010433775,0.10540837,0.015253887,0.0034460458,0.0039499844,0.0005708569,0.00014834704,0.009909901],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.8958602,0.050241165,0.023650242,0.004777538,0.022509374,0.002961467],"domain_scores_gemma":[0.8955351,0.07967909,0.010304482,0.009614623,0.003000953,0.0018657416],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","bibliometrics","sts","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.40647015,0.0016082097,0.035970878,0.0047519486,0.0014373958,0.0014296737,0.0045171417,0.00030231595,0.04573487],"category_scores_gemma":[0.10309399,0.00077990664,0.029638696,0.062264226,0.000441193,0.0005983483,0.0012716099,0.0009766486,0.0017659649],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00014976587,0.00003986346,0.009009805,0.00018203499,0.57204187,0.00016882199,0.0003241336,0.24374339,0.000024298792,0.0011936581,0.17190634,0.0012160101],"study_design_scores_gemma":[0.00052514154,0.00033873494,0.0035680342,0.000078372446,0.8590148,0.0000040775553,0.00023650978,0.1056609,0.000013764686,0.0069071497,0.022962939,0.00068958045],"about_ca_topic_score_codex":0.00037673072,"about_ca_topic_score_gemma":0.0023490626,"teacher_disagreement_score":0.85124063,"about_ca_system_score_codex":0.0001918039,"about_ca_system_score_gemma":0.00048373302,"threshold_uncertainty_score":0.9998626},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"simulation_or_modeling","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4366386744","doi":"10.1136/bmjebm-2022-112111","title":"Rapid reviews methods series: Guidance on assessing the certainty of evidence","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":30,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Public Health Agency","keywords":"Grading (engineering); Systematic review; Computer science; Rating system; Certainty; Management science; Medicine; Actuarial science; MEDLINE; Engineering; Mathematics; Business; Political science","score_opus":0.9546345879208633,"score_gpt":0.6949051985894442,"score_spread":0.2597293893314191,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4366386744","genre_codex":"commentary","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00785054,0.1790927,0.3232855,0.46836114,0.0040290924,0.010495954,0.000012848418,0.00011124884,0.0067610084],"genre_scores_gemma":[0.35981387,0.07536746,0.37034473,0.0829935,0.0070663923,0.0045592785,0.000043405915,0.00028733624,0.09952403],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.881878,0.08236209,0.020710647,0.002047578,0.012269771,0.00073188444],"domain_scores_gemma":[0.7892361,0.17696731,0.01677387,0.013092375,0.0034972574,0.0004330639],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.56694776,0.0005660722,0.0059288456,0.0007524002,0.00034081205,0.00034600633,0.003960373,0.00017041242,0.009986703],"category_scores_gemma":[0.7011788,0.00018533132,0.0018087956,0.007838656,0.00064585794,0.0007350892,0.00020564979,0.0005176863,0.002264584],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000052739488,0.000021567594,0.0032093665,0.00077530777,0.00008857375,0.000020650807,0.0006504897,0.000473132,0.0025001296,0.0009153329,0.6121005,0.37919223],"study_design_scores_gemma":[0.00038429385,0.00066283956,0.017379332,0.032320533,0.000745789,0.00002351068,0.0022970878,0.015454623,0.0014488541,0.0047357264,0.9241154,0.00043200105],"about_ca_topic_score_codex":0.000055478868,"about_ca_topic_score_gemma":0.000012387487,"teacher_disagreement_score":0.38536763,"about_ca_system_score_codex":0.00009005307,"about_ca_system_score_gemma":0.00043896722,"threshold_uncertainty_score":0.99851227},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4367675492","doi":"10.1136/bmjebm-2023-112343","title":"Correspondence on ‘The limits of shared decision making’ by Elwyn<i>et al</i>: We call for a more nuanced approach","year":2023,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Carleton University; University of Toronto","funders":"","keywords":"Computer science; Psychology; Sociology","score_opus":0.5148313715036523,"score_gpt":0.52626582144505,"score_spread":0.01143444994139764,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4367675492","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00018729875,0.008758619,0.006210134,0.9664044,0.002775841,0.013988601,0.0010692375,0.00031554158,0.00029036988],"genre_scores_gemma":[0.002799512,0.0018396255,0.0066496474,0.9663448,0.001690705,0.014222489,0.0018302037,0.00029171456,0.0043313215],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.98698854,0.004134198,0.003146013,0.001413785,0.0029818448,0.0013356073],"domain_scores_gemma":[0.9058151,0.08499533,0.00306967,0.0040599057,0.0018190979,0.00024089428],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0047433823,0.0008797338,0.0017445247,0.00073493615,0.0009371104,0.000018978588,0.002977427,0.0013529639,0.00040802703],"category_scores_gemma":[0.045254387,0.0006004686,0.00033680117,0.0013807239,0.00070027454,0.00022276462,0.00038558143,0.005778041,0.0002394396],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019519127,0.000062889,0.00038720385,0.0049156994,0.000050248058,0.000022673808,0.0036817407,0.00006800273,0.00012585391,0.000092557864,0.986985,0.001656218],"study_design_scores_gemma":[0.0024585696,0.0012232177,0.00040339804,0.12631643,0.00015683689,0.0000037984103,0.0022437156,0.004443754,0.000030369829,0.0005752971,0.86155224,0.0005923711],"about_ca_topic_score_codex":0.00045653098,"about_ca_topic_score_gemma":0.00006177389,"teacher_disagreement_score":0.12543276,"about_ca_system_score_codex":0.0007843119,"about_ca_system_score_gemma":0.0031722866,"threshold_uncertainty_score":0.9999435},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4376639745","doi":"10.1136/bmjebm-2022-112215","title":"Catalogue of bias: novelty bias","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Computational Drug Discovery Methods","field":"Computer Science","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Novelty; Computer science; Psychology; Social psychology","score_opus":0.3599031462954102,"score_gpt":0.4300701984081084,"score_spread":0.07016705211269819,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4376639745","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.100765176,0.0010086074,0.84708554,0.047732487,0.0016818065,0.00075516844,0.000024100824,0.0005031031,0.00044401572],"genre_scores_gemma":[0.9334265,0.00012499481,0.062391527,0.0028614087,0.0006511437,0.00009830346,0.00009084016,0.000028674838,0.000326588],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9965537,0.00052988215,0.0007415015,0.00053972367,0.0012688644,0.0003663235],"domain_scores_gemma":[0.9895143,0.008581285,0.00035723072,0.0010019455,0.00035995047,0.00018529345],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0057938322,0.00021967408,0.00046597538,0.0006041042,0.00007606278,0.00002924485,0.0011921971,0.00007005052,0.000052700398],"category_scores_gemma":[0.011458035,0.00017171213,0.00012072941,0.0029869555,0.0002709154,0.0005902924,0.00029735622,0.00017060358,0.00015615099],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00039527327,0.00048460532,0.031046167,0.0021265224,0.00022448135,0.00089650083,0.0050653736,0.25579897,0.0130165145,0.14578289,0.23159899,0.31356373],"study_design_scores_gemma":[0.0028425597,0.001198821,0.17838286,0.006978045,0.00008137389,0.00007054281,0.00025594461,0.75640184,0.017391456,0.025232479,0.010351803,0.0008122888],"about_ca_topic_score_codex":0.0005438872,"about_ca_topic_score_gemma":0.00003476648,"teacher_disagreement_score":0.83266133,"about_ca_system_score_codex":0.000078852296,"about_ca_system_score_gemma":0.00080016843,"threshold_uncertainty_score":0.99686885},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4378806110","doi":"10.1136/bmjebm-2022-112200","title":"Plain language summaries: enhancing patient-centred care and improving accessibility of health research","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Literacy and Information Accessibility","field":"Health Professions","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ontario Stroke Network; University Health Network; University of British Columbia","funders":"","keywords":"Plain language; Health care; Medicine; Computer science; Nursing; Political science","score_opus":0.264962570399348,"score_gpt":0.5450406684158028,"score_spread":0.2800780980164548,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4378806110","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.96385854,0.017012915,0.0005190259,0.0122812595,0.0007890886,0.004426777,0.00009521916,0.0002527399,0.0007644233],"genre_scores_gemma":[0.99432284,0.0003536866,0.0007819836,0.0036298593,0.00029838548,0.00029113502,0.00018862062,0.00002006323,0.00011345208],"study_design_codex":"qualitative","study_design_gemma":"qualitative","domain_scores_codex":[0.9901275,0.0038468062,0.0032118831,0.0005201745,0.0012222122,0.0010714353],"domain_scores_gemma":[0.98768497,0.008544801,0.0012020875,0.0008425145,0.0012730625,0.00045259428],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.014220525,0.00020660339,0.0006960557,0.000580692,0.0012979634,0.000018005245,0.00031809328,0.00020068104,0.0005349738],"category_scores_gemma":[0.031726193,0.00015233361,0.000049121823,0.001577327,0.00041477065,0.0009636683,0.0002803349,0.0010174176,0.00005815681],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00077151164,0.000059525773,0.23361802,0.039786402,0.000013216849,0.00001092264,0.37883702,0.000011136977,0.0011345492,0.00019682845,0.015993847,0.32956702],"study_design_scores_gemma":[0.0037773978,0.0016884826,0.14142285,0.033519305,0.000021580407,0.0000010629266,0.8020487,0.004830457,0.0014932471,0.00025443104,0.010587525,0.00035494994],"about_ca_topic_score_codex":0.0054331888,"about_ca_topic_score_gemma":0.00082484435,"teacher_disagreement_score":0.4232117,"about_ca_system_score_codex":0.0005657019,"about_ca_system_score_gemma":0.0035006634,"threshold_uncertainty_score":0.9983017},"labels":[{"model":"gemma","categories":["scholarly_communication"],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["scholarly_communication","open_science"],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4384663396","doi":"10.1136/bmjebm-2022-111921","title":"Reducing the residue of retractions in evidence synthesis: ways to minimise inappropriate citation and use of retracted data","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Academic integrity and plagiarism","field":"Social Sciences","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Regina","funders":"Alfred P. Sloan Foundation","keywords":"Citation; Systematic review; MEDLINE; Computer science; Medical literature; Data science; Medicine; Political science; Library science; Law","score_opus":0.35376102033369716,"score_gpt":0.42625773168766457,"score_spread":0.0724967113539674,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4384663396","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8193047,0.0012706527,0.0014789435,0.17554834,0.00047694048,0.0015004138,0.0000482213,0.00007336167,0.00029839305],"genre_scores_gemma":[0.99353236,0.0037545827,0.001726786,0.00045450253,0.00032080803,0.000054932792,0.000012738324,0.000010027705,0.00013327984],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99664915,0.0009135688,0.0007529228,0.00038326147,0.0010177542,0.00028334794],"domain_scores_gemma":[0.9761715,0.022236042,0.00041387478,0.00067092676,0.0003791818,0.0001285061],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.017688446,0.00012242136,0.00034043385,0.00041944496,0.00022033109,0.00002187372,0.00063100987,0.0003694937,0.000078605284],"category_scores_gemma":[0.21192709,0.000088419074,0.00002738354,0.0022736387,0.00064676267,0.0009975825,0.00010039895,0.00082166685,0.0000073705546],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0061907796,0.00032289492,0.0907215,0.002982691,0.00021657348,0.00019624906,0.23659325,0.002952592,0.12369359,0.022835374,0.3811357,0.13215879],"study_design_scores_gemma":[0.001567344,0.0011562422,0.7119565,0.12707621,0.0008772072,0.000015947766,0.092896126,0.018432207,0.022527112,0.0067342985,0.015739854,0.0010209342],"about_ca_topic_score_codex":0.02823934,"about_ca_topic_score_gemma":0.0033047737,"teacher_disagreement_score":0.621235,"about_ca_system_score_codex":0.00008541364,"about_ca_system_score_gemma":0.00067812146,"threshold_uncertainty_score":0.97823167},"labels":[{"model":"gemma","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4385063979","doi":"10.1136/bmjebm-2023-112270","title":"High-value care education can learn from the evidence-based medicine movement: moving beyond competencies and curricula to culture","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Health Sciences Centre; Women in Science and Engineering Newfoundland and Labrador; University of Toronto; Sunnybrook Health Science Centre","funders":"","keywords":"Curriculum; Movement (music); Value (mathematics); Medical education; Medicine; Pedagogy; Psychology; Computer science; Art","score_opus":0.46851158345077826,"score_gpt":0.5371929324341386,"score_spread":0.06868134898336031,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385063979","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.336037,0.02780373,0.00044526928,0.6266661,0.0035018788,0.0048306966,0.00008511438,0.00039289007,0.00023728878],"genre_scores_gemma":[0.7076984,0.0044508064,0.0022974017,0.27363926,0.007746851,0.0019122689,0.0005871347,0.00011415809,0.0015537218],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.98755926,0.005916906,0.001863488,0.0011899204,0.002323535,0.0011468912],"domain_scores_gemma":[0.97236913,0.022142297,0.0011128836,0.0015364517,0.001845317,0.0009939008],"candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.009942932,0.0006401371,0.0011030679,0.00047744857,0.0019818246,0.000044726345,0.00088902895,0.0004041216,0.0008952202],"category_scores_gemma":[0.04945205,0.0003957339,0.00009615788,0.002214546,0.00075046986,0.00065963186,0.00029966078,0.0017494755,0.00025455997],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009755588,0.00015389276,0.34857857,0.0069619096,0.0001453152,0.000118047355,0.11456458,0.0009039024,0.007728778,0.009799722,0.4499579,0.060111813],"study_design_scores_gemma":[0.0045432,0.0026772514,0.41615945,0.16677438,0.00090583833,0.0000050448198,0.28524086,0.0014618983,0.00043692064,0.0039657694,0.11655247,0.001276935],"about_ca_topic_score_codex":0.099400446,"about_ca_topic_score_gemma":0.009812913,"teacher_disagreement_score":0.37166142,"about_ca_system_score_codex":0.0010207223,"about_ca_system_score_gemma":0.0048779435,"threshold_uncertainty_score":0.99984944},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4385064669","doi":"10.1136/bmjebm-2023-112344","title":"Ten resources for understanding bias in health research: EBM live workshop 2022","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia; University of Alberta; University of Ottawa; University of Toronto","funders":"","keywords":"Psychology","score_opus":0.8979152418329284,"score_gpt":0.6624939927835675,"score_spread":0.2354212490493609,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385064669","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12086365,0.0038028695,0.0027404595,0.85991716,0.002329306,0.008990367,0.000019168585,0.000232161,0.0011048298],"genre_scores_gemma":[0.94476104,0.0065671396,0.0011005589,0.017916031,0.006121409,0.005891385,0.000115576375,0.00008330924,0.01744355],"study_design_codex":"not_applicable","study_design_gemma":"qualitative","domain_scores_codex":[0.9889951,0.004434444,0.0013569676,0.00067139935,0.0020890182,0.0024530892],"domain_scores_gemma":[0.94800144,0.049401853,0.00036698705,0.000598919,0.0005414036,0.0010894155],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.07743387,0.00018141133,0.0005117476,0.0020504112,0.002296393,0.000023957162,0.00057191856,0.00019939973,0.0010363138],"category_scores_gemma":[0.082762145,0.00013502031,0.000054603657,0.0054866895,0.0006140434,0.0003029829,0.00013599083,0.0015064044,0.00049067615],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00057718315,0.00004762235,0.03859135,0.0029573566,0.0000058011474,0.000018042665,0.029507602,0.000111479334,0.000052821266,0.0015904071,0.92217815,0.004362173],"study_design_scores_gemma":[0.0044418867,0.003194156,0.08681363,0.06159076,0.000009678999,0.0000022150714,0.60444474,0.018514978,0.000008979333,0.02518812,0.19535871,0.0004321615],"about_ca_topic_score_codex":0.006288881,"about_ca_topic_score_gemma":0.005261988,"teacher_disagreement_score":0.84200114,"about_ca_system_score_codex":0.0025038389,"about_ca_system_score_gemma":0.008269082,"threshold_uncertainty_score":0.99987686},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4385064816","doi":"10.1136/bmjebm-2023-112266","title":"Choosing wisely 10 years later: reflection and looking ahead","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto","funders":"","keywords":"Reflection (computer programming); Look-ahead; Computer science; Algorithm; Programming language","score_opus":0.7360109360885645,"score_gpt":0.6133176081544891,"score_spread":0.12269332793407539,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385064816","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7602816,0.0033280407,0.00037367243,0.22892879,0.0025274195,0.0021049522,0.0000063341963,0.0008131908,0.0016359991],"genre_scores_gemma":[0.9653117,0.0023228705,0.0006552454,0.02358415,0.0033007422,0.0003347517,0.00004485365,0.0000844715,0.004361239],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99399984,0.0027035067,0.0010764475,0.0005739322,0.0008453738,0.0008009217],"domain_scores_gemma":[0.9909285,0.007087379,0.00059172563,0.000557995,0.00040588577,0.00042852992],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["insufficient_payload"],"category_scores_codex":[0.012257414,0.00023645393,0.0005086888,0.0005003955,0.00075528293,0.000019817433,0.00019073328,0.00030848815,0.0010713788],"category_scores_gemma":[0.017589314,0.00020509609,0.000044740562,0.001146053,0.00023627821,0.0006120939,0.00011162862,0.001061051,0.0010695138],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0021008688,0.00010283989,0.44630787,0.008729258,0.000113760885,0.00055705925,0.019748032,0.000057488876,0.012772018,0.00092628226,0.2882029,0.22038163],"study_design_scores_gemma":[0.004291088,0.001278399,0.613379,0.039652843,0.00023672954,0.000028613907,0.008194261,0.0018824359,0.00030527764,0.0019475494,0.32808033,0.0007234916],"about_ca_topic_score_codex":0.0021915182,"about_ca_topic_score_gemma":0.00058090396,"teacher_disagreement_score":0.21965815,"about_ca_system_score_codex":0.00037347674,"about_ca_system_score_gemma":0.0010118262,"threshold_uncertainty_score":0.99984175},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W4385270131","doi":"10.1136/bmjebm-2023-112482","title":"Exploring advanced methods for network meta-analysis","year":2023,"lang":"en","type":"editorial","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Institute for Work & Health; St. Michael's Hospital","funders":"","keywords":"Network analysis; Computer science; Engineering","score_opus":0.9687966746492905,"score_gpt":0.6861621719315322,"score_spread":0.28263450271775836,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385270131","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":"methods","model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000013269049,0.033807382,0.4279063,0.0091105625,0.5253201,0.003504502,0.00010972817,0.00005603857,0.0001840363],"genre_scores_gemma":[0.000020843449,0.0019633332,0.22754534,0.00071372866,0.72596,0.01034685,0.0005281822,0.0001785023,0.03274321],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.87128913,0.054633774,0.03753914,0.006201898,0.028589094,0.0017469344],"domain_scores_gemma":[0.33784148,0.6009746,0.030055782,0.017704546,0.012375773,0.0010478206],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.5979968,0.0018120307,0.03919411,0.0033234323,0.0004924662,0.00086089835,0.0068907486,0.00083271216,0.016622847],"category_scores_gemma":[0.8151984,0.0008199602,0.029606305,0.016574599,0.00031120793,0.00091695215,0.00038050365,0.0011706464,0.0016531954],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000090912596,0.000016609134,0.000012943758,0.00047103874,0.09101371,0.000015359114,0.000089452995,0.010333916,0.0000049153655,0.00011812463,0.8815946,0.016238414],"study_design_scores_gemma":[0.00032949384,0.00020760225,0.000011939927,0.00046999127,0.34602225,3.231032e-7,0.0001494443,0.006011487,0.0000051525017,0.0026983698,0.64359665,0.00049729773],"about_ca_topic_score_codex":0.000116187955,"about_ca_topic_score_gemma":0.0001761446,"teacher_disagreement_score":0.5463408,"about_ca_system_score_codex":0.00023802936,"about_ca_system_score_gemma":0.0011136829,"threshold_uncertainty_score":0.9994625},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4385863795","doi":"10.1136/bmjebm-2023-112271","title":"Reducing low value care: opportunities and challenges for Choosing Wisely campaigns","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":11,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; St. Michael's Hospital; Ottawa Hospital; University of Ottawa","funders":"Canadian Institutes of Health Research","keywords":"Value (mathematics); Business; Computer science","score_opus":0.8963123753308139,"score_gpt":0.6013933185560272,"score_spread":0.2949190567747867,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4385863795","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19585533,0.052726083,0.0008212518,0.7390133,0.0034068092,0.006040175,0.000059926584,0.0008381318,0.001238958],"genre_scores_gemma":[0.8509769,0.09188624,0.0027542918,0.039388403,0.008488831,0.0036178487,0.00022465958,0.0002350457,0.002427781],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.99302953,0.0029363679,0.0013310165,0.0007452568,0.00086262764,0.0010952016],"domain_scores_gemma":[0.97622603,0.020594269,0.00076273986,0.0007671585,0.0009768971,0.0006729073],"candidate_categories":["metaresearch","metaepi_narrow","sts"],"consensus_categories":[],"category_scores_codex":[0.013951885,0.0003665371,0.0008108861,0.00049639295,0.0013144036,0.00001931373,0.0002953638,0.00035036827,0.00016311492],"category_scores_gemma":[0.029473685,0.00030571263,0.000079667596,0.00046160753,0.00040508193,0.0006275942,0.00012851118,0.0007992597,0.00006812516],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.002784233,0.00018964919,0.02000241,0.14645967,0.00028951274,0.0005525077,0.2404653,0.00024305182,0.0027048253,0.03118231,0.11811089,0.43701565],"study_design_scores_gemma":[0.0066509373,0.0027795334,0.035502132,0.11376096,0.0005455568,0.000025758405,0.523369,0.0029246393,0.0003612742,0.0026607239,0.3101038,0.0013156686],"about_ca_topic_score_codex":0.001649632,"about_ca_topic_score_gemma":0.00035042505,"teacher_disagreement_score":0.69962496,"about_ca_system_score_codex":0.00041343606,"about_ca_system_score_gemma":0.0029706392,"threshold_uncertainty_score":0.99998575},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4386767565","doi":"10.1136/bmjebm-2023-112582","title":"Enhanced access but increased testing with virtual care: insights from a multinational survey","year":2023,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Telemedicine and Telehealth Implementation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Canadian Institute for Health Information; University of Manitoba; Beef Farmers of Ontario","funders":"","keywords":"Multinational corporation; Business; Survey research; Computer science; Business administration","score_opus":0.18317465042704004,"score_gpt":0.41496418873076385,"score_spread":0.2317895383037238,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4386767565","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19377488,0.00275731,0.0092414,0.7806246,0.0028135194,0.008417823,0.0008711964,0.0011328483,0.00036637447],"genre_scores_gemma":[0.2398387,0.00020905912,0.0025120403,0.6920076,0.029161626,0.0013314986,0.03400444,0.00033626857,0.000598779],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99109524,0.00095144677,0.001901906,0.0015199165,0.0035629345,0.0009685563],"domain_scores_gemma":[0.9771902,0.017007163,0.0014439087,0.0011383414,0.0026717302,0.00054867275],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity"],"consensus_categories":[],"category_scores_codex":[0.0013605043,0.0010020941,0.0019267717,0.001229354,0.00028540465,0.000058280904,0.0006340941,0.0007835047,0.0006829079],"category_scores_gemma":[0.019413609,0.00071791327,0.00013135995,0.0021304155,0.00051128224,0.00043664387,0.00013468094,0.0024568902,0.00012393406],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0023519956,0.00008895161,0.046837874,0.002456955,0.00036886535,0.00262604,0.00074280304,0.000028336777,0.0018490943,0.0000040528053,0.9227146,0.019930467],"study_design_scores_gemma":[0.02968686,0.014146668,0.8284449,0.06696472,0.0026983775,0.00012378213,0.0014012626,0.0013063833,0.004132609,0.00007968814,0.04904709,0.001967638],"about_ca_topic_score_codex":0.044414286,"about_ca_topic_score_gemma":0.0024191565,"teacher_disagreement_score":0.8736675,"about_ca_system_score_codex":0.0007588836,"about_ca_system_score_gemma":0.004724757,"threshold_uncertainty_score":0.9998445},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4387504423","doi":"10.1136/bmjebm-2023-112352","title":"Making sustainable healthcare decisions: three turns towards sustainable guidelines","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Agriculture Sustainability and Environmental Impact","field":"Environmental Science","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"FP7 People: Marie-Curie Actions; European Commission","keywords":"Health care; Business; Environmental planning; Environmental resource management; Geography; Environmental science; Economics; Economic growth","score_opus":0.19240987544358096,"score_gpt":0.416253864773184,"score_spread":0.22384398932960303,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387504423","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.4333556,0.007486082,0.010108569,0.53161037,0.00076544547,0.00753132,0.000016633025,0.0013533097,0.007772672],"genre_scores_gemma":[0.9763482,0.0007267408,0.0016133485,0.0067967097,0.000653212,0.0003703667,0.00004906262,0.000058011585,0.013384355],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9942323,0.0002603815,0.0010252905,0.0009453092,0.0017424444,0.0017942558],"domain_scores_gemma":[0.9970066,0.0009036589,0.00029719673,0.0010236267,0.00020782877,0.0005610914],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0048374557,0.00051765266,0.00061582256,0.00024831446,0.00067222014,0.00007729569,0.0008124545,0.00024696763,0.003603215],"category_scores_gemma":[0.014264889,0.00035337903,0.00019118862,0.0026321856,0.00077528885,0.0011737866,0.0008069523,0.00043152488,0.0007265108],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00083825924,0.0002745173,0.18494035,0.0016738864,0.000065353066,0.004541326,0.0031602292,0.039550986,0.0007394097,0.0039712545,0.7037333,0.05651109],"study_design_scores_gemma":[0.0023220303,0.001989452,0.55006564,0.0034169485,0.0001808403,0.000115973686,0.06669069,0.004396628,0.00031476337,0.04743509,0.32165387,0.0014180634],"about_ca_topic_score_codex":0.008598693,"about_ca_topic_score_gemma":0.00057084387,"teacher_disagreement_score":0.5429926,"about_ca_system_score_codex":0.0021949026,"about_ca_system_score_gemma":0.00029143115,"threshold_uncertainty_score":0.9998918},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4387558756","doi":"10.1136/bmjebm-2022-112125","title":"Can a replication revolution resolve the duplication crisis in systematic reviews?","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Université de Montréal; Cochrane; Ottawa Hospital; Bruyère; University of Ottawa","funders":"","keywords":"Replication (statistics); Gene duplication; Political science; Biology; Genetics; Virology","score_opus":0.831327983766012,"score_gpt":0.5804957268733995,"score_spread":0.25083225689261257,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4387558756","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.022501973,0.05811914,0.057774372,0.8246053,0.0011117277,0.034075852,0.000013194798,0.00012405812,0.0016743938],"genre_scores_gemma":[0.9671437,0.00426149,0.00158726,0.008578913,0.0007737079,0.008707779,0.00006838243,0.000039929368,0.008838797],"study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9357973,0.033625863,0.019722914,0.0020584406,0.008291328,0.0005041835],"domain_scores_gemma":[0.9318575,0.035994552,0.01290013,0.016764645,0.0022140134,0.00026910743],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.42901596,0.00036260998,0.004083494,0.0010616152,0.00022743661,0.00021850812,0.002276991,0.00011718263,0.001267636],"category_scores_gemma":[0.50793314,0.0001400901,0.0008935707,0.009173257,0.00013925903,0.00031490054,0.00009406955,0.00025312733,0.0047304095],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000024960924,0.00003119952,0.008352663,0.007636478,0.000048935548,0.000009724682,0.0014030903,0.00032029455,0.00043117735,0.0028389655,0.9729241,0.0059784125],"study_design_scores_gemma":[0.0023221842,0.0007468469,0.22656752,0.18738359,0.003598855,0.00012086217,0.017996406,0.28013948,0.0002445806,0.05225843,0.22696061,0.0016606282],"about_ca_topic_score_codex":0.00029815658,"about_ca_topic_score_gemma":0.00030519505,"teacher_disagreement_score":0.94464177,"about_ca_system_score_codex":0.00024369537,"about_ca_system_score_gemma":0.00018729373,"threshold_uncertainty_score":0.99964535},"labels":[],"label_agreement":null},{"id":"W4388006999","doi":"10.1136/bmjebm-2023-112475","title":"How to use the regulatory data from Health Canada for secondary analyses on new drugs, biologics and vaccines","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Dalhousie University","funders":"Dalhousie University","keywords":"Medicine; Public health; Internet privacy; Business; Computer science","score_opus":0.7468377814715961,"score_gpt":0.5827913881962639,"score_spread":0.16404639327533221,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4388006999","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.19657588,0.0017471187,0.0022292768,0.7961205,0.0006217373,0.0022376925,0.00038392338,0.00007954325,0.0000043013974],"genre_scores_gemma":[0.6790181,0.0019764483,0.008004658,0.2989389,0.0049105557,0.0000760517,0.005305809,0.00007395882,0.0016955558],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99700123,0.0002792371,0.000960347,0.00068544934,0.000708918,0.00036484355],"domain_scores_gemma":[0.97519714,0.022003612,0.00047027573,0.0015424707,0.0002576788,0.00052881683],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.003728153,0.00024143711,0.0006208716,0.00016503193,0.00017821595,0.000044454795,0.00034855536,0.00006892782,0.0001138981],"category_scores_gemma":[0.0633927,0.00014054097,0.00004574471,0.0006062299,0.00008416343,0.00031136005,0.00015626024,0.00026516587,0.000006321041],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0009862694,0.000016115957,0.0056848936,0.00011218702,0.000087238135,0.00001914026,0.0001235041,0.00003725702,0.0006102802,0.000018241586,0.9261884,0.06611644],"study_design_scores_gemma":[0.007819463,0.0034867413,0.42812535,0.0038032427,0.0008906423,0.000014813733,0.007780022,0.0058542844,0.0005178239,0.00044791025,0.54084516,0.00041453028],"about_ca_topic_score_codex":0.43911657,"about_ca_topic_score_gemma":0.34300613,"teacher_disagreement_score":0.49718162,"about_ca_system_score_codex":0.00019979775,"about_ca_system_score_gemma":0.004597661,"threshold_uncertainty_score":0.94449675},"labels":[{"model":"gemma","categories":["metaresearch","open_science"],"domain":"reproducibility","study_design":"not_applicable","genre":"methods","about_ca_system":true,"about_ca_topic":true,"confidence":"high"},{"model":"gpt","categories":["metaresearch","open_science"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":true,"confidence":"high"}],"label_agreement":"split"},{"id":"W4389207515","doi":"10.1136/bmjebm-2023-112397","title":"Integrating a climate lens into the design of education programmes for health professionals","year":2023,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta; McMaster University; University of Waterloo","funders":"","keywords":"Lens (geology); Health professionals; Through-the-lens metering; Medical education; Sociology; Engineering ethics; Public relations; Engineering; Political science; Medicine; Health care","score_opus":0.16808932245712369,"score_gpt":0.5137538397847501,"score_spread":0.3456645173276264,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4389207515","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.03770132,0.0025663876,0.074487,0.8721562,0.0021400037,0.010589146,0.0000037422355,0.0002488546,0.00010733699],"genre_scores_gemma":[0.68283325,0.0012999661,0.20940891,0.08883229,0.003964274,0.011403427,0.00077010144,0.00011862071,0.0013691677],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9971264,0.00029850437,0.0010628444,0.00031446898,0.0007916531,0.0004061462],"domain_scores_gemma":[0.99549353,0.0021146638,0.00062298466,0.00053070125,0.0011418384,0.00009630343],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.00936949,0.00018548855,0.00047304615,0.00044807204,0.0003003966,0.000008979059,0.00020472438,0.00009985096,0.000070877824],"category_scores_gemma":[0.026203018,0.00010511079,0.00007534206,0.0019363326,0.0003566797,0.00014158692,0.00003060808,0.0002980545,0.000015614165],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00068244745,0.00050681684,0.0032790538,0.004503942,0.000052118266,0.0000017078607,0.010437472,0.000052906136,0.0024456347,0.0065764314,0.4167125,0.554749],"study_design_scores_gemma":[0.01630851,0.025865683,0.06508093,0.31408986,0.001553647,0.00016013107,0.2017205,0.13474676,0.0151840085,0.016708901,0.20688832,0.0016927393],"about_ca_topic_score_codex":0.00014788841,"about_ca_topic_score_gemma":0.000010217401,"teacher_disagreement_score":0.78332394,"about_ca_system_score_codex":0.00027393296,"about_ca_system_score_gemma":0.004504192,"threshold_uncertainty_score":0.9819997},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4390977446","doi":"10.1136/bmjebm-2023-112608","title":"Preventing overdiagnosis and overuse: proposed guidance for guideline panels","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":5,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Overdiagnosis; Guideline; Medicine; Intensive care medicine; Medical physics; Internal medicine; Pathology","score_opus":0.7344073371348866,"score_gpt":0.6267851970459801,"score_spread":0.10762214008890647,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4390977446","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.09204859,0.06692986,0.038128298,0.78020036,0.0057661785,0.01488552,0.00012999595,0.000997997,0.0009132184],"genre_scores_gemma":[0.82776475,0.0064462763,0.03017133,0.10951555,0.013609934,0.0048439824,0.00014988027,0.00027499802,0.0072232834],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99255794,0.0022895893,0.002349688,0.0009264445,0.00093058,0.00094573427],"domain_scores_gemma":[0.973584,0.023606889,0.0007209615,0.0007513557,0.0008587937,0.00047802],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.021133885,0.00038328033,0.0007941376,0.0002831175,0.0007116876,0.000039634804,0.0003107142,0.00033667567,0.0010161351],"category_scores_gemma":[0.05570081,0.0002914203,0.000115170966,0.0005756429,0.00027112456,0.0008465193,0.00014629727,0.00093453325,0.00011772952],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0011437566,0.00017031089,0.11666357,0.046794154,0.00021242477,0.00017495656,0.0054826927,0.000020436051,0.0063756257,0.006945192,0.7021727,0.113844186],"study_design_scores_gemma":[0.003456877,0.0013289734,0.020706328,0.09782654,0.000501877,0.000017985463,0.0026593546,0.009100388,0.0008491032,0.0035409525,0.85931474,0.0006969003],"about_ca_topic_score_codex":0.0018234766,"about_ca_topic_score_gemma":0.00024056656,"teacher_disagreement_score":0.73571616,"about_ca_system_score_codex":0.00041759168,"about_ca_system_score_gemma":0.002612876,"threshold_uncertainty_score":0.9999538},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4391022170","doi":"10.1136/bmjebm-2023-112784","title":"Development of the Reporting Infographics and Visual Abstracts of Comparative studies (RIVA-C) checklist and guide","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of British Columbia","funders":"National Health and Medical Research Council; Medical Research Council","keywords":"Infographic; Checklist; Psychological intervention; Delphi method; Reading (process); Medical education; Psychology; Computer science; Medicine","score_opus":0.48597114545384495,"score_gpt":0.5992018600550433,"score_spread":0.11323071460119832,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391022170","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.9686548,0.012559062,0.00021228007,0.0147591755,0.00019892326,0.0008729342,0.0000011967107,0.000053244483,0.0026883718],"genre_scores_gemma":[0.9952177,0.0010007798,0.0033210055,0.00008869166,0.000101973215,0.000044316635,5.567072e-7,0.00000576016,0.00021923694],"study_design_codex":"qualitative","study_design_gemma":"observational","domain_scores_codex":[0.9970034,0.00018769188,0.0013354708,0.00021763693,0.0010580358,0.00019776779],"domain_scores_gemma":[0.9960151,0.0025271676,0.0007130563,0.00014991438,0.0005049466,0.00008984773],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.011233697,0.00010957612,0.0003760089,0.00018959293,0.00022554016,0.000020857575,0.00017463857,0.00006491862,0.0000128352585],"category_scores_gemma":[0.015452032,0.00006786363,0.00003599185,0.00075055065,0.0026083782,0.00014715464,0.00012713856,0.00020166162,3.817772e-7],"study_design_candidate":"qualitative","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0003950234,0.00028015452,0.20738438,0.011530855,0.0010136301,0.00031380611,0.43118557,0.00005274229,0.07347546,0.023970818,0.06159219,0.18880537],"study_design_scores_gemma":[0.0012186769,0.00096686126,0.40598187,0.086126335,0.00037260252,0.000038818296,0.25124168,0.00095806347,0.13202335,0.005595656,0.11465654,0.00081955147],"about_ca_topic_score_codex":0.000782552,"about_ca_topic_score_gemma":0.0012086096,"teacher_disagreement_score":0.1985975,"about_ca_system_score_codex":0.00007457762,"about_ca_system_score_gemma":0.0010970278,"threshold_uncertainty_score":0.99284124},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4391035776","doi":"10.1136/bmjebm-2023-112530","title":"Rapid review methods series: Guidance on the use of supportive software","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa","funders":"","keywords":"Systematic review; Software; Computer science; Automation; Process (computing); Software engineering; Engineering; MEDLINE","score_opus":0.9106387882086024,"score_gpt":0.6251835614097659,"score_spread":0.2854552267988365,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4391035776","genre_codex":"review","genre_gemma":"methods","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00030940343,0.42596024,0.35617474,0.21109702,0.0013068953,0.004203352,0.000037754293,0.000038656584,0.00087194797],"genre_scores_gemma":[0.01677877,0.1301414,0.5938496,0.15262966,0.0027659647,0.0028758557,0.00006274723,0.00020591918,0.10069005],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.95049214,0.028363148,0.012040583,0.0012794974,0.0074869995,0.0003376207],"domain_scores_gemma":[0.83425343,0.1470923,0.006474257,0.008666946,0.0031977398,0.0003153326],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.28069344,0.00043680528,0.0044301045,0.00040703767,0.00011074073,0.00024593828,0.0019256246,0.00008310277,0.04736626],"category_scores_gemma":[0.5873651,0.00014619462,0.0016771986,0.0037196446,0.00045240938,0.0005897171,0.000101808975,0.00031207275,0.0012931242],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000024280022,0.00001520931,0.00032366507,0.0024461225,0.00015024717,0.000032337717,0.00019299591,0.000031950316,0.00009784828,0.0018769944,0.81395054,0.1808578],"study_design_scores_gemma":[0.00008170334,0.0003851155,0.00055541954,0.057203114,0.00083630363,0.000023094348,0.00019067278,0.0016392445,0.00037716623,0.0018194884,0.9367147,0.00017399743],"about_ca_topic_score_codex":0.00002531484,"about_ca_topic_score_gemma":0.000006007078,"teacher_disagreement_score":0.30667165,"about_ca_system_score_codex":0.00005738764,"about_ca_system_score_gemma":0.0004900721,"threshold_uncertainty_score":0.9994845},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4392545360","doi":"10.1136/bmjebm-2024-112868","title":"Overly complex methods may impair pragmatic use of core evidence-based medicine principles","year":2024,"lang":"en","type":"editorial","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":13,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Core (optical fiber); Psychology; Epistemology; Medicine; Computer science; Philosophy","score_opus":0.6553516264834451,"score_gpt":0.6381847810439233,"score_spread":0.01716684543952185,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392545360","genre_codex":"editorial","genre_gemma":"editorial","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"editorial","genre_consensus":"editorial","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0009452309,0.022302087,0.008877452,0.13704506,0.81754297,0.011810145,0.00038946365,0.0004918694,0.0005957061],"genre_scores_gemma":[0.0021183486,0.008377424,0.03521528,0.007724916,0.9291053,0.003666135,0.0018801008,0.00030743255,0.011605091],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.97196394,0.009476703,0.0059316223,0.0020421236,0.007974083,0.0026114976],"domain_scores_gemma":[0.8162705,0.17044613,0.0034746998,0.0027991773,0.0048721507,0.0021373753],"candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.052145045,0.0012728476,0.0035675021,0.0024139367,0.0011146554,0.00006311254,0.0021524862,0.0019903763,0.008138145],"category_scores_gemma":[0.34085634,0.0008408026,0.00039431127,0.0037798933,0.00313547,0.0011182361,0.00038870153,0.0056184866,0.00048739376],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0010840062,0.00012106228,0.0017621205,0.039922807,0.000089498586,0.000064762164,0.0022214395,0.000095199975,0.00079432933,0.00024079326,0.94987595,0.0037280424],"study_design_scores_gemma":[0.0017753562,0.0035519777,0.002653259,0.20643505,0.00049701746,0.000002668911,0.0014613969,0.00884538,0.000027232772,0.00042369388,0.7737373,0.0005896832],"about_ca_topic_score_codex":0.01674113,"about_ca_topic_score_gemma":0.00096949696,"teacher_disagreement_score":0.28871128,"about_ca_system_score_codex":0.0025784157,"about_ca_system_score_gemma":0.046977304,"threshold_uncertainty_score":0.9995774},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4392583344","doi":"10.1136/bmjebm-2023-112695","title":"CARE-radiology statement explanation and elaboration: reporting guideline for radiological case reports","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":16,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Fundamental Research Funds for the Central Universities; Lanzhou University","keywords":"Guideline; Radiological weapon; Checklist; Medicine; Delphi method; Multidisciplinary approach; Documentation; Medical physics; Radiology; Psychology; Pathology; Computer science","score_opus":0.3723114543237489,"score_gpt":0.5789761313316922,"score_spread":0.20666467700794333,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392583344","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.12732464,0.047599826,0.36199254,0.44574305,0.0034277095,0.009458978,0.000030221485,0.0012806004,0.0031424174],"genre_scores_gemma":[0.97499406,0.0008380647,0.01999598,0.0009797234,0.0020146223,0.0008513239,0.00008601341,0.000018518294,0.00022167034],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99572045,0.00048505765,0.0021426785,0.00057831925,0.000690182,0.00038333383],"domain_scores_gemma":[0.9952175,0.002672487,0.00072635186,0.0003085143,0.00086905865,0.0002060728],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.024242196,0.00014684442,0.00034352837,0.0002575278,0.0003861485,0.000074462725,0.000093821334,0.0001723943,0.00019745756],"category_scores_gemma":[0.051691648,0.00011125453,0.000052383606,0.0005087298,0.0005556772,0.00029666946,0.00003325521,0.00020252312,0.0000018901047],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00029540315,0.00004612197,0.00561794,0.0011921029,0.00008515822,0.10286946,0.015053725,0.00032057188,0.0031914269,0.021658164,0.7202884,0.12938151],"study_design_scores_gemma":[0.0009132005,0.0025115123,0.000320587,0.0041877576,0.00022258608,0.011908587,0.04418162,0.009317409,0.0012051453,0.01260216,0.911971,0.0006584729],"about_ca_topic_score_codex":0.0015661431,"about_ca_topic_score_gemma":0.0006431842,"teacher_disagreement_score":0.8476694,"about_ca_system_score_codex":0.0004129661,"about_ca_system_score_gemma":0.0011471092,"threshold_uncertainty_score":0.9562964},"labels":[],"label_agreement":null},{"id":"W4392797608","doi":"10.1136/bmjebm-2023-112722","title":"Rapid reviews methods series: assessing the appropriateness of conducting a rapid review","year":2024,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Ottawa Public Health; Public Health Agency of Canada; University of Ottawa","funders":"","keywords":"Scope (computer science); Context (archaeology); Risk analysis (engineering); Resource (disambiguation); Process (computing); Management science; Computer science; Systematic review; Operations research; Medicine; MEDLINE; Political science; Engineering","score_opus":0.9588174281060408,"score_gpt":0.7032138568070663,"score_spread":0.2556035712989745,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4392797608","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[9.843276e-8,0.94687694,0.027116157,0.009356133,0.0026855087,0.012718708,0.000022101292,0.000027150636,0.0011971745],"genre_scores_gemma":[2.0789179e-7,0.9540018,0.036902085,0.0020855824,0.0015152993,0.0021085392,0.000053311283,0.0001102809,0.0032229275],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.7213483,0.19639416,0.06369156,0.004234786,0.013328492,0.0010027254],"domain_scores_gemma":[0.7686756,0.12030807,0.07870916,0.025617918,0.005898513,0.0007906923],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"category_scores_codex":[0.6782458,0.0020370865,0.044699073,0.0013655917,0.0003397198,0.00084817293,0.0075199683,0.0005075359,0.022354279],"category_scores_gemma":[0.4678687,0.00064291194,0.012090751,0.011827136,0.00093807175,0.0008872623,0.00057433656,0.0016647818,0.0021209614],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0000016920457,0.000012897505,6.55416e-7,0.18709004,0.00034586992,0.000020525262,0.00007450215,3.7007553e-7,0.0000026212178,0.00012387919,0.17977276,0.6325542],"study_design_scores_gemma":[0.00006310485,0.00007756165,2.3212279e-7,0.41492042,0.013438697,0.00018231597,0.00018324869,0.000042369826,0.0000016059196,0.00022192218,0.5705234,0.00034516997],"about_ca_topic_score_codex":0.000017323679,"about_ca_topic_score_gemma":0.000002555329,"teacher_disagreement_score":0.632209,"about_ca_system_score_codex":0.00020141593,"about_ca_system_score_gemma":0.0021859163,"threshold_uncertainty_score":0.9996022},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4394722187","doi":"10.1136/bmjebm-2023-112767","title":"Trivalent and quadrivalent seasonal influenza vaccine in adults aged 60 and older: a systematic review and network meta-analysis","year":2024,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Influenza Virus Research Studies","field":"Medicine","cited_by":15,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Public Health Ontario; University of Ottawa; McMaster University; University of Toronto; Public Health Agency of Canada; Ottawa Hospital; Institute for Work & Health; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Canada Excellence Research Chairs, Government of Canada; Public Health Agency; Health Research Board","keywords":"Medicine; Meta-analysis; Confidence interval; Odds ratio; Influenza vaccine; Placebo; Randomized controlled trial; Systematic review; Internal medicine; MEDLINE; Pediatrics; Vaccination; Immunology; Alternative medicine","score_opus":0.3326713281150513,"score_gpt":0.4941296525993264,"score_spread":0.1614583244842751,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4394722187","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000030456014,0.9755553,0.00002249302,0.0067196344,0.00008691592,0.017393876,0.00007308129,0.000094677715,0.000023558167],"genre_scores_gemma":[0.00006442527,0.98805726,0.00024542873,0.005582029,0.00044159315,0.00503791,0.00007272122,0.00011616163,0.0003824695],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.98823005,0.00252578,0.004221283,0.0018789549,0.0021068628,0.0010370432],"domain_scores_gemma":[0.99021643,0.0055974266,0.0012845567,0.0015533478,0.0004629144,0.00088534027],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaepi_narrow"],"category_scores_codex":[0.011374508,0.0016053086,0.0219813,0.001738857,0.00017240547,0.00007709358,0.00036520124,0.00042965633,0.00033164138],"category_scores_gemma":[0.01209473,0.0008780344,0.0017890031,0.0038278135,0.00049510755,0.00020027926,0.0004929127,0.001665483,0.000030418612],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00018830005,0.000059750484,0.00022921414,0.9195483,0.0679452,0.0009950805,0.00017326942,0.0000021533951,7.675391e-8,0.000023354049,0.0055708536,0.005264497],"study_design_scores_gemma":[0.0013369558,0.0003934987,0.00023367838,0.51790106,0.4669777,0.00018280295,0.000033799213,0.00006808707,3.8727137e-8,0.000010293439,0.012475571,0.0003865225],"about_ca_topic_score_codex":0.00022916452,"about_ca_topic_score_gemma":0.00017900465,"teacher_disagreement_score":0.40164718,"about_ca_system_score_codex":0.00041038386,"about_ca_system_score_gemma":0.0007859378,"threshold_uncertainty_score":0.99966943},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":true,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4396967666","doi":"10.1136/bmjebm-2023-112622","title":"Direct-to-consumer advertising: a modifiable driver of overdiagnosis and overtreatment","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"York University","funders":"","keywords":"Overdiagnosis; Advertising; Medicine; Business; Internal medicine","score_opus":0.48704821847550556,"score_gpt":0.5796537362761751,"score_spread":0.09260551780066956,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4396967666","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.7180406,0.1303296,0.0006182318,0.1339191,0.0039288597,0.003644816,0.00022434261,0.00042839468,0.008866089],"genre_scores_gemma":[0.98260933,0.004348237,0.00023271503,0.01069677,0.0005349828,0.00025859312,0.000012247082,0.000029459468,0.0012776764],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9974312,0.00049367483,0.0006033943,0.0005561744,0.00037675843,0.00053876964],"domain_scores_gemma":[0.9950296,0.003602445,0.00009006249,0.00028414067,0.00014289274,0.000850804],"candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0014880309,0.00031035196,0.00056327356,0.0002652011,0.00016447836,0.000015821108,0.00017652987,0.00038874347,0.0031887083],"category_scores_gemma":[0.0009923119,0.00024071393,0.00009517407,0.000564722,0.00046991187,0.00021965125,0.000078973026,0.0007892591,0.00011713276],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0049025775,0.0011795395,0.14855309,0.00722427,0.0016875218,0.0019925016,0.0042237523,0.0034598394,0.10648931,0.0073988596,0.4916901,0.22119863],"study_design_scores_gemma":[0.0037210875,0.0018298706,0.008083837,0.010848491,0.001868754,0.000071698385,0.00026406895,0.020856274,0.17023492,0.000567966,0.7809726,0.00068040285],"about_ca_topic_score_codex":0.00039719488,"about_ca_topic_score_gemma":0.000013103736,"teacher_disagreement_score":0.2892825,"about_ca_system_score_codex":0.00018667581,"about_ca_system_score_gemma":0.00038908742,"threshold_uncertainty_score":0.9977225},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4398251068","doi":"10.1136/bmjebm-2023-112681","title":"Guidelines for the development and validation of patient-reported outcome measures: a scoping review","year":2024,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Cancer survivorship and care","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Prom; PsycINFO; CLARITY; Guideline; MEDLINE; Scopus; Medicine; Data extraction; Computer science; Medical physics; Pathology","score_opus":0.5391998939064311,"score_gpt":0.5249969430455457,"score_spread":0.01420295086088541,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4398251068","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0000045122197,0.9769788,0.0015133078,0.009339353,0.0011776708,0.01088226,0.00001192848,0.000061180996,0.0000309931],"genre_scores_gemma":[0.000014119139,0.9924526,0.00087172433,0.0025839887,0.0007209881,0.0029105784,0.00015213876,0.00007237339,0.0002214416],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9940474,0.00019382006,0.0036732864,0.0006180631,0.0011860942,0.00028134914],"domain_scores_gemma":[0.9939337,0.0021698852,0.0013767942,0.0008243536,0.0015135432,0.00018174778],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.004998679,0.0005595496,0.003208339,0.0003051165,0.000088505905,0.00001635057,0.0002005748,0.00020028165,0.000068454836],"category_scores_gemma":[0.015862701,0.0002769219,0.0005170788,0.0007335463,0.000176206,0.00006522065,0.00006941367,0.0003214736,0.000006076746],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00003300675,0.0000090732265,0.000025436313,0.42793885,0.0002312245,0.0000134971415,0.00012255588,2.4805647e-7,4.199811e-7,0.000006732953,0.0053414614,0.5662775],"study_design_scores_gemma":[0.00025142543,0.00016899327,0.0000012964018,0.5639592,0.008385647,0.000055069962,0.000073442396,0.0000033456718,0.000014648328,0.0000036404876,0.42693037,0.00015294818],"about_ca_topic_score_codex":0.000117706295,"about_ca_topic_score_gemma":0.00012444558,"teacher_disagreement_score":0.56612456,"about_ca_system_score_codex":0.00020929913,"about_ca_system_score_gemma":0.0036211929,"threshold_uncertainty_score":0.9999683},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4400491807","doi":"10.1136/bmjebm-2023-112719","title":"Cancer screening attendance rates in transgender and gender-diverse patients: a systematic review and meta-analysis","year":2024,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"LGBTQ Health, Identity, and Policy","field":"Psychology","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":false,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"","funders":"Academy of Medical Sciences; Wellcome Trust","keywords":"Transgender; Meta-analysis; Attendance; Cancer; Medicine; Clinical psychology; Psychology; Demography; Gerontology; Sociology; Internal medicine; Political science; Psychoanalysis","score_opus":0.5172235772782489,"score_gpt":0.5521889312201794,"score_spread":0.034965353941930566,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400491807","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00000343547,0.98695827,0.00013088043,0.0058196057,0.00027572818,0.0063715326,0.00024573985,0.0000512759,0.00014355741],"genre_scores_gemma":[0.0000451095,0.9897208,0.00006355862,0.0045230994,0.00026018277,0.003154617,0.00008478389,0.000074627365,0.002073206],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9923697,0.002420481,0.0025913948,0.001275483,0.00071609154,0.0006268086],"domain_scores_gemma":[0.99573964,0.0016004835,0.0010995833,0.0009644709,0.00021995432,0.00037589166],"candidate_categories":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0037573208,0.0009157216,0.010020858,0.001374338,0.00011873438,0.00005317876,0.00041355216,0.00037033897,0.002286016],"category_scores_gemma":[0.00073132035,0.00055524736,0.0012359908,0.0023866314,0.0003155069,0.00020321137,0.000069056376,0.00080327375,0.000083434265],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000009819199,0.000035690682,0.00050012267,0.94641435,0.03907523,0.00008467903,0.0003485784,5.046936e-7,4.5143125e-9,0.00007045493,0.008245942,0.005214614],"study_design_scores_gemma":[0.00038444152,0.00009080246,0.00021714656,0.19638313,0.7934641,0.000021029773,0.00005904522,0.000005238709,3.0686127e-9,0.000026722055,0.008914986,0.0004333218],"about_ca_topic_score_codex":0.0068891975,"about_ca_topic_score_gemma":0.0013623108,"teacher_disagreement_score":0.7543889,"about_ca_system_score_codex":0.0001907592,"about_ca_system_score_gemma":0.00041253067,"threshold_uncertainty_score":0.99972403},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4400582003","doi":"10.1136/bmjebm-2024-112886","title":"Pharmacological interventions for preventing upper gastrointestinal bleeding in people admitted to intensive care units: a network meta-analysis","year":2024,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Nosocomial Infections in ICU","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Psychological intervention; Intensive care; Medicine; Upper gastrointestinal bleeding; Intensive care medicine; Gastrointestinal bleeding; Meta-analysis; Intensive care unit; Internal medicine; Nursing","score_opus":0.4932109188940237,"score_gpt":0.5329212801273934,"score_spread":0.039710361233369684,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400582003","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00005662241,0.97100925,0.010969893,0.005469476,0.0009416177,0.011137398,0.000113683236,0.00023922419,0.00006285354],"genre_scores_gemma":[0.00089063786,0.9555333,0.014649049,0.0040567173,0.0048512244,0.017175745,0.0011877168,0.00026731341,0.0013882949],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.99280775,0.0008648513,0.0031171339,0.0014363748,0.0007750976,0.0009987842],"domain_scores_gemma":[0.9890119,0.0055074682,0.0009391905,0.00079315575,0.0031508505,0.0005974571],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0035942472,0.0010807231,0.008574062,0.0030617975,0.0001894171,0.00006789981,0.0004981289,0.00034607798,0.0016631499],"category_scores_gemma":[0.024962965,0.0007182218,0.006066192,0.010611233,0.00020243142,0.00014315612,0.00031494643,0.0017582757,0.000084284395],"study_design_candidate":"meta_analysis","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0012942804,0.00053308334,0.003678261,0.3319251,0.298799,0.0019999489,0.002090504,0.0029692866,0.000044955723,0.00036654802,0.23895073,0.117348306],"study_design_scores_gemma":[0.00076349505,0.0017404638,0.000036543523,0.21220528,0.67171645,0.0003591409,0.0011541577,0.00017662099,0.0000037312006,0.000021859392,0.11134353,0.00047873412],"about_ca_topic_score_codex":0.00022950115,"about_ca_topic_score_gemma":0.00048596488,"teacher_disagreement_score":0.37291744,"about_ca_system_score_codex":0.0013468657,"about_ca_system_score_gemma":0.0015995945,"threshold_uncertainty_score":0.99952686},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4400879434","doi":"10.1136/bmjebm-2024-112899","title":"Rapid review method series: interim guidance for the reporting of rapid reviews","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":55,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital; University of Alberta; Public Health Agency of Canada; Bruyère; University of Ottawa","funders":"Canadian Institutes of Health Research; Canada Research Chairs","keywords":"Interim; Systematic review; Set (abstract data type); Computer science; Psychology; MEDLINE; Political science","score_opus":0.9004752747875066,"score_gpt":0.647176774034603,"score_spread":0.2532985007529035,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4400879434","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000012307698,0.5960172,0.3266341,0.07120948,0.0009676515,0.00426934,0.000006935545,0.000013053375,0.00086995616],"genre_scores_gemma":[0.006081666,0.60342824,0.30366465,0.03809823,0.004126733,0.006502778,0.000036015303,0.00013556783,0.037926078],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9181783,0.018894266,0.05374631,0.0019555688,0.006711117,0.00051443034],"domain_scores_gemma":[0.73149306,0.15770067,0.08257787,0.019966453,0.007655666,0.00060627784],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.62129295,0.00051269284,0.008255812,0.00037242027,0.00016265569,0.00022888329,0.0025236981,0.00008886765,0.018057235],"category_scores_gemma":[0.7240558,0.0001770049,0.0035242166,0.0033758346,0.00028854128,0.0004984011,0.00013212246,0.0002813146,0.0004013348],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000019399626,0.0000080960135,0.00006162098,0.010725636,0.00012511991,0.000010764346,0.00016641921,0.000007610145,0.00038218978,0.0005594808,0.5713201,0.41661355],"study_design_scores_gemma":[0.00011664315,0.0003165945,0.00008848333,0.09394831,0.0014920922,0.0000775343,0.0002625198,0.0052050375,0.00027826728,0.0010056585,0.8970376,0.0001712299],"about_ca_topic_score_codex":0.000024150895,"about_ca_topic_score_gemma":0.000009566395,"teacher_disagreement_score":0.4164423,"about_ca_system_score_codex":0.00006049069,"about_ca_system_score_gemma":0.00046176952,"threshold_uncertainty_score":0.9828404},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch","research_integrity"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4401198502","doi":"10.1136/bmjebm-2023-112815","title":"Successes, shortcomings and learning opportunities for evidence-based medicine from the COVID-19 pandemic","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Coronavirus disease 2019 (COVID-19); Pandemic; 2019-20 coronavirus outbreak; Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Virology; Computer science; Medicine; Infectious disease (medical specialty); Internal medicine; Disease","score_opus":0.7098921007921705,"score_gpt":0.591445131736033,"score_spread":0.1184469690561375,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401198502","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.07799783,0.059353817,0.022481214,0.83280283,0.0027257758,0.0040193144,0.00003684711,0.00036056494,0.0002217925],"genre_scores_gemma":[0.84439266,0.018299714,0.0007773698,0.12053656,0.00867505,0.0029062056,0.00022267867,0.00006668145,0.004123064],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99237037,0.0027885283,0.0013568109,0.0008552416,0.0014867971,0.0011422331],"domain_scores_gemma":[0.89020365,0.1069101,0.00035757915,0.0005736959,0.0005513247,0.0014036193],"candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.027385317,0.00036355277,0.000673496,0.00045641462,0.002606625,0.000067269655,0.0006802962,0.00026161945,0.002853076],"category_scores_gemma":[0.09984336,0.00020436078,0.00008028949,0.0009567268,0.0016557266,0.0007677188,0.00009683508,0.0015917646,0.00004385555],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00095693197,0.000025939902,0.18705532,0.0071009574,0.00004453335,0.000067920046,0.012677944,0.00010858252,0.000774625,0.0010380098,0.756614,0.0335352],"study_design_scores_gemma":[0.0019196933,0.0016537451,0.012550814,0.045768633,0.00022416616,0.000013085021,0.038223207,0.018739631,0.000020445996,0.0018415025,0.8786595,0.0003855812],"about_ca_topic_score_codex":0.024830602,"about_ca_topic_score_gemma":0.0018720236,"teacher_disagreement_score":0.76639485,"about_ca_system_score_codex":0.0007417671,"about_ca_system_score_gemma":0.013919005,"threshold_uncertainty_score":0.99869186},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4401991382","doi":"10.1136/bmjebm-2024-113031","title":"Meta-analysis of continuous outcomes: a user’s guide for analysis and interpretation","year":2024,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Interpretation (philosophy); Computer science; Meta-analysis; Psychology; Medicine; Internal medicine; Programming language","score_opus":0.8238197139119845,"score_gpt":0.6024751740025874,"score_spread":0.22134453990939706,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4401991382","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.014764114,0.038113907,0.9222157,0.021873256,0.00026772497,0.0021816732,0.00011464152,0.000022054875,0.00044694895],"genre_scores_gemma":[0.96630293,0.00011867561,0.015339506,0.0020608911,0.0000841606,0.00047671498,0.000057920948,0.000019360245,0.01553986],"study_design_codex":"meta_analysis","study_design_gemma":"meta_analysis","domain_scores_codex":[0.9750881,0.0051438753,0.0127552645,0.0014714904,0.005269775,0.00027144654],"domain_scores_gemma":[0.9370376,0.052181385,0.00456027,0.0035205327,0.002458399,0.0002418461],"candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.15776965,0.00045714495,0.01351259,0.0048160017,0.000076900455,0.00042296707,0.0011207766,0.000108300344,0.01166231],"category_scores_gemma":[0.12148391,0.00018769738,0.011670987,0.011364219,0.00022769978,0.000419542,0.00007101832,0.000121099,0.0000805313],"study_design_candidate":"meta_analysis","study_design_consensus":"meta_analysis","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00009201311,0.00005966535,0.06955238,0.00069550925,0.8321115,0.00002449677,0.0017677413,0.0033123507,0.00030118274,0.0027163397,0.07270124,0.016665589],"study_design_scores_gemma":[0.00018445299,0.0001582983,0.0112749,0.00014906592,0.76126367,0.0000015144931,0.0005946385,0.21116342,0.00003541587,0.00079486636,0.014207496,0.0001722967],"about_ca_topic_score_codex":0.0002748791,"about_ca_topic_score_gemma":0.0005558083,"teacher_disagreement_score":0.9515388,"about_ca_system_score_codex":0.000047451365,"about_ca_system_score_gemma":0.00016679533,"threshold_uncertainty_score":0.9982269},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4406820295","doi":"10.1136/bmjebm-2024-113134","title":"Proposed framework for unifying disease definitions in guideline development","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Guideline; Disease; Computer science; Management science; Medicine; Engineering; Pathology","score_opus":0.5406540875678616,"score_gpt":0.5688595876034911,"score_spread":0.028205500035629494,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4406820295","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"methods","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.017801324,0.0015637684,0.60098106,0.37368062,0.0006574467,0.0048963,0.0000125526085,0.00013869627,0.0002682364],"genre_scores_gemma":[0.4919508,0.00024745706,0.44608784,0.058815274,0.00083737756,0.0009731295,0.00069548964,0.000046402092,0.00034621343],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9955395,0.00018237928,0.0026190127,0.0005681186,0.00065420964,0.00043677638],"domain_scores_gemma":[0.9894859,0.008223293,0.00042413516,0.00065100123,0.00085597683,0.00035968344],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.004664916,0.00027846033,0.0006064255,0.0005906634,0.00015825193,0.00001815526,0.00018182003,0.0001415238,0.0005057497],"category_scores_gemma":[0.09507141,0.00022902952,0.00011914262,0.0010350052,0.00017842252,0.0002205926,0.000057901296,0.00041541125,0.000024670428],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.027562382,0.0031810638,0.26676887,0.007697556,0.0005102592,0.00060259044,0.0009869862,0.0017527107,0.0036262642,0.09469452,0.26545522,0.3271616],"study_design_scores_gemma":[0.054281283,0.0031806633,0.22823304,0.12009007,0.0040117432,0.000036104837,0.0049392343,0.038355548,0.0047523337,0.14359003,0.3966768,0.0018531436],"about_ca_topic_score_codex":0.00012977223,"about_ca_topic_score_gemma":0.00011618844,"teacher_disagreement_score":0.4741495,"about_ca_system_score_codex":0.00046578323,"about_ca_system_score_gemma":0.004199839,"threshold_uncertainty_score":0.93395525},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4407132017","doi":"10.1136/bmjebm-2023-112389","title":"Rapid reviews methods series: guidance on rapid scoping, mapping and evidence and gap map (‘Big Picture Reviews’)","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":18,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"St. Michael's Hospital; Ottawa Hospital","funders":"","keywords":"Series (stratigraphy); Data science; Computer science; Information retrieval; Biology","score_opus":0.4793826492030341,"score_gpt":0.5678402709174526,"score_spread":0.08845762171441851,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407132017","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00018228323,0.7794216,0.0339619,0.17462005,0.0009805742,0.006910065,0.0000016694037,0.00021145526,0.0037103684],"genre_scores_gemma":[0.0031936145,0.86522776,0.104890116,0.018003566,0.0016777037,0.0018184257,0.0000032706969,0.00004216292,0.0051433914],"study_design_codex":"design_other","study_design_gemma":"not_applicable","domain_scores_codex":[0.9897276,0.0058310353,0.0014345425,0.0010688129,0.0011377679,0.0008002184],"domain_scores_gemma":[0.9897173,0.007634221,0.0006188886,0.0011628107,0.00043232302,0.00043446472],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04251332,0.00046929662,0.0013553043,0.0005603981,0.00076624175,0.0001447721,0.0009076035,0.00029603456,0.0005129893],"category_scores_gemma":[0.082266144,0.00035385453,0.000136855,0.0017718339,0.0020698763,0.00064883777,0.00024219487,0.0007341266,0.000030131761],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00020457845,0.000034407854,0.0014512736,0.010042578,0.000025260442,0.000040586045,0.0025768469,0.000001065391,0.0027000087,0.0031378826,0.16249031,0.8172952],"study_design_scores_gemma":[0.00041468567,0.00048181,0.0006625156,0.28852776,0.00007469059,0.000005942105,0.0010023825,0.00003896961,0.00092202803,0.002175558,0.7053387,0.00035494615],"about_ca_topic_score_codex":0.00052672747,"about_ca_topic_score_gemma":0.00023621673,"teacher_disagreement_score":0.81694025,"about_ca_system_score_codex":0.0003082581,"about_ca_system_score_gemma":0.001006762,"threshold_uncertainty_score":0.99989134},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4407402958","doi":"10.1136/bmjebm-2024-112919","title":"What makes a ‘good’ decision with artificial intelligence? A grounded theory study in paediatric care","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":true,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; York University; SickKids Foundation; University of Toronto; Canadian Institute for Advanced Research; Vector Institute","funders":"Canadian Institutes of Health Research; Hospital for Sick Children","keywords":"Grounded theory; Nonprobability sampling; Snowball sampling; Psychology; Psychological intervention; Qualitative research; Health care; Medical education; Applied psychology; Nursing; Medicine","score_opus":0.2084493904989476,"score_gpt":0.4812033996992079,"score_spread":0.2727540092002603,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407402958","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.91498345,0.01779484,0.025748504,0.032374687,0.002747389,0.0057435804,0.0000012789833,0.00013827573,0.0004680056],"genre_scores_gemma":[0.9930165,0.0010234745,0.0010426565,0.0030835138,0.000981804,0.0005912311,0.000021798562,0.000031654792,0.00020736892],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9956106,0.00052493747,0.0014625004,0.0008105795,0.0010204406,0.00057094917],"domain_scores_gemma":[0.99325967,0.0044807633,0.0002499827,0.0008935348,0.0008389215,0.0002771553],"candidate_categories":["metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.0034341593,0.00038667506,0.0007921835,0.0013328249,0.00020330577,0.00008547858,0.0002904698,0.00021541334,0.00033752035],"category_scores_gemma":[0.0051038,0.0002692867,0.00009913089,0.0028585547,0.00034860783,0.000525586,0.00005060086,0.0006174269,0.00006238695],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.007285833,0.0008480716,0.1390783,0.000621456,0.00005065664,0.0003085159,0.011412128,0.0002454524,0.000074948715,0.0026100893,0.00042399066,0.83704054],"study_design_scores_gemma":[0.0011938036,0.013614103,0.08043845,0.04064772,0.0011715727,0.00006554471,0.8000445,0.0016688785,0.005216945,0.054174904,0.00092115876,0.00084243197],"about_ca_topic_score_codex":0.0013830124,"about_ca_topic_score_gemma":0.0037413603,"teacher_disagreement_score":0.83619815,"about_ca_system_score_codex":0.00064311846,"about_ca_system_score_gemma":0.0024221598,"threshold_uncertainty_score":0.9999759},"labels":[{"model":"gemma","categories":["sts"],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4407402962","doi":"10.1136/bmjebm-2023-112588","title":"Expanded disease definitions in Alzheimer’s disease and the new era of disease-modifying drugs","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Dementia and Cognitive Impairment Research","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Calgary; University of Alberta","funders":"","keywords":"Disease; Medicine; Pathology","score_opus":0.09279324019988813,"score_gpt":0.3975662631856471,"score_spread":0.3047730229857589,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4407402962","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.45913514,0.09151218,0.0027761564,0.43841755,0.000253443,0.0059099356,0.000040157203,0.00009144163,0.0018639669],"genre_scores_gemma":[0.9883184,0.0035412817,0.00012091443,0.0067998883,0.0002087428,0.00049545954,0.000063218526,0.0000178738,0.00043421052],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99710697,0.00038263586,0.0006608596,0.000466823,0.0009627176,0.00042000998],"domain_scores_gemma":[0.99589443,0.0020586415,0.00013742683,0.0006188153,0.00025735077,0.0010333317],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0014740493,0.00026132682,0.00055704784,0.0004002405,0.00014248394,0.000024485404,0.00019540508,0.000040626204,0.00041889894],"category_scores_gemma":[0.005325935,0.00016531508,0.00016306793,0.0008207495,0.0010364269,0.00018156355,0.00010876081,0.00037487596,0.000009532535],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.06069879,0.0011332342,0.83456,0.0027817525,0.0005478137,0.0006594407,0.000885412,0.00005451804,0.00042171564,0.01729824,0.01928921,0.061669912],"study_design_scores_gemma":[0.02612607,0.00035020104,0.9247127,0.026266053,0.0036684624,0.000002360252,0.00074393896,0.002904005,0.00026785242,0.014063666,0.0006475342,0.00024712956],"about_ca_topic_score_codex":0.00034489686,"about_ca_topic_score_gemma":0.000016324113,"teacher_disagreement_score":0.52918327,"about_ca_system_score_codex":0.0000683697,"about_ca_system_score_gemma":0.0026318822,"threshold_uncertainty_score":0.6741354},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4408245171","doi":"10.1136/bmjebm-2024-113077","title":"Rating certainty when the target threshold is the null and the point estimate is close to the null","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":9,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"Einstein Stiftung Berlin; National Natural Science Foundation of China","keywords":"Certainty; Null hypothesis; Null (SQL); Harm; Confidence interval; Point estimation; Point (geometry); Range (aeronautics); Grading (engineering); Econometrics; Mathematics; Statistics; Computer science; Psychology; Social psychology; Data mining; Engineering","score_opus":0.5544545072542626,"score_gpt":0.533779388139586,"score_spread":0.020675119114676566,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408245171","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.009351308,0.021710891,0.029775148,0.92800164,0.000651152,0.005321885,0.000014140179,0.0000142868,0.0051595704],"genre_scores_gemma":[0.7165158,0.00041544155,0.0035984947,0.25585034,0.0009456702,0.00102852,0.0000039704355,0.00003017279,0.021611637],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.9746692,0.011079651,0.006529301,0.0012185246,0.0059741857,0.0005291013],"domain_scores_gemma":[0.9257597,0.0599296,0.0034981547,0.009029307,0.0015373081,0.00024593883],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.22326648,0.00050421094,0.002601177,0.00021641956,0.0012182159,0.0010039604,0.0043192743,0.000087850116,0.007346946],"category_scores_gemma":[0.1373863,0.00011545572,0.0010137409,0.0020731129,0.00094572816,0.00021893939,0.00047594134,0.00051400124,0.0006686706],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000124449,0.000015265288,0.0039338795,0.00008633265,0.00021276454,0.000006079075,0.008129672,0.0003654196,0.000033264638,0.009124847,0.96762097,0.010347043],"study_design_scores_gemma":[0.0021293415,0.00023573186,0.0057477965,0.0042176982,0.002196837,0.00003820795,0.014854908,0.20721635,0.0002951888,0.12864043,0.6339806,0.00044694656],"about_ca_topic_score_codex":0.00043957305,"about_ca_topic_score_gemma":0.00016124432,"teacher_disagreement_score":0.70716447,"about_ca_system_score_codex":0.000052747313,"about_ca_system_score_gemma":0.00031508997,"threshold_uncertainty_score":0.9935605},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4408383217","doi":"10.1136/bmjebm-2024-113527","title":"Improving peer review of systematic reviews and related review types by involving librarians and information specialists as methodological peer reviewers: a randomised controlled trial","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Ottawa Hospital","funders":"","keywords":"Peer review; Psychological intervention; Systematic review; Medicine; Randomized controlled trial; Family medicine; Inclusion (mineral); Intervention (counseling); MEDLINE; Medical education; Psychology; Nursing; Internal medicine; Social psychology","score_opus":0.6790106756671164,"score_gpt":0.5482039948117332,"score_spread":0.13080668085538316,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408383217","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000106783235,0.88385516,0.006676424,0.065288305,0.0006428916,0.039974794,0.0000065579616,0.000015615746,0.003433477],"genre_scores_gemma":[0.0006044837,0.944857,0.0053932164,0.031783804,0.00023083603,0.0039588194,0.00006528744,0.000019765559,0.013086781],"study_design_codex":"not_applicable","study_design_gemma":"systematic_review","domain_scores_codex":[0.8564191,0.0869972,0.04312805,0.0014938957,0.011460891,0.0005008814],"domain_scores_gemma":[0.65450734,0.23898953,0.071442574,0.0112239225,0.022706,0.0011306166],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5551242,0.0006588887,0.019062959,0.0005923758,0.00016007069,0.00033452275,0.0012069484,0.00022310358,0.004929542],"category_scores_gemma":[0.9314585,0.00027667233,0.0018805341,0.0026608207,0.0003079614,0.0014939982,0.00016977894,0.00038194354,0.00016263878],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004836698,0.000032091604,0.00003787976,0.4803014,0.00032135422,0.0000048450706,0.00014469493,4.169878e-7,0.000051463667,0.0015781774,0.50049776,0.012193169],"study_design_scores_gemma":[0.07677841,0.00033015065,0.00002422087,0.7830864,0.007735647,0.000029423614,0.0001442759,0.00077846274,0.00000651994,0.0009266523,0.12984477,0.00031507068],"about_ca_topic_score_codex":0.000030372112,"about_ca_topic_score_gemma":0.0000018508043,"teacher_disagreement_score":0.37633422,"about_ca_system_score_codex":0.00006577734,"about_ca_system_score_gemma":0.0004536362,"threshold_uncertainty_score":0.9999685},"labels":[],"label_agreement":null},{"id":"W4408488905","doi":"10.1136/bmjebm-2024-113050","title":"Identifying actionable statements in Chinese health guidelines: a cross-sectional study","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact; Cochrane","funders":"National Natural Science Foundation of China","keywords":"Guideline; Grading (engineering); Specialty; Family medicine; Medicine; Global Positioning System; Medical education; Computer science; Engineering; Pathology","score_opus":0.5934378988714254,"score_gpt":0.6822762201283746,"score_spread":0.08883832125694924,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408488905","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.8614223,0.0018833706,0.020399243,0.107909516,0.002293387,0.005627732,0.000011869779,0.00017160585,0.00028099454],"genre_scores_gemma":[0.96714145,0.00034947417,0.0039158873,0.02522832,0.00095769664,0.00047594003,0.00013555598,0.000035032703,0.0017606533],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99166054,0.00054621184,0.0046924567,0.00087716425,0.0016663618,0.00055728265],"domain_scores_gemma":[0.9934561,0.003216907,0.00091910374,0.0007781028,0.0013367265,0.0002930357],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.013916336,0.00036405004,0.0008992731,0.0010081119,0.00024205867,0.000067146284,0.00024954998,0.00011670285,0.0011365112],"category_scores_gemma":[0.045871176,0.0002892183,0.00013696913,0.001960222,0.0001622716,0.0009922191,0.00011610371,0.000629481,0.00006496163],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019297502,0.0007269288,0.97493434,0.0005321069,0.00010939831,0.000088646404,0.0001721355,0.00055878016,0.0003677878,0.000041670668,0.014499822,0.0060386104],"study_design_scores_gemma":[0.013918427,0.0014344928,0.9737404,0.003595424,0.00010814866,0.000023520513,0.001637795,0.0023084849,0.000025632362,0.0006395967,0.0023797252,0.00018835458],"about_ca_topic_score_codex":0.009275998,"about_ca_topic_score_gemma":0.001895097,"teacher_disagreement_score":0.10571916,"about_ca_system_score_codex":0.0011667717,"about_ca_system_score_gemma":0.0029260216,"threshold_uncertainty_score":0.999956},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4408555504","doi":"10.1136/bmjebm-2024-112974","title":"Analgesic effects of non-surgical and non-interventional treatments for low back pain: a systematic review and meta-analysis of placebo-controlled randomised trials","year":2025,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Musculoskeletal pain and rehabilitation","field":"Medicine","cited_by":22,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Memorial University of Newfoundland","funders":"","keywords":"Medicine; Placebo; Meta-analysis; Analgesic; Randomized controlled trial; Low back pain; Chronic pain; Physical therapy; Confidence interval; Strictly standardized mean difference; Relative risk; Clinical trial; Anesthesia; Internal medicine; Alternative medicine","score_opus":0.09458008446055835,"score_gpt":0.4391426482895198,"score_spread":0.34456256382896144,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408555504","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000007299719,0.93345076,0.0020871616,0.0010526467,0.00006641797,0.06297216,0.00034155668,0.00000903516,0.000012937561],"genre_scores_gemma":[0.0001378326,0.9859987,0.00036358924,0.00025671467,0.0000781994,0.010952271,0.0012078121,0.000027192362,0.0009777002],"study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","domain_scores_codex":[0.98560905,0.0066499775,0.0059237736,0.00073020975,0.0008271882,0.00025981263],"domain_scores_gemma":[0.9092165,0.085320175,0.0038897356,0.0007747065,0.00054937816,0.0002494694],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad"],"consensus_categories":["metaepi_broad"],"category_scores_codex":[0.028333178,0.00079247664,0.03770564,0.0013113366,0.00004538558,0.000009184338,0.00017002891,0.00033042772,0.0003926552],"category_scores_gemma":[0.06302699,0.0003806059,0.010102854,0.0012334866,0.0003295249,0.000059137805,0.00003080135,0.0001881513,0.0000013019205],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0047015967,0.0002053033,0.0000016199772,0.8040515,0.18798006,0.000008821257,0.000015649113,8.601828e-7,0.0000033425326,0.000004844254,0.001102799,0.0019235794],"study_design_scores_gemma":[0.041431446,0.0013894758,0.0000056298973,0.44473824,0.5120553,0.000002836715,0.000006444274,0.00013779351,0.0000012118549,0.0000039119354,0.0001008081,0.00012696092],"about_ca_topic_score_codex":0.000032258315,"about_ca_topic_score_gemma":0.0000034850075,"teacher_disagreement_score":0.3593133,"about_ca_system_score_codex":0.0001020083,"about_ca_system_score_gemma":0.00054884737,"threshold_uncertainty_score":0.9998646},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W4408660425","doi":"10.1136/bmjebm-2024-113006","title":"The path of Chile towards the institutionalisation of evidence-based health policy","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":2,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University","funders":"","keywords":"Institutionalisation; Path (computing); Political science; Path dependence; Economics; Computer science; Neoclassical economics; Law","score_opus":0.32637414548648297,"score_gpt":0.5374067296486567,"score_spread":0.2110325841621737,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4408660425","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.005036432,0.024610056,0.0055520074,0.9563512,0.0016501208,0.0034364255,0.000043713284,0.00007120471,0.003248837],"genre_scores_gemma":[0.8053371,0.007166864,0.0004446775,0.18378463,0.0013207651,0.00060935423,0.000049292386,0.0000200502,0.0012673091],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.99357444,0.002234576,0.0020510394,0.0003307459,0.0011724636,0.0006367203],"domain_scores_gemma":[0.9828908,0.013445567,0.0013555114,0.001129219,0.0009937544,0.0001851679],"candidate_categories":["metaresearch","sts"],"consensus_categories":[],"category_scores_codex":[0.011139962,0.00025168556,0.00077417254,0.00037985315,0.0014200311,0.000006079735,0.00074484374,0.00017399601,0.0002158195],"category_scores_gemma":[0.016766429,0.00011696501,0.00016981193,0.00149483,0.0009815779,0.00023459838,0.00011759406,0.00065630756,0.000014943489],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0026553376,0.00021550803,0.097592644,0.013922226,0.00014926364,0.0000059701815,0.0037958112,0.00045608927,0.0005689895,0.24897741,0.4856263,0.14603442],"study_design_scores_gemma":[0.00553035,0.001882619,0.52347285,0.076564245,0.00025093625,0.0000023415284,0.0033592954,0.00096068194,0.0008406088,0.009017539,0.3777391,0.00037940446],"about_ca_topic_score_codex":0.018870667,"about_ca_topic_score_gemma":0.0010510702,"teacher_disagreement_score":0.8003006,"about_ca_system_score_codex":0.0012212576,"about_ca_system_score_gemma":0.047832962,"threshold_uncertainty_score":0.99987996},"labels":[{"model":"gemma","categories":["sts"],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4409047603","doi":"10.1136/bmjebm-2024-113322","title":"Leading with options or issues to support purposeful shared decision-making in clinical practice","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":4,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Ottawa; Ottawa Hospital","funders":"","keywords":"Clinical Practice; Psychology; Process management; Medicine; Nursing; Business","score_opus":0.4683492103640248,"score_gpt":0.6216772883195422,"score_spread":0.15332807795551734,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409047603","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.039410505,0.014402214,0.034889996,0.8794832,0.0046191043,0.01747921,0.000041017352,0.00063690037,0.009037819],"genre_scores_gemma":[0.7192343,0.0013414198,0.18121222,0.089798614,0.0009457731,0.003803841,0.000052865977,0.00007615196,0.0035348497],"study_design_codex":"observational","study_design_gemma":"not_applicable","domain_scores_codex":[0.9914398,0.0030368918,0.0029678645,0.0007808219,0.0009909492,0.0007836662],"domain_scores_gemma":[0.94435704,0.05076309,0.0008801833,0.0020108933,0.0015762035,0.00041257803],"candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0062362724,0.00031720698,0.00090915733,0.00088130415,0.0007411124,0.000022319797,0.00095366774,0.0003283861,0.0014455611],"category_scores_gemma":[0.1333689,0.00022530255,0.00006935575,0.0023750162,0.00023482749,0.00074535643,0.00036917013,0.0016303546,0.0004415047],"study_design_candidate":"not_applicable","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.014928588,0.00054012274,0.52547544,0.0024214613,0.00010090339,0.00041245756,0.018143341,0.0007014049,0.000044993438,0.0040482003,0.32712907,0.10605403],"study_design_scores_gemma":[0.008463705,0.005360915,0.18077852,0.24399076,0.00032343477,0.000024657023,0.05636337,0.0020623114,0.00001593435,0.0013937666,0.5003921,0.0008305504],"about_ca_topic_score_codex":0.00086616434,"about_ca_topic_score_gemma":0.003034138,"teacher_disagreement_score":0.7896846,"about_ca_system_score_codex":0.00079608604,"about_ca_system_score_gemma":0.005219998,"threshold_uncertainty_score":0.99946725},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4409283522","doi":"10.1136/bmjebm-2024-113066","title":"Novel AI applications in systematic review: GPT-4 assisted data extraction, analysis, review of bias","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children; University of Toronto","funders":"","keywords":"Data extraction; Concordance; Comparability; Computer science; Meta-analysis; Systematic review; Statistics; Data mining; Artificial intelligence; Medicine; MEDLINE; Mathematics; Internal medicine; Biology","score_opus":0.8422006587199937,"score_gpt":0.6309828565757105,"score_spread":0.2112178021442832,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409283522","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000017321066,0.60899454,0.33971003,0.04215346,0.00014266236,0.007843895,0.000049195594,0.000010725601,0.0010781658],"genre_scores_gemma":[0.082199074,0.6787473,0.037296884,0.1561381,0.00068422256,0.011648525,0.002573401,0.00010763688,0.030604886],"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","domain_scores_codex":[0.944508,0.015642062,0.029284943,0.001947154,0.008272148,0.00034569285],"domain_scores_gemma":[0.8758671,0.060002137,0.02553068,0.030513536,0.007701524,0.00038497924],"candidate_categories":["metaresearch","bibliometrics","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.30745146,0.00049164245,0.012483934,0.0021458066,0.00010652253,0.00011385581,0.0050457995,0.00011500141,0.00561469],"category_scores_gemma":[0.45812473,0.00022937424,0.0016991157,0.021817638,0.00019793198,0.0005578843,0.00024326239,0.00031940945,0.0002729697],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000007035688,0.00019400685,0.00956817,0.65205747,0.0010543863,0.000004426006,0.00002008026,0.000043581582,0.00009587085,0.0008410975,0.3346226,0.001491321],"study_design_scores_gemma":[0.00033471506,0.000033044504,0.008648696,0.9242257,0.025822753,0.0000102463755,0.00020363227,0.0064741885,0.000012288249,0.0002060724,0.0337791,0.00024958607],"about_ca_topic_score_codex":0.00017443876,"about_ca_topic_score_gemma":0.00022639877,"teacher_disagreement_score":0.30241317,"about_ca_system_score_codex":0.000106755324,"about_ca_system_score_gemma":0.0007461686,"threshold_uncertainty_score":0.9989741},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"simulation_or_modeling","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4409642815","doi":"10.1136/bmjebm-2024-113171","title":"Challenges in the selection and measurement of outcomes in psychiatric trials","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Health Systems, Economic Evaluations, Quality of Life","field":"Economics, Econometrics and Finance","cited_by":6,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; St. Joseph’s Healthcare Hamilton; Impact","funders":"Danmarks Frie Forskningsfond","keywords":"Selection (genetic algorithm); Psychiatry; Psychology; Medicine; Computer science; Artificial intelligence","score_opus":0.7992621237512433,"score_gpt":0.5470442325903778,"score_spread":0.25221789116086546,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4409642815","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.100889035,0.0944278,0.0014357453,0.7980061,0.0008373085,0.003160601,0.00007035889,0.000017704344,0.0011553125],"genre_scores_gemma":[0.98748404,0.003936728,0.00038367734,0.0076451977,0.00017072752,0.0003456228,0.0000075444127,0.0000065040463,0.000019939123],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.99280983,0.0016659563,0.0047258595,0.0003767766,0.0002089551,0.00021260197],"domain_scores_gemma":[0.99143,0.0065262946,0.0015712503,0.00034846825,0.00008088228,0.000043112756],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.12979004,0.00013917025,0.0013715808,0.0010067133,0.00004497339,0.000010553066,0.00020061049,0.00009722961,0.000057717334],"category_scores_gemma":[0.04939642,0.000110726876,0.00006871263,0.00054823153,0.00006328936,0.0001682042,0.0000130205835,0.00015749423,0.0000115092735],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00017481881,0.00027415366,0.9058133,0.00245567,0.00007237457,0.0000011838648,0.0017263107,0.00029317482,0.00001353998,0.052618716,0.032565538,0.00399124],"study_design_scores_gemma":[0.0033032468,0.0002909707,0.9692701,0.0023884336,0.00002836693,0.0000014414146,0.0018869766,0.0010552534,0.000009455082,0.013832123,0.007778713,0.00015490358],"about_ca_topic_score_codex":0.0021223438,"about_ca_topic_score_gemma":0.0034621968,"teacher_disagreement_score":0.886595,"about_ca_system_score_codex":0.00033425805,"about_ca_system_score_gemma":0.00029052593,"threshold_uncertainty_score":0.9586109},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4410343249","doi":"10.1136/bmjebm-2025-113825","title":"Reporting guideline for the use of Generative Artificial intelligence tools in MEdical Research: the GAMER Statement","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":49,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Queen's University","funders":"Chinese Academy of Medical Sciences; Lanzhou University","keywords":"Declaration; Delphi method; Checklist; Transparency (behavior); Delphi; Computer science; Guideline; Artificial intelligence; Medical education; Psychology; Data science; Medicine; Computer security","score_opus":0.8920578848643578,"score_gpt":0.6593057290771102,"score_spread":0.23275215578724762,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4410343249","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.041551623,0.0036291534,0.24710749,0.70171666,0.0011146098,0.0047752187,0.0000050197564,0.000024509853,0.000075680735],"genre_scores_gemma":[0.9635655,0.0019745668,0.0066788746,0.02367564,0.0020195302,0.0015150288,0.000048248614,0.000026071943,0.00049653574],"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.99148726,0.0006849143,0.005233718,0.0004609276,0.001592694,0.000540486],"domain_scores_gemma":[0.9611146,0.03272915,0.001488466,0.0011764419,0.0032633739,0.00022800705],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.032676578,0.0001886593,0.0005692339,0.00040070646,0.00030863954,0.000036369158,0.00035134223,0.00015622035,0.00046414032],"category_scores_gemma":[0.22029842,0.00009697883,0.000120869445,0.0015289439,0.0011168309,0.00018067166,0.00007598887,0.0007014473,0.000007193173],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.001797611,0.00026752392,0.0056769205,0.0005231568,0.000075836746,0.000051217852,0.003036028,0.0024912383,0.0015637893,0.0082497625,0.09826717,0.8779997],"study_design_scores_gemma":[0.00057064876,0.004319024,0.0066663306,0.040647022,0.00060264417,0.00005783594,0.056071598,0.42423666,0.18366565,0.032676287,0.24992839,0.0005579235],"about_ca_topic_score_codex":0.005603437,"about_ca_topic_score_gemma":0.0017361343,"teacher_disagreement_score":0.9220139,"about_ca_system_score_codex":0.00030806864,"about_ca_system_score_gemma":0.0040105544,"threshold_uncertainty_score":0.99606305},"labels":[],"label_agreement":null},{"id":"W4411037095","doi":"10.1136/bmjebm-2025-113804","title":"Top 15 Choosing Wisely international campaign recommendations to reduce low-value care","year":2025,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Healthcare cost, quality, practices","field":"Health Professions","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"","keywords":"Value (mathematics); Business; Computer science","score_opus":0.6366719556149726,"score_gpt":0.6106770473141256,"score_spread":0.02599490830084694,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411037095","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0006088264,0.002399324,0.0031842026,0.9605717,0.020024432,0.006089974,0.00035164025,0.00038684224,0.0063830568],"genre_scores_gemma":[0.0017543262,0.0008202662,0.0045233783,0.93898535,0.033516444,0.0021923753,0.0025927357,0.00013355984,0.015481536],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.983334,0.0081482595,0.0031118977,0.0016088397,0.002331384,0.0014656216],"domain_scores_gemma":[0.97170556,0.02027867,0.00189621,0.0020064532,0.0034498782,0.00066322833],"candidate_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity","insufficient_payload"],"category_scores_codex":[0.008365056,0.0008513659,0.0014854436,0.0015037828,0.0012607782,0.00006184875,0.0016551699,0.0017664848,0.0058904765],"category_scores_gemma":[0.057235375,0.0007677735,0.00023477692,0.0013344112,0.0002635858,0.0006735749,0.00043539266,0.00816467,0.0009724272],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":true,"study_design_scores_codex":[0.00025598067,0.000035193494,0.0017008873,0.0056623397,0.00009984733,0.0002054999,0.0051212134,0.000020016125,0.000056224515,0.00016925209,0.9739525,0.0127210915],"study_design_scores_gemma":[0.001194499,0.00026678154,0.0007809262,0.08736485,0.00027158015,0.0000070276,0.0057795234,0.00004684957,0.00005336838,0.00016052662,0.90347856,0.0005955074],"about_ca_topic_score_codex":0.0068734316,"about_ca_topic_score_gemma":0.00065990264,"teacher_disagreement_score":0.08170251,"about_ca_system_score_codex":0.0039112945,"about_ca_system_score_gemma":0.011727303,"threshold_uncertainty_score":0.9999125},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W4411952500","doi":"10.1136/bmjebm-2024-113617","title":"Understanding synthetic data: artificial datasets for real-world evidence","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":7,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McGill University","funders":"","keywords":"Computer science; Artificial intelligence","score_opus":0.5243495068028722,"score_gpt":0.4894084368243749,"score_spread":0.034941069978497274,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4411952500","genre_codex":"methods","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.000118741285,0.0020285377,0.81691974,0.17655288,0.0018760418,0.0016385277,0.00006324221,0.00033416387,0.00046813942],"genre_scores_gemma":[0.92645323,0.0006974111,0.05718803,0.0122120315,0.0018061331,0.0004673128,0.000425506,0.00005569949,0.00069465395],"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","domain_scores_codex":[0.9950661,0.00063338433,0.0010622902,0.0015217923,0.0009864168,0.0007300247],"domain_scores_gemma":[0.9813021,0.013448229,0.00042466857,0.004284071,0.0002560719,0.00028490656],"candidate_categories":["metaresearch","metaepi_narrow"],"consensus_categories":[],"category_scores_codex":[0.007633783,0.00036089454,0.00062429043,0.00073839986,0.0005183475,0.00018925956,0.0036846078,0.000115933566,0.00008476258],"category_scores_gemma":[0.022985118,0.00030442185,0.000075845426,0.0018767108,0.0003361723,0.0015894928,0.00075857766,0.00048231476,0.000028366907],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.00065173465,0.00012713388,0.009061246,0.00415182,0.00008903632,0.00017513699,0.0004048429,0.0018161999,0.0008885038,0.53381073,0.39417067,0.054652944],"study_design_scores_gemma":[0.0012498347,0.0008992286,0.003534209,0.04558376,0.0002696539,0.000019994426,0.00023262223,0.87049747,0.00080175325,0.04241838,0.03367751,0.00081559026],"about_ca_topic_score_codex":0.0017040292,"about_ca_topic_score_gemma":0.0012937565,"teacher_disagreement_score":0.9263345,"about_ca_system_score_codex":0.0006276823,"about_ca_system_score_gemma":0.0016558492,"threshold_uncertainty_score":0.9999408},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4412521978","doi":"10.1136/bmjebm-2025-113715","title":"Co-production and implementation of an evidence collation strategy for a novel point-of-care information resource: gpevidence.org","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Hospital for Sick Children","funders":"National Institute for Health and Care Research","keywords":"Collation; Production (economics); Resource (disambiguation); Point (geometry); Information resource; Computer science; Business; Knowledge management; Mathematics; Economics","score_opus":0.07713404106508719,"score_gpt":0.4283125794856482,"score_spread":0.351178538420561,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4412521978","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.93966615,0.0066141407,0.045697816,0.0061936313,0.00021930311,0.0014928643,0.00003368845,0.000023600202,0.00005878286],"genre_scores_gemma":[0.9949123,0.0003672764,0.0038697221,0.00033926268,0.00017121449,0.00009582269,0.00020810033,0.000005316527,0.000030979936],"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","domain_scores_codex":[0.9986864,0.000082971426,0.00058163574,0.00024841528,0.00025882266,0.00014175635],"domain_scores_gemma":[0.99852157,0.00026400134,0.00037263066,0.00028654528,0.00049968733,0.000055580054],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001208088,0.0001281623,0.00024145139,0.00016810582,0.000076419026,0.00000928521,0.0001310713,0.00014465339,0.000012322968],"category_scores_gemma":[0.004406238,0.00010704386,0.00004212933,0.00024178211,0.00032666686,0.0000596355,0.000021455948,0.000057425215,2.4498306e-7],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0027166007,0.000056139957,0.03079923,0.0034720805,0.000077034405,4.9555746e-7,0.0013629314,0.0004407408,0.6893231,0.00034060574,0.0066105435,0.26480046],"study_design_scores_gemma":[0.0033643267,0.008561702,0.09696009,0.009114004,0.00028304468,0.000010386756,0.023299105,0.0008078614,0.848332,0.00017040834,0.00877322,0.00032387918],"about_ca_topic_score_codex":0.00012409563,"about_ca_topic_score_gemma":0.00007876086,"teacher_disagreement_score":0.2644766,"about_ca_system_score_codex":0.000029553761,"about_ca_system_score_gemma":0.0003893556,"threshold_uncertainty_score":0.5274997},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4414833709","doi":"10.1136/bmjebm-2025-113704","title":"Addressing misleading medical information on social media: a scoping review of current interventions","year":2025,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Misinformation and Its Impacts","field":"Social Sciences","cited_by":6,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Impact","funders":"National Health and Medical Research Council; Medical Research Council","keywords":"Psychological intervention; Social media; Public health; Public policy; Medical information; MEDLINE; Digital health","score_opus":0.632896534837736,"score_gpt":0.6151385818461708,"score_spread":0.01775795299156524,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4414833709","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[1.18543795e-7,0.98214304,0.0015590049,0.006598961,0.0017871811,0.00291276,0.000025273488,0.000080908896,0.0048927404],"genre_scores_gemma":[0.000005468201,0.99641114,0.000045244626,0.001935496,0.001222433,0.00011970521,0.00019313733,0.000011689394,0.00005568845],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.991809,0.0014139543,0.0030834032,0.00023650039,0.003014222,0.00044291705],"domain_scores_gemma":[0.99012655,0.006074909,0.0024585659,0.00035296383,0.00059559743,0.00039140382],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.010968602,0.00038826623,0.0023211574,0.0009961808,0.00046008336,0.000060918934,0.0007456231,0.00040043186,0.0031212806],"category_scores_gemma":[0.11990889,0.00029031845,0.00080045895,0.001681661,0.000582158,0.000963389,0.000083849176,0.00075924234,0.000091473885],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":true,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000004229459,0.000017400438,1.0360426e-7,0.394318,0.000014251732,0.0000010739811,0.0012984255,8.933902e-8,2.2878035e-9,0.0009988586,0.029549718,0.5737978],"study_design_scores_gemma":[0.00021967477,0.000032965043,7.3974707e-7,0.68865293,0.00044511005,0.0000012950159,0.00038818934,0.0000027967458,2.0092449e-7,0.000013373653,0.3100863,0.00015642853],"about_ca_topic_score_codex":0.00004562599,"about_ca_topic_score_gemma":0.00008355062,"teacher_disagreement_score":0.5736414,"about_ca_system_score_codex":0.00046476355,"about_ca_system_score_gemma":0.010082385,"threshold_uncertainty_score":0.9999549},"labels":[{"model":"gemma","categories":["scholarly_communication"],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":[],"domain":null,"study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W4415279421","doi":"10.1136/bmjebm-2024-113391","title":"Over 1000 terms have been used to describe evidence synthesis: a scoping review","year":2025,"lang":"en","type":"review","venue":"BMJ evidence-based medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Public Health Ontario; Impact; McGill University; McMaster University; Queen's University; St. Michael's Hospital","funders":"National Health and Medical Research Council","keywords":"Identification (biology); Scope (computer science); Context (archaeology); Key (lock); Perspective (graphical); Systematic review","score_opus":0.8775033471675885,"score_gpt":0.6368221377890774,"score_spread":0.24068120937851112,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415279421","genre_codex":"review","genre_gemma":"review","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"review","genre_consensus":"review","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[8.759518e-7,0.9527014,0.008381511,0.013847043,0.0010144606,0.022646582,0.000044808723,0.00004861874,0.0013147134],"genre_scores_gemma":[0.000007998706,0.9704414,0.001847973,0.009013877,0.0010499612,0.006707755,0.000024336901,0.00009310302,0.010813584],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.88622254,0.04753166,0.039871857,0.005409221,0.019610357,0.0013543525],"domain_scores_gemma":[0.74084526,0.20902778,0.024465704,0.020533154,0.0035096635,0.0016184441],"candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"category_scores_codex":[0.3014744,0.0023899167,0.04077239,0.0031505285,0.00036539513,0.0009912113,0.009891462,0.00063581776,0.035333283],"category_scores_gemma":[0.669393,0.0010985795,0.0088324305,0.008785521,0.00037554116,0.0009702796,0.00081817375,0.0010060173,0.006871364],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.000010215979,0.000025684618,0.0000939474,0.27074444,0.00029141066,0.000082081926,0.000055181783,0.0000033242286,6.2057137e-7,0.000036922338,0.19921944,0.52943677],"study_design_scores_gemma":[0.00008993633,0.000102957165,0.000012688538,0.58312535,0.009521466,0.000026730228,0.000026125486,0.000029561545,0.00000140388,0.000041581472,0.40643367,0.0005885561],"about_ca_topic_score_codex":0.0001247146,"about_ca_topic_score_gemma":0.0001829948,"teacher_disagreement_score":0.5288482,"about_ca_system_score_codex":0.0006311312,"about_ca_system_score_gemma":0.005275949,"threshold_uncertainty_score":0.99914646},"labels":[{"model":"gemma","categories":["metaresearch","bibliometrics"],"domain":"reporting","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W4415406762","doi":"10.1136/bmjebm-2025-113675","title":"Comparing traditional and AI-enhanced strategies for developing patient decision aids: a multiple case study","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Healthcare Decision-Making and Restraints","field":"Psychology","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":true,"ca_institutions":"Université Laval","funders":"","keywords":"Context (archaeology); Key (lock); Data collection; Identification (biology)","score_opus":0.24972038070113398,"score_gpt":0.478647880144518,"score_spread":0.228927499443384,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415406762","genre_codex":"empirical","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":"empirical","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.781031,0.0009257795,0.21171781,0.0029533764,0.0012045673,0.0019572712,0.000008036637,0.000064472006,0.0001376635],"genre_scores_gemma":[0.99088746,0.000015600892,0.0061960877,0.0019905618,0.00024572638,0.00060591125,0.000011031447,0.000016911477,0.0000307066],"study_design_codex":"design_other","study_design_gemma":"qualitative","domain_scores_codex":[0.9973669,0.00031612473,0.0008844797,0.00061920815,0.0004342193,0.00037904858],"domain_scores_gemma":[0.9904946,0.008427497,0.00018332401,0.00039848586,0.00034140798,0.00015466473],"candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0016183327,0.00024754854,0.0005321371,0.0004119987,0.00036033508,0.00004754704,0.00015876403,0.0001312901,0.00006269923],"category_scores_gemma":[0.003253368,0.00018723242,0.000060474882,0.00042315712,0.00017656459,0.0001541506,0.000029037561,0.00027060983,0.000006354111],"study_design_candidate":"design_other","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.004534769,0.0013109108,0.029151587,0.00085655565,0.0003709502,0.011199796,0.024886409,0.00082415494,0.00062743004,0.005629676,0.008586228,0.9120215],"study_design_scores_gemma":[0.043844003,0.01976539,0.20321219,0.063570134,0.00065678847,0.0057172426,0.6031477,0.01289362,0.0005727803,0.037948478,0.006685715,0.001985978],"about_ca_topic_score_codex":0.001077737,"about_ca_topic_score_gemma":0.00072860357,"teacher_disagreement_score":0.91003555,"about_ca_system_score_codex":0.00013621837,"about_ca_system_score_gemma":0.0007244844,"threshold_uncertainty_score":0.76351166},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W4415524767","doi":"10.1136/bmjebm-2025-114275","title":"Correspondence on ‘Less is more for patients, practitioners, public and planet: a taxonomy for the harms of too much medicine’","year":2025,"lang":"en","type":"letter","venue":"BMJ evidence-based medicine","topic":"Pharmaceutical industry and healthcare","field":"Pharmacology, Toxicology and Pharmaceutics","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Centre for Addiction and Mental Health","funders":"","keywords":"Taxonomy (biology); Focus (optics)","score_opus":0.6890357720182831,"score_gpt":0.5719885798623117,"score_spread":0.11704719215597148,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4415524767","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.00029749292,0.005622052,0.00116491,0.9774376,0.003607662,0.009052815,0.0023895407,0.000057350488,0.000370589],"genre_scores_gemma":[0.006653063,0.0024098882,0.00015755033,0.97204566,0.0073171034,0.004880637,0.0019670674,0.00005139455,0.004517627],"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","domain_scores_codex":[0.99537045,0.0008707362,0.0012510435,0.00087782874,0.00073698227,0.0008929842],"domain_scores_gemma":[0.9622447,0.03478817,0.001018438,0.00062193174,0.000928102,0.0003986526],"candidate_categories":["metaepi_narrow","research_integrity","insufficient_payload"],"consensus_categories":["research_integrity"],"category_scores_codex":[0.0034082644,0.0006636565,0.0011543266,0.0004321142,0.00059112615,0.000022426239,0.00082466815,0.0019491016,0.0020966586],"category_scores_gemma":[0.007016827,0.00043200966,0.00015282078,0.00041055056,0.0017934688,0.00020503171,0.00008277355,0.0043153595,0.000009333768],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0036349792,0.00014807534,0.0014273014,0.0043354724,0.000321312,0.000037731526,0.00026556788,0.000014527259,0.000044348402,0.00044725643,0.97177076,0.017552692],"study_design_scores_gemma":[0.005634671,0.0020969932,0.00020512922,0.0038007128,0.0013567659,0.00001093443,0.00047215124,0.0021433844,0.0004495742,0.00023478815,0.98320585,0.00038906268],"about_ca_topic_score_codex":0.00014497829,"about_ca_topic_score_gemma":0.000009322443,"teacher_disagreement_score":0.033917435,"about_ca_system_score_codex":0.00020358436,"about_ca_system_score_gemma":0.0012684683,"threshold_uncertainty_score":0.9998132},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4416430973","doi":"10.1136/bmjebm-2025-113904","title":"A primer to planetary health in evidence-based medicine and clinical decision-making","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Climate Change and Health Impacts","field":"Environmental Science","cited_by":0,"is_retracted":false,"has_abstract":false,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"Impact; Queen's University; McMaster University; Western University","funders":"","keywords":"Primer (cosmetics); MEDLINE; Public health; Biomedicine","score_opus":0.23348970559746515,"score_gpt":0.5002367385224676,"score_spread":0.26674703292500246,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4416430973","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.28849998,0.02003157,0.005354584,0.6807314,0.0011876004,0.0034200714,0.000010333867,0.000112613714,0.00065184414],"genre_scores_gemma":[0.7499115,0.0060158884,0.0054059746,0.2377103,0.00070036936,0.00010896831,0.000017544147,0.00003064661,0.00009883265],"study_design_codex":"observational","study_design_gemma":"observational","domain_scores_codex":[0.9947017,0.0007015808,0.0017711944,0.0009654928,0.0009751132,0.000884939],"domain_scores_gemma":[0.98267007,0.015247893,0.0003691059,0.00077853363,0.000042804673,0.0008916022],"candidate_categories":["metaresearch","metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.011477882,0.000382752,0.0010711702,0.00054927665,0.00021187078,0.000018848823,0.00046372687,0.00018727231,0.0028158005],"category_scores_gemma":[0.021889543,0.00028000172,0.00006467337,0.0014904484,0.0006788506,0.00034280456,0.0002080209,0.0005101785,0.00015059297],"study_design_candidate":"observational","study_design_consensus":"observational","about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0026131584,0.0001413727,0.48694018,0.00048159566,0.000010209184,0.00012987287,0.0006137216,0.00055568846,0.00013154797,0.0000270395,0.2308249,0.2775307],"study_design_scores_gemma":[0.0038086905,0.0032513998,0.867913,0.10491435,0.000074545525,0.000013579556,0.00033496204,0.0042828796,0.00001917453,0.00089586026,0.014134841,0.0003566843],"about_ca_topic_score_codex":0.004965078,"about_ca_topic_score_gemma":0.003931762,"teacher_disagreement_score":0.4614115,"about_ca_system_score_codex":0.00065427425,"about_ca_system_score_gemma":0.0005395833,"threshold_uncertainty_score":0.9999652},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W4417300166","doi":"10.1136/bmjebm-2025-113997","title":"Reporting GUideline for Intervention DEscription in Rehabilitation (GUIDE-Rehab): a tool to open the ‘black box’ of rehabilitation complex interventions","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Cerebral Palsy and Movement Disorders","field":"Medicine","cited_by":7,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"University of Toronto; Ontario Tech University; University of Alberta","funders":"","keywords":"Rehabilitation; Psychological intervention; Delphi method; Intervention (counseling); Guideline; Transparency (behavior); Quality (philosophy); Rehabilitation counseling","score_opus":0.16649353114868318,"score_gpt":0.47692042697819914,"score_spread":0.310426895829516,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417300166","genre_codex":"commentary","genre_gemma":"empirical","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"empirical","genre_consensus":null,"domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.32583177,0.00068891403,0.2863507,0.3680224,0.00056138315,0.017408006,0.000016611573,0.00006824517,0.0010519637],"genre_scores_gemma":[0.92942846,0.000023691802,0.0599277,0.006933358,0.00013566687,0.0018612952,0.00025241583,0.000024734878,0.001412692],"study_design_codex":"not_applicable","study_design_gemma":"observational","domain_scores_codex":[0.9931959,0.00034455504,0.005193857,0.0005262626,0.0004625272,0.00027693564],"domain_scores_gemma":[0.99393284,0.0023645835,0.0017206962,0.0007754752,0.0011246877,0.00008169237],"candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.014966092,0.0002218372,0.00078324106,0.00065248366,0.00010818867,0.000032080836,0.000310421,0.000095280215,0.00026822012],"category_scores_gemma":[0.09104953,0.00016439914,0.00042503516,0.0011368694,0.0002849194,0.00037771938,0.00013020843,0.00017031195,0.000003999552],"study_design_candidate":"observational","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.010026153,0.0022503496,0.19436088,0.015340695,0.00025463602,0.000007864962,0.004215942,0.0028284295,0.040729214,0.016511006,0.565506,0.14796887],"study_design_scores_gemma":[0.020581018,0.01521591,0.737325,0.1169015,0.0009567603,0.000005889698,0.022129467,0.013028551,0.001369762,0.016079163,0.05590237,0.000504596],"about_ca_topic_score_codex":0.00243719,"about_ca_topic_score_gemma":0.0022061751,"teacher_disagreement_score":0.6035967,"about_ca_system_score_codex":0.00038644474,"about_ca_system_score_gemma":0.0003296431,"threshold_uncertainty_score":0.91660696},"labels":[],"label_agreement":null},{"id":"W4417321183","doi":"10.1136/bmjebm-2025-113925","title":"Current knowledge on the adaptation of clinical practice guidelines: a scoping review","year":2025,"lang":"en","type":"article","venue":"BMJ evidence-based medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"route_ca_aff":true,"route_ca_fund":false,"route_ca_venue":false,"route_about_ca":false,"ca_institutions":"McMaster University; Hamilton Health Sciences","funders":"Chinese University of Hong Kong","keywords":"Adaptation (eye); Quality (philosophy); Clinical Practice; Current (fluid); Context (archaeology)","score_opus":0.7839957976028011,"score_gpt":0.70014721165082,"score_spread":0.08384858595198119,"validation_status":"score_only:v0-immature-baseline","prediction":{"id":"W4417321183","genre_codex":"commentary","genre_gemma":"commentary","domain_codex":null,"domain_gemma":null,"model_version":"codex-gemma-dda1882f352a","genre_candidate":"commentary","genre_consensus":"commentary","domain_candidate":null,"domain_consensus":null,"prediction_status":"machine_predicted_unvalidated","genre_scores_codex":[0.0014187183,0.24225453,0.03663807,0.70885676,0.0021841826,0.0066617304,0.000003832589,0.00007768338,0.0019044722],"genre_scores_gemma":[0.07250183,0.42230168,0.03573089,0.4609784,0.0063857734,0.0012864706,0.0001696091,0.00009184511,0.00055349414],"study_design_codex":"design_other","study_design_gemma":"systematic_review","domain_scores_codex":[0.9900617,0.0016466111,0.006244663,0.00057638215,0.0011966568,0.0002739889],"domain_scores_gemma":[0.91580486,0.072633184,0.0028119844,0.0015195978,0.0069835573,0.00024683028],"candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.031056978,0.0002818681,0.0012053564,0.0002277521,0.0001058544,0.000009734065,0.00029135807,0.000114775496,0.00051279506],"category_scores_gemma":[0.65476763,0.00016748432,0.00030001818,0.0012183387,0.0003444609,0.0003034366,0.000079505706,0.00074429245,0.000086623695],"study_design_candidate":"systematic_review","study_design_consensus":null,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_system_candidate":false,"about_ca_system_consensus":false,"study_design_scores_codex":[0.0019798742,0.0006759519,0.0015710121,0.015163285,0.00012996743,0.000018437804,0.00008471678,0.00005953877,0.00005146924,0.0017711683,0.35133055,0.627164],"study_design_scores_gemma":[0.0044440054,0.0019412059,0.0021870416,0.81494045,0.0026091784,0.000019047273,0.0010594252,0.00604449,0.00011263254,0.00014557273,0.1663096,0.00018732181],"about_ca_topic_score_codex":0.00019146912,"about_ca_topic_score_gemma":0.0000641261,"teacher_disagreement_score":0.7997772,"about_ca_system_score_codex":0.00012040979,"about_ca_system_score_gemma":0.0047050915,"threshold_uncertainty_score":0.99773073},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"}]}