{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":326,"total_is_capped":false,"direct_labels_cover":325,"predictions_cover":326,"direct_label_status":"direct model label, unvalidated","prediction_status":"machine_predicted_unvalidated (Codex and Gemma teacher distillation)","score_status":"score_only:v0-immature-baseline (scores rank; they never assert a category)","snapshot":{"source":"OpenAlex, pinned release, all 482 partitions","release":"2026-06-24","frame_built":"2026-07-12","author_layer_release":"2026-06-26"},"query_hash":"25b05772a27d","filters":{"venue":"Evidence-Based Medicine"}},"results":[{"id":"W2603520596","doi":"10.1136/ebmed-2017-110668","title":"Rating the certainty in evidence in the absence of a single estimate of effect","year":2017,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":675,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Certainty; Grading (engineering); Meta-analysis; Computer science; Econometrics; Psychology; Actuarial science; Mathematics; Economics; Medicine; Engineering","authors":[{"name":"M. Hassan Murad","is_ca":false},{"name":"Reem A. Mustafa","is_ca":true},{"name":"Holger J. Schünemann","is_ca":true},{"name":"Shahnaz Sultan","is_ca":false},{"name":"Nancy Santesso","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7369926022121754,"gpt":0.6868045677958105,"spread":0.05018803441636499,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4318945,0.002995816,0.01263793,0.027402,0.002038905,0.01303516,0.00641451,0.01097642,0.006963977],"category_scores_gemma":[0.8051954,0.002792445,0.01899253,0.01138143,0.006631623,0.0141322,0.01203993,0.01054376,0.001686012],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005904675,"about_ca_system_score_gemma":0.008958176,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003611693,"about_ca_topic_score_gemma":0.004771421,"domain_scores_codex":[0.4091236,0.2345431,0.2805566,0.01204744,0.06165636,0.002072882],"domain_scores_gemma":[0.1845463,0.6737552,0.0675102,0.02144679,0.0493724,0.003369021],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.006023416,0.0002223394,0.02281801,0.364826,0.05988569,0.001348702,0.008896084,0.004998556,0.003042337,0.03855469,0.06712187,0.4222624],"study_design_scores_gemma":[0.00500846,0.00223389,0.01872773,0.3392673,0.06679119,0.005041557,0.007361478,0.01344047,0.006365508,0.2726915,0.2603091,0.002761838],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.03698652,0.426051,0.32303,0.1048259,0.02899355,0.02530569,0.01847712,0.001390276,0.03493989],"genre_scores_gemma":[0.3033439,0.10908,0.5122558,0.02278981,0.007059347,0.03526894,0.006537536,0.0004893165,0.003175389],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.5681055,"threshold_uncertainty_score":0.7005752,"prediction_status":"machine_predicted_unvalidated"},"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":"high"}],"label_agreement":"split"},{"id":"W2070999639","doi":"10.1136/ebm.11.6.184","title":"The GAD-7 scale was accurate for diagnosing generalised anxiety disorder","year":2006,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Anxiety, Depression, Psychometrics, Treatment, Cognitive Processes","field":"Psychology","cited_by":330,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Generalized anxiety disorder; Scale (ratio); Psychology; Clinical psychology; Anxiety; Psychiatry; Cartography; Geography","authors":[{"name":"Richard P. Swinson","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08237169929046008,"gpt":0.3742362222208535,"spread":0.2918645229303934,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002650767,0.0004966718,0.0009565732,0.0008800132,0.0003406696,0.0006152382,0.0003513726,0.0005885802,0.005931322],"category_scores_gemma":[0.0103185,0.000251401,0.0009717557,0.0007080589,0.0003337471,0.000650508,0.0006711566,0.001219013,0.002344842],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004635701,"about_ca_system_score_gemma":0.0007742458,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001674677,"about_ca_topic_score_gemma":0.00354744,"domain_scores_codex":[0.9975377,0.0009356143,0.0004398428,0.0002168842,0.0007243241,0.0001456136],"domain_scores_gemma":[0.9969343,0.000902831,0.0006154865,0.0002369057,0.001181571,0.0001289735],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004154696,0.001291965,0.4741895,0.002783522,0.001409631,0.001318562,0.002160911,0.00164515,0.009476504,0.003593317,0.1441189,0.3538573],"study_design_scores_gemma":[0.001255232,0.00211154,0.8774179,0.0008563019,0.0006092781,0.003746838,0.0009206398,0.002098492,0.002065683,0.002476163,0.1062843,0.0001577052],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"commentary","genre_scores_codex":[0.7612019,0.009386563,0.02581469,0.006335011,0.002436869,0.012506,0.03776636,0.001144962,0.1434075],"genre_scores_gemma":[0.9428837,0.002921942,0.02523224,0.001925871,0.0004357777,0.003750674,0.009752704,0.0001574567,0.01293955],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.005931322,"threshold_uncertainty_score":0.01984221,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","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":"medium"}],"label_agreement":"split"},{"id":"W2089009990","doi":"10.1136/ebm.13.6.162-a","title":"GRADE: assessing the quality of evidence for diagnostic recommendations","year":2008,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":231,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"European Commission","keywords":"Grading (engineering); Diagnostic test; Medicine; Clinical Practice; Test (biology); Odds; Diagnostic odds ratio; Medical physics; MEDLINE; Diagnostic accuracy; Quality (philosophy); Intensive care medicine; Family medicine; Pediatrics; Logistic regression; Radiology","authors":[{"name":"Holger J. Schünemann","is_ca":true},{"name":"Andrew D Oxman","is_ca":false},{"name":"Jan Brożek","is_ca":false},{"name":"Paul Glasziou","is_ca":false},{"name":"Patrick Bossuyt","is_ca":false},{"name":"Stephanie Chang","is_ca":false},{"name":"Paola Muti","is_ca":false},{"name":"Roman Jaeschke","is_ca":true},{"name":"Gordon Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9476189088113064,"gpt":0.6841482826036769,"spread":0.2634706262076295,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1357964,0.003367509,0.01300689,0.02811885,0.00215711,0.01085124,0.01302032,0.009470046,0.02361264],"category_scores_gemma":[0.594062,0.003482701,0.01258747,0.01670825,0.003637849,0.006062301,0.005275395,0.01460036,0.008077205],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01038476,"about_ca_system_score_gemma":0.02154461,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005403937,"about_ca_topic_score_gemma":0.004863356,"domain_scores_codex":[0.7774477,0.1062614,0.06587166,0.004354734,0.04451822,0.001546155],"domain_scores_gemma":[0.3369024,0.4320124,0.05691347,0.01277562,0.1561332,0.00526304],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004653661,0.00003882527,0.0006268854,0.1436167,0.004833749,0.0001062136,0.0002382461,0.0004452653,0.0001736361,0.002925018,0.7588227,0.08770732],"study_design_scores_gemma":[0.002191043,0.000239044,0.003103753,0.1896303,0.01425921,0.0004294839,0.0002630421,0.002683209,0.0006713969,0.0153423,0.7707797,0.0004074205],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"editorial","genre_scores_codex":[0.001285285,0.3124282,0.0504646,0.1934272,0.3384894,0.0396846,0.03493249,0.005529442,0.02375876],"genre_scores_gemma":[0.03032644,0.334843,0.3338626,0.068418,0.1008732,0.08623575,0.02408145,0.003434871,0.0179248],"genre_candidate":"editorial","genre_consensus":"editorial","teacher_disagreement_score":0.8642036,"threshold_uncertainty_score":0.7181685,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","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":"W4244401793","doi":"10.1136/ebm.11.6.162-a","title":"Of studies, syntheses, synopses, summaries, and systems: the \"5S\" evolution of information services for evidence-based healthcare decisions","year":2006,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Electronic Health Records Systems","field":"Health Professions","cited_by":151,"is_retracted":false,"has_abstract":false,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Computer science; Health care; Information system; Data science; Healthcare system; Real world evidence; Knowledge management; Information retrieval; Medicine; Engineering; Political science","authors":[{"name":"R. Brian Haynes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1694791498193305,"gpt":0.4373667516424011,"spread":0.2678876018230706,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4071284,0.001901042,0.004824973,0.04872732,0.004105199,0.03006351,0.005400242,0.00705425,0.005332006],"category_scores_gemma":[0.5360126,0.002387709,0.00374335,0.03815811,0.02725805,0.04326132,0.01660366,0.008154208,0.0008421057],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01757379,"about_ca_system_score_gemma":0.05707841,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006468547,"about_ca_topic_score_gemma":0.01249316,"domain_scores_codex":[0.5809658,0.32373,0.05909487,0.006518519,0.02816386,0.001526916],"domain_scores_gemma":[0.2604155,0.6137217,0.03427169,0.04848883,0.03750306,0.00559922],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003595044,0.0001827766,0.004452754,0.03675912,0.002345263,0.0001236015,0.01862263,0.001308022,0.0003836885,0.3834873,0.06296934,0.489006],"study_design_scores_gemma":[0.0004326413,0.0004646509,0.004584539,0.08196445,0.001834669,0.0004440824,0.01780709,0.003179084,0.0006991326,0.5636821,0.3245859,0.0003216986],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00624974,0.5296969,0.0688935,0.3569273,0.02085688,0.001036763,0.001211335,0.0006588736,0.01446857],"genre_scores_gemma":[0.1949267,0.3316668,0.3860761,0.05776161,0.01998901,0.004256941,0.001939691,0.0005687036,0.002814632],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5928717,"threshold_uncertainty_score":0.7311163,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["scholarly_communication"],"domain":null,"study_design":"not_applicable","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["scholarly_communication"],"domain":null,"study_design":"theoretical_or_conceptual","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2163264591","doi":"10.1136/ebm.11.1.2-a","title":"An emerging consensus on grading recommendations?","year":2006,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":141,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Grading (engineering); Excellence; Health care; Evidence-based medicine; Rating system; Medicine; Quality (philosophy); Quality management system; Evidence-based practice; Medical education; Quality management; Psychology; Operations management; Alternative medicine; Political science; Engineering; Management system","authors":[{"name":"Gordon Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4244408403965483,"gpt":0.5398832852316038,"spread":0.1154424448350555,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3002256,0.003715478,0.008214357,0.02860707,0.005373142,0.02825339,0.02166981,0.02806973,0.01259836],"category_scores_gemma":[0.6037473,0.003667158,0.008410289,0.0194342,0.01499153,0.03544541,0.01663427,0.04207612,0.01524051],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02213975,"about_ca_system_score_gemma":0.05748529,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01399556,"about_ca_topic_score_gemma":0.018823,"domain_scores_codex":[0.6652235,0.1345621,0.1101446,0.01628782,0.06917888,0.004603019],"domain_scores_gemma":[0.4256743,0.2627175,0.03966067,0.02764596,0.2263453,0.01795621],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001151184,0.00007587033,0.0006857023,0.01376346,0.0005969921,0.0001273798,0.001765349,0.0003315299,0.0001532063,0.05482189,0.5512421,0.3763215],"study_design_scores_gemma":[0.0002350678,0.0001110567,0.001181901,0.1495326,0.0005212407,0.0005756978,0.001854125,0.0007531113,0.0002304611,0.1389061,0.7058137,0.0002849086],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0004453384,0.1287557,0.02743252,0.7742262,0.05704403,0.001076096,0.001018412,0.0005988407,0.009402938],"genre_scores_gemma":[0.02308091,0.2962881,0.2562022,0.3615943,0.04346544,0.006294637,0.004952467,0.001216907,0.006905186],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.6997744,"threshold_uncertainty_score":0.8629464,"prediction_status":"machine_predicted_unvalidated"},"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":"W2461069230","doi":"10.1136/ebmed-2016-110447","title":"EBHC pyramid 5.0 for accessing preappraised evidence and guidance","year":2016,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":140,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"McMaster University","keywords":"Health informatics; Pyramid (geometry); Conceptual framework; Informatics; Computer science; Data science; Information system; Knowledge management; Medicine; Engineering; Sociology; Public health; Nursing; Mathematics","authors":[{"name":"Brian S. Alper","is_ca":false},{"name":"R. Brian Haynes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4781214244581729,"gpt":0.5704565639656052,"spread":0.09233513950743233,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.06490267,0.0007142519,0.001253091,0.006461147,0.004022303,0.008648772,0.004067574,0.04626732,0.03377992],"category_scores_gemma":[0.1835719,0.001307577,0.002902955,0.004591485,0.005789285,0.01385501,0.01170695,0.03512721,0.02344871],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0127427,"about_ca_system_score_gemma":0.0367064,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01395704,"about_ca_topic_score_gemma":0.02700189,"domain_scores_codex":[0.9141357,0.03863795,0.01482175,0.001960177,0.02617185,0.004272647],"domain_scores_gemma":[0.8143539,0.1008723,0.01099236,0.009450291,0.04697821,0.01735281],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003939937,0.00004605752,0.000546627,0.0003211991,0.0000157128,0.0007239733,0.0004342481,0.0002394745,0.0002336161,0.0336314,0.9252134,0.03855499],"study_design_scores_gemma":[0.00008408032,0.00006004464,0.001178595,0.001974216,0.00001621467,0.001145814,0.0002581337,0.0009845671,0.0001317762,0.03188738,0.9621847,0.00009463579],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003171531,0.001309165,0.005550812,0.9506025,0.002371115,0.0003645759,0.0004595123,0.0003818868,0.03864332],"genre_scores_gemma":[0.006730623,0.002465499,0.05421516,0.9074665,0.007117984,0.001128196,0.0007993749,0.0002227944,0.01985391],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.9350973,"threshold_uncertainty_score":0.3432422,"prediction_status":"machine_predicted_unvalidated"},"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":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2113321769","doi":"10.1136/eb-2013-101243","title":"The two-step Fagan's nomogram: ad hoc interpretation of a diagnostic test result without calculation","year":2013,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":136,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Prince Edward Island","funders":"","keywords":"Nomogram; Fagan inspection; Computer science; Test (biology); Medicine; Software engineering; Geology","authors":[{"name":"Charles Caraguel","is_ca":false},{"name":"Raphaël Vanderstichel","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1152710325688526,"gpt":0.3861066713332384,"spread":0.2708356387643858,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05881313,0.002507877,0.0040079,0.009951892,0.002152605,0.007085541,0.003103006,0.002322837,0.00537121],"category_scores_gemma":[0.1582203,0.001069661,0.004688522,0.004286156,0.003245012,0.004458294,0.004553264,0.004601576,0.002177442],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002642382,"about_ca_system_score_gemma":0.006619722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004084047,"about_ca_topic_score_gemma":0.005720122,"domain_scores_codex":[0.9612862,0.02344061,0.004656325,0.002820207,0.007215939,0.0005808303],"domain_scores_gemma":[0.9309035,0.04861327,0.004177265,0.004789312,0.0103368,0.001179884],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001331357,0.000138429,0.03856027,0.004304432,0.002528051,0.0008319115,0.001777812,0.02164517,0.001159854,0.05780335,0.07973793,0.7901815],"study_design_scores_gemma":[0.0008666254,0.002314342,0.03958431,0.007947457,0.003804449,0.01002192,0.002515501,0.2629789,0.006293549,0.2753058,0.3870088,0.001358276],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0131793,0.01661008,0.9359894,0.006405906,0.003434965,0.003404412,0.001362868,0.001363334,0.01824979],"genre_scores_gemma":[0.1731427,0.009080163,0.8047752,0.00172385,0.001950162,0.003113984,0.001181198,0.0006379249,0.004394765],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.05881313,"threshold_uncertainty_score":0.3110372,"prediction_status":"machine_predicted_unvalidated"},"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":[],"domain":null,"study_design":"design_other","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W1976868198","doi":"10.1136/ebm.6.3.71","title":"Using Evidence in Health and Social Care","year":2001,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Primary Care and Health Outcomes","field":"Health Professions","cited_by":68,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University Health Network; University of Toronto; Mount Sinai Hospital","funders":"","keywords":"Social care; Health care; Subject (documents); Psychology; Public relations; Nursing; Medicine; Political science; Computer science; Library science","authors":[{"name":"Howard Abrams","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.420018676887856,"gpt":0.5470936460772179,"spread":0.1270749691893619,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2131747,0.002037282,0.005232819,0.01230113,0.007477264,0.0459953,0.00594512,0.01604844,0.00645276],"category_scores_gemma":[0.2678107,0.001906386,0.002444902,0.009700457,0.06331333,0.04337584,0.02416853,0.02668321,0.002285096],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01675432,"about_ca_system_score_gemma":0.03755651,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004187228,"about_ca_topic_score_gemma":0.004051663,"domain_scores_codex":[0.5302466,0.3988402,0.01778834,0.007148003,0.04175615,0.004220645],"domain_scores_gemma":[0.5383947,0.4016332,0.01392808,0.02360204,0.01765128,0.004790716],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.00003367793,0.00004022331,0.0003097027,0.004220028,0.0002315601,0.000167853,0.0139613,0.0004140419,0.0001176735,0.8663885,0.05355952,0.06055599],"study_design_scores_gemma":[0.00003686581,0.00006239991,0.0002371409,0.02210975,0.00008370002,0.0001794914,0.006984528,0.0003249311,0.0002193405,0.6401533,0.3295417,0.00006700218],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.002084101,0.2290719,0.06203801,0.4888285,0.02361261,0.001042943,0.0002581044,0.0003708874,0.192693],"genre_scores_gemma":[0.255697,0.2697192,0.2366895,0.1836526,0.02364617,0.005540388,0.0004762459,0.0004842183,0.02409481],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7868253,"threshold_uncertainty_score":0.9702957,"prediction_status":"machine_predicted_unvalidated"},"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":"W2155843659","doi":"10.1136/ebm.6.5.133","title":"Creating knowledge management skills in primary care residents: a description of a new pathway to evidence-based practice in the community","year":2001,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":59,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto","funders":"","keywords":"Primary care; Knowledge management; Psychology; Medical education; Medicine; Family medicine; Computer science","authors":[{"name":"M. Blair Evans","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2730220787516308,"gpt":0.511557426295035,"spread":0.2385353475434042,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.009048806,0.0005320305,0.000475862,0.001513535,0.003751981,0.005508697,0.002938741,0.004582928,0.003547675],"category_scores_gemma":[0.008298294,0.0005118194,0.001120518,0.001279146,0.004135773,0.005231409,0.005917991,0.006182337,0.000872523],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003777028,"about_ca_system_score_gemma":0.01676203,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002916584,"about_ca_topic_score_gemma":0.01349594,"domain_scores_codex":[0.9933563,0.00463714,0.0003288695,0.0002832854,0.0008980898,0.0004963731],"domain_scores_gemma":[0.9946252,0.001950178,0.0003174615,0.0002756833,0.0005926529,0.002238993],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002434916,0.006051036,0.009558496,0.01039834,0.0001168201,0.02577478,0.1140421,0.001248703,0.003599344,0.06128775,0.08756917,0.6801099],"study_design_scores_gemma":[0.0006339592,0.004644758,0.02515693,0.01170413,0.0001629109,0.06107504,0.09336878,0.003229272,0.002364621,0.07011914,0.7271361,0.0004044386],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.2001347,0.06830513,0.0848976,0.5348385,0.008696814,0.008719319,0.0002614339,0.000549313,0.09359717],"genre_scores_gemma":[0.465942,0.06681116,0.364645,0.06291682,0.003981669,0.008651147,0.000272152,0.000264822,0.02651527],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.009048806,"threshold_uncertainty_score":0.0478552,"prediction_status":"machine_predicted_unvalidated"},"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":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2016126857","doi":"10.1136/ebm.6.1.4","title":"The rocky road: qualitative research as evidence","year":2001,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":53,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Geography","authors":[{"name":"Mita Giacomini","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9004402786253913,"gpt":0.8176259546885531,"spread":0.08281432393683819,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5955059,0.002712986,0.008944204,0.01457804,0.01128235,0.04292317,0.01291434,0.02052635,0.01367501],"category_scores_gemma":[0.659247,0.004207924,0.003039531,0.0136824,0.1057105,0.06566869,0.04146446,0.04036398,0.004649881],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02573374,"about_ca_system_score_gemma":0.04888067,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007797525,"about_ca_topic_score_gemma":0.00577528,"domain_scores_codex":[0.2073178,0.7293739,0.01810878,0.009476002,0.03325534,0.002468138],"domain_scores_gemma":[0.157801,0.7445291,0.01111418,0.05255635,0.02772633,0.006273187],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002632965,0.0001442614,0.000598029,0.009625304,0.0002453531,0.000293828,0.06073192,0.0009704289,0.0002668594,0.7857163,0.06092608,0.08021834],"study_design_scores_gemma":[0.0001996896,0.0001223771,0.0002079937,0.02536334,0.00005094544,0.0002034078,0.01485843,0.001565809,0.0001756962,0.8202479,0.1368608,0.000143506],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"editorial","genre_scores_codex":[0.002140376,0.09064019,0.2628627,0.5956908,0.01742092,0.004966429,0.00070694,0.0005806895,0.02499097],"genre_scores_gemma":[0.08639742,0.05325105,0.6473652,0.1718141,0.01141013,0.0238187,0.0005397939,0.000839776,0.004563772],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.4044941,"threshold_uncertainty_score":0.4988133,"prediction_status":"machine_predicted_unvalidated"},"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":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W1971758413","doi":"10.1136/ebm.12.1.2-a","title":"Misunderstandings, misperceptions, and mistakes","year":2007,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":45,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; University of Calgary","funders":"","keywords":"Medicine; Family medicine","authors":[{"name":"S. Straus","is_ca":true},{"name":"Brian Haynes","is_ca":true},{"name":"Paul Glasziou","is_ca":true},{"name":"K. Dickersin","is_ca":false},{"name":"G. Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9078811001668722,"gpt":0.5696515659896233,"spread":0.3382295341772489,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1630977,0.002217223,0.002589079,0.005236727,0.01200187,0.01597374,0.008753378,0.02189465,0.002880055],"category_scores_gemma":[0.4071047,0.001826407,0.001837034,0.00334383,0.06754469,0.03383572,0.01589806,0.05589845,0.002896152],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01311265,"about_ca_system_score_gemma":0.01643538,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006885724,"about_ca_topic_score_gemma":0.007471799,"domain_scores_codex":[0.7443317,0.1605998,0.01890043,0.01448624,0.05678811,0.004893807],"domain_scores_gemma":[0.4914089,0.3739794,0.033484,0.01990599,0.06894942,0.01227234],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001733634,0.00007598145,0.001901567,0.002167723,0.0002304153,0.002517526,0.2170189,0.0003636364,0.0006109502,0.08696859,0.6132739,0.07469744],"study_design_scores_gemma":[0.00006446105,0.00009982209,0.0007821713,0.01055189,0.00009825904,0.004233333,0.1528347,0.0008834942,0.001245174,0.1911234,0.6376916,0.0003918647],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.00342495,0.01610717,0.004299935,0.9376171,0.03472071,0.00003248461,0.000044872,0.0001250638,0.003627748],"genre_scores_gemma":[0.09105739,0.02084277,0.008766889,0.845493,0.02873593,0.0001524246,0.00008081934,0.0004785967,0.004392343],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.8369023,"threshold_uncertainty_score":0.8625535,"prediction_status":"machine_predicted_unvalidated"},"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":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2738079722","doi":"10.1136/ebmed-2017-110714","title":"Mental health literacy in primary care: Canadian Research and Education for the Advancement of Child Health (CanREACH)","year":2017,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Child and Adolescent Health","field":"Health Professions","cited_by":42,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary; Alberta Health Services","funders":"","keywords":"Referral; Medicine; Mental health; Family medicine; Primary care; Psychiatry","authors":[{"name":"Eden McCaffrey","is_ca":true},{"name":"Samuel Chang","is_ca":true},{"name":"Geraldine Farrelly","is_ca":true},{"name":"Abdul Rahman","is_ca":true},{"name":"David Cawthorpe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1578522742340356,"gpt":0.5266581172409056,"spread":0.36880584300687,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01090279,0.0003848174,0.0005905021,0.003735731,0.002529063,0.00239548,0.001375957,0.0006858007,0.004735863],"category_scores_gemma":[0.02756859,0.0003093111,0.0006507015,0.006073923,0.001248515,0.0009702899,0.001431759,0.001258533,0.0002576006],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.04875384,"about_ca_system_score_gemma":0.1212096,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9823303,"about_ca_topic_score_gemma":0.9841591,"domain_scores_codex":[0.99433,0.001268016,0.0002802715,0.0003628598,0.002872325,0.0008865612],"domain_scores_gemma":[0.9794123,0.005676828,0.001561267,0.0004644569,0.009110116,0.003775124],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003303685,0.001088501,0.4129653,0.00664,0.0002440716,0.0002001746,0.009337986,0.0004486681,0.0002950221,0.004874737,0.03545742,0.5281178],"study_design_scores_gemma":[0.00007764841,0.0002160369,0.9719642,0.002296393,0.0001386761,0.0000511938,0.002727479,0.0003392483,0.0001283073,0.0003764859,0.02165723,0.00002713018],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"other","genre_scores_codex":[0.6253303,0.1756056,0.002647381,0.06863871,0.001270765,0.003729031,0.0166241,0.00020384,0.1059503],"genre_scores_gemma":[0.9484838,0.03426379,0.005128968,0.003927595,0.0001944183,0.0008926854,0.002493027,0.00002685994,0.004588812],"genre_candidate":"other","genre_consensus":null,"teacher_disagreement_score":0.04875384,"threshold_uncertainty_score":0.3537356,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":true,"confidence":"high"},{"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":"W2079429547","doi":"10.1136/ebm.14.4.101","title":"What kind of randomised trials do patients and clinicians need?","year":2009,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Cerebrovascular and Carotid Artery Diseases","field":"Medicine","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Health Sciences Centre; Sunnybrook Health Science Centre","funders":"","keywords":"Medicine; Intensive care medicine","authors":[{"name":"Merrick Zwarenstein","is_ca":true},{"name":"Shaun Treweek","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.06865936897527998,"gpt":0.355519507897034,"spread":0.286860138921754,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4754732,0.003456096,0.02401308,0.01388005,0.005865244,0.03314834,0.01339053,0.05725291,0.02080212],"category_scores_gemma":[0.7547393,0.004299353,0.007463639,0.0106635,0.04294351,0.08573802,0.0127703,0.03380192,0.01566107],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01428534,"about_ca_system_score_gemma":0.03073568,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005049612,"about_ca_topic_score_gemma":0.003139408,"domain_scores_codex":[0.4802752,0.3921805,0.05281909,0.01606434,0.05460414,0.004056677],"domain_scores_gemma":[0.1818348,0.6753157,0.0313873,0.04316537,0.05262148,0.01567529],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"qualitative","study_design_scores_codex":[0.003100477,0.0006872292,0.002667257,0.02973124,0.002176818,0.0006172544,0.004677082,0.001287023,0.0004627868,0.1948718,0.4305258,0.3291952],"study_design_scores_gemma":[0.004942366,0.0005142348,0.0006896796,0.03684913,0.0005628096,0.0008313065,0.003037259,0.0009934407,0.0002301828,0.577868,0.3730272,0.0004542518],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.0004151292,0.1717233,0.01737773,0.7870576,0.01725659,0.001228091,0.0003033259,0.0002636374,0.00437465],"genre_scores_gemma":[0.01443257,0.1001518,0.1215947,0.7184206,0.03367225,0.008842362,0.0005443573,0.0004999064,0.001841628],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.5245268,"threshold_uncertainty_score":0.646835,"prediction_status":"machine_predicted_unvalidated"},"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":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W1968960679","doi":"10.1136/ebm.6.4.104","title":"Evidence-based practice: a primer for health care professionals","year":2001,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":39,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Scope (computer science); Health professionals; Health care; Evidence-based practice; Primer (cosmetics); Nursing; Nursing practice; Library science; Engineering ethics; Medical education; Psychology; Sociology; Medicine; Alternative medicine; Political science; Engineering; Computer science; Law","authors":[{"name":"J W Sellors","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3837056646440681,"gpt":0.6139096012606841,"spread":0.2302039366166159,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.05123127,0.002874698,0.00433816,0.00633711,0.004395979,0.02541614,0.007351705,0.02115158,0.01265475],"category_scores_gemma":[0.07616612,0.003131973,0.002120183,0.007259839,0.01820075,0.03518841,0.01437324,0.05402165,0.01302238],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008346877,"about_ca_system_score_gemma":0.02082192,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002322611,"about_ca_topic_score_gemma":0.003008784,"domain_scores_codex":[0.9488343,0.0334115,0.005119953,0.001944684,0.009410113,0.001279527],"domain_scores_gemma":[0.8955854,0.08550204,0.003581116,0.002175828,0.008631993,0.004523641],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0000540594,0.0001714242,0.000149019,0.005048986,0.00005870578,0.0004552892,0.01761604,0.0004860292,0.0004583592,0.1436937,0.676512,0.1552963],"study_design_scores_gemma":[0.00002163699,0.00005634142,0.00009802578,0.009736397,0.0000090046,0.0005434046,0.00264357,0.0001696084,0.00005251548,0.09680802,0.8898222,0.00003919568],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"review","genre_scores_codex":[0.0003074368,0.2857302,0.0371643,0.6126568,0.03035397,0.0007553787,0.0002227332,0.0004971897,0.03231198],"genre_scores_gemma":[0.009654582,0.3630197,0.1727496,0.3645077,0.04072736,0.005343922,0.0005872913,0.0010268,0.04238311],"genre_candidate":"review","genre_consensus":null,"teacher_disagreement_score":0.05123127,"threshold_uncertainty_score":0.2709401,"prediction_status":"machine_predicted_unvalidated"},"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":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W2756094298","doi":"10.1136/ebmed-2017-110798","title":"What does expert opinion in guidelines mean? a meta-epidemiological study","year":2017,"lang":"en","type":"review","venue":"Evidence-Based Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":38,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Impact","funders":"","keywords":"Guideline; Observational study; CLARITY; Medicine; Expert opinion; Evidence-based medicine; Evidence-based practice; Family medicine; MEDLINE; Alternative medicine; Pathology; Intensive care medicine","authors":[{"name":"Oscar J. Ponce","is_ca":false},{"name":"Neri Alejandro Álvarez‐Villalobos","is_ca":false},{"name":"Raj Shah","is_ca":false},{"name":"Mohammed Khaled","is_ca":false},{"name":"Rebecca L. Morgan","is_ca":true},{"name":"Shahnaz Sultan","is_ca":false},{"name":"Yngve Falck–Ytter","is_ca":false},{"name":"Larry J. Prokop","is_ca":false},{"name":"Philipp Dahm","is_ca":false},{"name":"Reem A. Mustafa","is_ca":false},{"name":"M. Hassan Murad","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.934234501038434,"gpt":0.7009597189922305,"spread":0.2332747820462034,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch","metaepi_broad"],"consensus_categories":[],"category_scores_codex":[0.08695167,0.001664276,0.01042071,0.0141925,0.001055499,0.007593118,0.003315478,0.004793522,0.007043593],"category_scores_gemma":[0.3649171,0.00158146,0.01537464,0.01678502,0.001972853,0.007856677,0.002507731,0.003912268,0.0006155501],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005780005,"about_ca_system_score_gemma":0.006240275,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005941249,"about_ca_topic_score_gemma":0.009110675,"domain_scores_codex":[0.8551418,0.09441846,0.02828054,0.007598736,0.01338903,0.001171465],"domain_scores_gemma":[0.4589089,0.4699189,0.04572413,0.006695644,0.01737259,0.00137988],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.00213779,0.0001146153,0.03382529,0.5740592,0.2283228,0.0003456453,0.001694136,0.0006151187,0.0001341601,0.003158265,0.006842752,0.1487501],"study_design_scores_gemma":[0.001557649,0.000609621,0.03620443,0.5032052,0.4210836,0.0008123619,0.001190456,0.001028627,0.0002531941,0.007994121,0.02590259,0.0001581586],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.002505329,0.9931096,0.0009176985,0.002017992,0.0003844374,0.0001651412,0.0002442048,0.000008022264,0.0006476319],"genre_scores_gemma":[0.1660007,0.8218055,0.004130989,0.004545182,0.001581101,0.0009429006,0.0006865091,0.0000496539,0.0002574737],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.9895793,"threshold_uncertainty_score":0.4598498,"prediction_status":"machine_predicted_unvalidated"},"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":"low"},{"model":"gpt","categories":["metaresearch","metaepi_narrow","metaepi_broad"],"domain":"methods","study_design":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2157007996","doi":"10.1136/ebmed-2014-110158","title":"Mixed kinds of evidence: synthesis designs and critical appraisal for systematic mixed studies reviews including qualitative, quantitative and mixed methods studies","year":2015,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":34,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University","funders":"","keywords":"Multimethodology; Critical appraisal; Systematic review; Management science; Context (archaeology); Qualitative research; Computer science; Qualitative property; Psychological intervention; Psychology; Data science; MEDLINE; Medicine; Sociology; Engineering; Alternative medicine; Social science; Mathematics education; Nursing; Chemistry; Geography","authors":[{"name":"Pierre Pluye","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9713863695536488,"gpt":0.7222219156480846,"spread":0.2491644539055642,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5675545,0.002675164,0.006740576,0.0139562,0.005499168,0.01657089,0.006560271,0.02802541,0.01023649],"category_scores_gemma":[0.8200393,0.004759293,0.007162087,0.00921957,0.01599434,0.01945929,0.01562314,0.03888759,0.008191872],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01389948,"about_ca_system_score_gemma":0.04142522,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001690427,"about_ca_topic_score_gemma":0.003186274,"domain_scores_codex":[0.2326776,0.6172043,0.09098067,0.007804633,0.05000167,0.001331083],"domain_scores_gemma":[0.0800617,0.7910931,0.03257395,0.02623112,0.06736674,0.002673392],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0007088269,0.00008350987,0.0007777,0.03780511,0.001419955,0.0005797027,0.005335078,0.000550646,0.0004718544,0.09464192,0.5623097,0.295316],"study_design_scores_gemma":[0.001177335,0.0003768563,0.0008976735,0.09901911,0.0009301693,0.001393849,0.001450391,0.003747224,0.0009868699,0.3612727,0.5281703,0.0005774776],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"methods","genre_scores_codex":[0.0003314053,0.05192485,0.1113314,0.7856946,0.03979014,0.004069594,0.0004848629,0.0003882495,0.005984872],"genre_scores_gemma":[0.009959321,0.04335647,0.473683,0.3861478,0.04795347,0.03580199,0.0003154594,0.0004908938,0.00229159],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.4324455,"threshold_uncertainty_score":0.5332823,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"methods","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":"medium"}],"label_agreement":"split"},{"id":"W2124587829","doi":"10.1136/eb-2012-100872","title":"Tranexamic acid reduces blood transfusion in surgical patients while its effects on thromboembolic events and mortality are uncertain","year":2012,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Blood transfusion and management","field":"Medicine","cited_by":32,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Toronto Western Hospital; University of Toronto","funders":"","keywords":"Tranexamic acid; Medicine; Antifibrinolytic; Fibrinolysis; Perioperative; Blood transfusion; Anesthesia; Surgery; Internal medicine; Blood loss","authors":[{"name":"Raviraj Raveendran","is_ca":true},{"name":"Jean Wong","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.05930321580026039,"gpt":0.30653109437467,"spread":0.2472278785744096,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004640526,0.001437255,0.005037392,0.002673525,0.001138861,0.002416944,0.003572201,0.01192318,0.04544415],"category_scores_gemma":[0.07391234,0.0006854734,0.003745292,0.002904341,0.001015346,0.002645301,0.0007887656,0.008363882,0.01283224],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003218232,"about_ca_system_score_gemma":0.005583909,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02170474,"about_ca_topic_score_gemma":0.02790714,"domain_scores_codex":[0.9954031,0.001522379,0.0006327697,0.0004443923,0.001755773,0.0002415538],"domain_scores_gemma":[0.9652803,0.01877635,0.003081257,0.0004073045,0.01140754,0.001047318],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0004117567,0.00001236441,0.0001197437,0.01181965,0.0005048597,0.00007731906,0.00004885883,0.00005027828,0.00007576423,0.0006135788,0.9683974,0.01786851],"study_design_scores_gemma":[0.001566004,0.000348307,0.003366385,0.07432184,0.003663143,0.0002800336,0.0002288839,0.0003299932,0.0003647415,0.00456509,0.910804,0.0001615972],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0005275109,0.2519339,0.0004047602,0.4879152,0.2342752,0.0002712259,0.006115322,0.0001816424,0.01837525],"genre_scores_gemma":[0.01462888,0.2343724,0.0008477143,0.4765828,0.235265,0.0005532261,0.002608001,0.000201954,0.03494005],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.04544415,"threshold_uncertainty_score":0.1520259,"prediction_status":"machine_predicted_unvalidated"},"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":"W2138654288","doi":"10.1136/ebm.7.1.4","title":"\"Double blind, you are the weakest link -- goodbye!\"","year":2002,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":26,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Blinding; Double blind; Masking (illustration); Medicine; Randomized controlled trial; Psychology; Alternative medicine; Placebo; Pathology","authors":[{"name":"P.J. Devereaux","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7776670886860587,"gpt":0.5311319646305034,"spread":0.2465351240555553,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.03588162,0.004578772,0.005990166,0.004366788,0.003889905,0.01016523,0.006553921,0.01459949,0.01035479],"category_scores_gemma":[0.1456436,0.001777302,0.003461434,0.003474189,0.007434236,0.00756035,0.002557228,0.03149975,0.01450852],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003447403,"about_ca_system_score_gemma":0.006298121,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001767495,"about_ca_topic_score_gemma":0.003093185,"domain_scores_codex":[0.9654197,0.01351898,0.005043237,0.002201692,0.01317987,0.0006364752],"domain_scores_gemma":[0.8710389,0.06987441,0.00813702,0.004793531,0.04128546,0.004870667],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00007090349,0.000009358766,0.0000176571,0.0009925663,0.00004959908,0.00004058236,0.00002858066,0.00001847599,0.0000414549,0.0008055905,0.9880366,0.009888557],"study_design_scores_gemma":[0.0001432655,0.00004018427,0.0001102983,0.001799803,0.0001101242,0.0002062931,0.00004299425,0.0001750262,0.0001581992,0.003167284,0.9940148,0.0000317774],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"editorial","genre_scores_codex":[0.00002041126,0.0101759,0.0007277455,0.03036957,0.9574085,0.00006154176,0.0000660941,0.0001079207,0.001062241],"genre_scores_gemma":[0.0008069591,0.01718354,0.001775815,0.07213566,0.8963054,0.0001939388,0.000081413,0.0001603197,0.01135697],"genre_candidate":"editorial","genre_consensus":"editorial","teacher_disagreement_score":0.9641184,"threshold_uncertainty_score":0.1897624,"prediction_status":"machine_predicted_unvalidated"},"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":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2108816030","doi":"10.1136/ebm.10.4.101","title":"Finding the gold in Medline: clinical queries","year":2005,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":25,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"MEDLINE; Computer science; Treasure; Search engine indexing; Information retrieval; World Wide Web; Data science; History; Political science","authors":[{"name":"R. Brian Haynes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07816964072642624,"gpt":0.4094760960115682,"spread":0.331306455285142,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.01548395,0.002431229,0.003078986,0.01135979,0.002356,0.008707066,0.002906923,0.009206734,0.01354689],"category_scores_gemma":[0.07179484,0.0012831,0.00207967,0.00613323,0.002373776,0.006209508,0.002050155,0.01236052,0.008187511],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005438401,"about_ca_system_score_gemma":0.006471211,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003452844,"about_ca_topic_score_gemma":0.009099022,"domain_scores_codex":[0.9916328,0.00257616,0.002228378,0.0006764298,0.002709877,0.0001762277],"domain_scores_gemma":[0.9441994,0.0361426,0.002366657,0.0008211144,0.01345248,0.003017735],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00002123914,0.00000633613,0.00002195483,0.0004918459,0.00001931387,0.00007105722,0.00003815444,0.00002339898,0.00004430984,0.0002454144,0.9913117,0.007705344],"study_design_scores_gemma":[0.00007015897,0.00002774162,0.0002580161,0.002058989,0.0001025816,0.0003118445,0.0001323601,0.0001985483,0.0001421578,0.001255415,0.995411,0.00003129452],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"editorial","genre_scores_codex":[0.0001761077,0.03380458,0.001513861,0.2348817,0.7255152,0.0001553066,0.0004650211,0.0003762589,0.003112077],"genre_scores_gemma":[0.002477668,0.06003497,0.003330465,0.1702416,0.7433624,0.0002723669,0.0006411071,0.0003757677,0.01926363],"genre_candidate":"editorial","genre_consensus":"editorial","teacher_disagreement_score":0.984516,"threshold_uncertainty_score":0.0818879,"prediction_status":"machine_predicted_unvalidated"},"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":"W2043409515","doi":"10.1136/ebm.13.4.98-a","title":"Knowledge-to-action cycle","year":2008,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Intensive Care Unit Cognitive Disorders","field":"Medicine","cited_by":24,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary","funders":"","keywords":"Delirium; Knowledge translation; Action (physics); Psychological intervention; Medicine; Intensive care medicine; Hip fracture; Cognition; Nursing; Psychology; Psychiatry; Knowledge management; Osteoporosis","authors":[{"name":"Sharon E. Straus","is_ca":true},{"name":"Jayna Holroyd‐Leduc","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.125523433825342,"gpt":0.3721375491546728,"spread":0.2466141153293308,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03199666,0.002001475,0.001493674,0.00406246,0.005238327,0.01043954,0.005309812,0.007341329,0.01873826],"category_scores_gemma":[0.04304758,0.001177031,0.002105376,0.002413244,0.01439254,0.01058384,0.01865278,0.007304786,0.0051205],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01606912,"about_ca_system_score_gemma":0.05537202,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00703227,"about_ca_topic_score_gemma":0.004363654,"domain_scores_codex":[0.956544,0.02723549,0.002224077,0.004053214,0.007728254,0.002215104],"domain_scores_gemma":[0.9560251,0.02743417,0.002310992,0.002760707,0.005823753,0.005645224],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003129487,0.001397784,0.006409802,0.007199361,0.0002430686,0.001657386,0.06289753,0.003750912,0.001626565,0.4862882,0.03548223,0.3927343],"study_design_scores_gemma":[0.0002883553,0.0005358896,0.004463888,0.006132458,0.0001166889,0.0008935093,0.02184942,0.003840521,0.001108012,0.6204918,0.3401485,0.0001308518],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"other","genre_gemma":"empirical","genre_scores_codex":[0.04806658,0.01253864,0.3017941,0.1547352,0.001976962,0.01517415,0.001040545,0.001354255,0.4633195],"genre_scores_gemma":[0.5585964,0.01119375,0.367498,0.0250034,0.0006862355,0.01063135,0.001333534,0.0003844475,0.02467281],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03199666,"threshold_uncertainty_score":0.1692165,"prediction_status":"machine_predicted_unvalidated"},"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":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W1992388321","doi":"10.1136/eb-2013-101634","title":"Suggestions for improving guideline utility and trustworthiness","year":2013,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":23,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Guideline; Rigour; Medicine; Audit; Evidence-based medicine; Documentation; Medical education; Public relations; Alternative medicine; Political science; Business; Accounting; Computer science; Pathology","authors":[{"name":"Ian Scott","is_ca":false},{"name":"Gordon Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2840592520046301,"gpt":0.4815833579619859,"spread":0.1975241059573558,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6087804,0.002563736,0.005673614,0.02080486,0.008444561,0.03814064,0.01410387,0.03351821,0.01611547],"category_scores_gemma":[0.8887044,0.004325811,0.007772211,0.01426972,0.02639292,0.0573596,0.03063863,0.03918403,0.005947975],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.02538317,"about_ca_system_score_gemma":0.1015338,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01432191,"about_ca_topic_score_gemma":0.01963171,"domain_scores_codex":[0.2998411,0.5153965,0.08346891,0.01492608,0.07854607,0.007821379],"domain_scores_gemma":[0.05901165,0.6665322,0.02890525,0.04787295,0.1841946,0.01348328],"domain_codex":"methods","domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0003502423,0.0004437263,0.01002107,0.009152151,0.0005949293,0.0006646081,0.01866212,0.001711926,0.0003978929,0.1004885,0.4585237,0.3989891],"study_design_scores_gemma":[0.001396591,0.0006473419,0.01012269,0.09265643,0.000866532,0.001118769,0.02129284,0.007261693,0.001386017,0.4126205,0.4496588,0.0009716909],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.001394238,0.01169133,0.02699608,0.9402288,0.008947467,0.001583304,0.0001754037,0.0004840891,0.008499217],"genre_scores_gemma":[0.1015483,0.02598888,0.44114,0.3960633,0.01673409,0.01185376,0.0008187794,0.0009978571,0.004855034],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.3912196,"threshold_uncertainty_score":0.4824434,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"},{"model":"gpt","categories":["metaresearch"],"domain":"evaluation","study_design":"theoretical_or_conceptual","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2138571613","doi":"10.1136/ebmed-2014-110086","title":"Mendelian randomisation meta-analysis sheds doubt on protective associations between ‘moderate’ alcohol consumption and coronary heart disease","year":2014,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Alcohol Consumption and Health Effects","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Victoria","funders":"Economic and Social Research Council; Medical Research Council","keywords":"Confounding; Observational study; Medicine; Mendelian randomization; Internal medicine; Alcohol consumption; Meta-analysis; Genotype; Alcohol; Genetics; Biology; Gene; Genetic variants","authors":[{"name":"Tanya Chikritzhs","is_ca":false},{"name":"Timothy S. Naimi","is_ca":false},{"name":"Tim Stockwell","is_ca":true},{"name":"Wenbin Liang","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3952768025304359,"gpt":0.4326952756226826,"spread":0.03741847309224666,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.04918437,0.002482351,0.008913763,0.003790831,0.002305906,0.004929003,0.009506412,0.02215185,0.01556566],"category_scores_gemma":[0.2502809,0.001676729,0.007661635,0.003996765,0.004489333,0.005368544,0.002168663,0.02925559,0.00727017],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006864942,"about_ca_system_score_gemma":0.009178426,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01973222,"about_ca_topic_score_gemma":0.02174788,"domain_scores_codex":[0.9586759,0.02437515,0.003729534,0.002913585,0.009499262,0.0008066491],"domain_scores_gemma":[0.8043711,0.1583171,0.006374389,0.004229918,0.02461399,0.002093408],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.0005232574,0.00001014301,0.000198677,0.005451645,0.001700038,0.0001567,0.0001348538,0.0001540775,0.00009459139,0.002205945,0.9753377,0.01403236],"study_design_scores_gemma":[0.004619074,0.0004313032,0.003143592,0.05548806,0.01306087,0.0007230669,0.0004483837,0.002029386,0.0008353905,0.04897751,0.8696524,0.0005908876],"study_design_candidate":"meta_analysis","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.0003666645,0.1179448,0.001964402,0.7094614,0.1649519,0.0001979222,0.001987993,0.0003054491,0.002819497],"genre_scores_gemma":[0.01384117,0.0757358,0.002559889,0.6920682,0.2054134,0.0007655605,0.0008676112,0.0005162687,0.008232075],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.04918437,"threshold_uncertainty_score":0.2601149,"prediction_status":"machine_predicted_unvalidated"},"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":[],"domain":null,"study_design":"meta_analysis","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W1991116672","doi":"10.1136/ebm.13.4.101","title":"Evidence-based medicine targets the individual patient, part 1: how clinicians can use study results to determine optimal individual care","year":2008,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Patient Satisfaction in Healthcare","field":"Health Professions","cited_by":19,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; McMaster University","funders":"","keywords":"Acknowledgement; Medicine; Acute otitis media; Randomized controlled trial; Otitis; Evidence-based medicine; Socioeconomic status; Alternative medicine; Intensive care medicine; Family medicine; Surgery","authors":[{"name":"Dirk Bassler","is_ca":false},{"name":"Jason W. Busse","is_ca":true},{"name":"P. J Karanicolas","is_ca":true},{"name":"G. H. Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4990194589023669,"gpt":0.4604585242727393,"spread":0.03856093462962756,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3523085,0.002498751,0.007753178,0.008017236,0.003078488,0.01551166,0.006654092,0.02889833,0.00326678],"category_scores_gemma":[0.6427767,0.002378757,0.00318901,0.004047995,0.01859445,0.02239529,0.005397185,0.02020282,0.002555808],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006515933,"about_ca_system_score_gemma":0.01786788,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002835311,"about_ca_topic_score_gemma":0.00553136,"domain_scores_codex":[0.6025417,0.3245002,0.03689326,0.004676197,0.02994504,0.001443686],"domain_scores_gemma":[0.2154787,0.7174823,0.01960899,0.01297763,0.03164073,0.002811606],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0007691956,0.0001983583,0.00532731,0.04675556,0.002836704,0.001801925,0.006692527,0.002215869,0.001071954,0.07916521,0.4926226,0.3605427],"study_design_scores_gemma":[0.0008247466,0.0005341158,0.005144211,0.1115167,0.001499374,0.002938722,0.004958383,0.004337267,0.001455284,0.4701159,0.3959319,0.0007433508],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.0008896173,0.12198,0.03021736,0.8212181,0.01984804,0.0007085383,0.0001307868,0.000152255,0.004855298],"genre_scores_gemma":[0.06535168,0.1612124,0.1577468,0.5534112,0.0561006,0.003230532,0.0001736126,0.0002810247,0.002492044],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.6476915,"threshold_uncertainty_score":0.798719,"prediction_status":"machine_predicted_unvalidated"},"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":"other","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2053786830","doi":"10.1136/ebm.13.5.130-a","title":"Evidence-based medicine targets the individual patient, part 2: guides and tools for individual decision-making","year":2008,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; McMaster University","funders":"","keywords":"Clinical decision making; Randomized controlled trial; Evidence-based medicine; Medicine; Adverse effect; Patient care; MEDLINE; Alternative medicine; Intensive care medicine; Psychology; Nursing; Surgery; Internal medicine","authors":[{"name":"Dirk Bassler","is_ca":false},{"name":"Jason W. Busse","is_ca":true},{"name":"P. J Karanicolas","is_ca":true},{"name":"G. H. Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5598851550812675,"gpt":0.4876308129443874,"spread":0.07225434213688009,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.1860677,0.002142934,0.005546489,0.006134201,0.00264153,0.01487166,0.006338426,0.02740672,0.005623056],"category_scores_gemma":[0.3500162,0.002459901,0.002107478,0.004549132,0.0199347,0.01876926,0.006154666,0.03040097,0.007409271],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006337243,"about_ca_system_score_gemma":0.01869604,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003507378,"about_ca_topic_score_gemma":0.005296926,"domain_scores_codex":[0.8062478,0.1403278,0.02420631,0.002946631,0.02486787,0.001403623],"domain_scores_gemma":[0.5605525,0.3497164,0.01749548,0.01593906,0.04945724,0.006839179],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0001123868,0.0001290388,0.0005706243,0.006800318,0.0002483418,0.0004530603,0.004655817,0.000930873,0.0004345105,0.09245913,0.7393822,0.1538238],"study_design_scores_gemma":[0.0001712667,0.0001106199,0.0009850505,0.01620284,0.0001471756,0.001059066,0.001994015,0.001799138,0.0004218885,0.3137716,0.6630915,0.0002457497],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003607164,0.03721489,0.03372281,0.9079718,0.01262855,0.0003072097,0.0001585682,0.0002277079,0.00740766],"genre_scores_gemma":[0.02217465,0.05719106,0.3747937,0.5049651,0.02906227,0.002579472,0.0003501881,0.0003689381,0.008514685],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.1860677,"threshold_uncertainty_score":0.984032,"prediction_status":"machine_predicted_unvalidated"},"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":"high"}],"label_agreement":"split"},{"id":"W2123780592","doi":"10.1136/ebm.11.3.73","title":"Watchful waiting was as safe as surgical repair for minimally symptomatic inguinal hernias","year":2006,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Hernia repair and management","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Watchful waiting; Medicine; Inguinal hernia; Surgery; General surgery; Hernia","authors":[{"name":"P J O’Dwyer","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04726706895209426,"gpt":0.3271762909705744,"spread":0.2799092220184802,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0015175,0.0003520434,0.00180067,0.0003054217,0.0004515062,0.0009614043,0.0003368693,0.001224901,0.008508435],"category_scores_gemma":[0.006922137,0.0002711213,0.002258644,0.0002669201,0.0008255749,0.001236918,0.0004085193,0.001681529,0.0002941265],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003193966,"about_ca_system_score_gemma":0.0008841593,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001334431,"about_ca_topic_score_gemma":0.001830037,"domain_scores_codex":[0.9993642,0.0003053376,0.0001018205,0.0000989052,0.00005845182,0.0000711722],"domain_scores_gemma":[0.9967334,0.002033932,0.0005998389,0.0001218492,0.0000535931,0.0004573753],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"nonrandomized_trial","study_design_scores_codex":[0.8405898,0.002767562,0.006405618,0.008735304,0.002886711,0.0001382586,0.0003495573,0.0005616566,0.001457086,0.0009969277,0.005170348,0.1299411],"study_design_scores_gemma":[0.6318067,0.2757497,0.04553861,0.008528425,0.01512597,0.0004864874,0.0008824421,0.002844555,0.002831877,0.005175092,0.01083297,0.0001973297],"study_design_candidate":"nonrandomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"commentary","genre_scores_codex":[0.9255379,0.05453724,0.001543672,0.006659288,0.003378898,0.001158387,0.001388479,0.0001966331,0.005599341],"genre_scores_gemma":[0.9875953,0.007191132,0.0005691538,0.001812979,0.001223501,0.0004546879,0.0003738198,0.00002138462,0.0007580207],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.008508435,"threshold_uncertainty_score":0.0284636,"prediction_status":"machine_predicted_unvalidated"},"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":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2049989275","doi":"10.1136/ebm.14.4.100","title":"The devil is in the details...or not? A primer on individual patient data meta-analysis","year":2009,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; University of Toronto","funders":"","keywords":"Primer (cosmetics); Meta-analysis; Psychology; Computer science; Medicine; Internal medicine; Chemistry","authors":[{"name":"S. C. Sud","is_ca":true},{"name":"James D. Douketis","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9355634960450253,"gpt":0.5838908827533396,"spread":0.3516726132916858,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.09828657,0.001829513,0.006463337,0.008260129,0.001074089,0.00785541,0.005324364,0.008483759,0.008646434],"category_scores_gemma":[0.2029404,0.001767183,0.006225267,0.008195495,0.00552236,0.014371,0.003832272,0.01879503,0.004502321],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00308941,"about_ca_system_score_gemma":0.006764818,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002823891,"about_ca_topic_score_gemma":0.002731123,"domain_scores_codex":[0.9222555,0.06252085,0.006305486,0.002692288,0.005861937,0.0003639444],"domain_scores_gemma":[0.8501197,0.1271531,0.00406159,0.008289802,0.009431887,0.0009438654],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003288365,0.00008050589,0.001078071,0.02488462,0.002760284,0.0003850175,0.001106079,0.001535087,0.0003591326,0.262409,0.4082997,0.2967735],"study_design_scores_gemma":[0.0001605869,0.0001059479,0.0005270351,0.02294747,0.0009130103,0.0006494362,0.0002504924,0.001284292,0.0002971536,0.4579186,0.5148188,0.0001271279],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"review","genre_gemma":"methods","genre_scores_codex":[0.0002944221,0.4863179,0.1041638,0.3670589,0.03172876,0.0008885982,0.001643819,0.0003380244,0.007565672],"genre_scores_gemma":[0.01177536,0.448706,0.2383125,0.2498374,0.03597828,0.004957532,0.0009037911,0.0007859162,0.008743258],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9017134,"threshold_uncertainty_score":0.5197953,"prediction_status":"machine_predicted_unvalidated"},"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":"high"}],"label_agreement":"split"},{"id":"W2131621623","doi":"10.1136/ebm0005","title":"Making GRADE accessible: a proposal for graphic display of evidence quality assessments","year":2011,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":17,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Canadian Institutes of Health Research","keywords":"Grading (engineering); Outlier; Computer science; Quality of evidence; Quality (philosophy); Identification (biology); Information retrieval; Data mining; Artificial intelligence; Engineering","authors":[{"name":"Khalid S. Khan","is_ca":false},{"name":"E Borowiack","is_ca":false},{"name":"Carolien Roos","is_ca":false},{"name":"Monika Kowalska","is_ca":false},{"name":"A Zapalska","is_ca":false},{"name":"Ben W. Mol","is_ca":false},{"name":"Luciano Mignini","is_ca":false},{"name":"Catherine Meads","is_ca":false},{"name":"J Walczak","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9670815865336823,"gpt":0.6774248858922133,"spread":0.289656700641469,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.260081,0.004979757,0.00560377,0.03674971,0.003725724,0.02840058,0.01582882,0.01795562,0.05575243],"category_scores_gemma":[0.5941375,0.006962909,0.01152845,0.03369026,0.01066927,0.03222361,0.01935585,0.0255465,0.04820405],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.007563384,"about_ca_system_score_gemma":0.02043983,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006673343,"about_ca_topic_score_gemma":0.007260695,"domain_scores_codex":[0.7544459,0.1548608,0.04669398,0.006888798,0.03406541,0.003045029],"domain_scores_gemma":[0.5055519,0.2943533,0.03152345,0.0460787,0.1126206,0.009871956],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000516178,0.0002419886,0.0007175361,0.005076773,0.0003838072,0.0005554837,0.003545041,0.002256287,0.001057206,0.1446458,0.4634244,0.3775795],"study_design_scores_gemma":[0.0009445337,0.0003678715,0.0009827133,0.00670785,0.0004411979,0.0006199935,0.0009602799,0.008154074,0.001526989,0.2290792,0.7495777,0.0006375572],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.0005802005,0.001693386,0.8530369,0.07578594,0.009884142,0.008647786,0.003759475,0.03238363,0.0142285],"genre_scores_gemma":[0.00166259,0.0009233113,0.9851493,0.002733956,0.0008993821,0.004220207,0.0006621156,0.00196429,0.001784899],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7399189,"threshold_uncertainty_score":0.9124519,"prediction_status":"machine_predicted_unvalidated"},"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"],"domain":"methods","study_design":"theoretical_or_conceptual","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2095996801","doi":"10.1136/ebm.10.3.66-a","title":"Does clinical experience make up for failure to keep up to date?","year":2005,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":16,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Computer science; Psychology","authors":[{"name":"G R Norman","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.08451577484482581,"gpt":0.4603450181899665,"spread":0.3758292433451407,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03803993,0.001954632,0.004531838,0.003621823,0.003657827,0.009826454,0.00423119,0.0143446,0.009458878],"category_scores_gemma":[0.2791769,0.001316322,0.002916356,0.002868348,0.006905388,0.01619715,0.002535824,0.02813181,0.006613456],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003922406,"about_ca_system_score_gemma":0.006465492,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002239833,"about_ca_topic_score_gemma":0.006267241,"domain_scores_codex":[0.9618988,0.0130256,0.006686382,0.002375892,0.01507987,0.0009334175],"domain_scores_gemma":[0.7384975,0.1838673,0.01281166,0.005761904,0.05203019,0.007031385],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003721558,0.000007523087,0.0001091841,0.001313469,0.000068997,0.0002559165,0.000254342,0.00002169597,0.00003633495,0.001510225,0.977507,0.01887815],"study_design_scores_gemma":[0.00009749769,0.00005000105,0.0006072544,0.00563889,0.0001929987,0.001341377,0.0009907236,0.0001305984,0.0001209023,0.006783477,0.9839584,0.00008806376],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"editorial","genre_gemma":"editorial","genre_scores_codex":[0.00008563892,0.03158048,0.0007470054,0.3163556,0.6488825,0.00003224416,0.00005355314,0.00007141601,0.002191661],"genre_scores_gemma":[0.003172428,0.04419221,0.001224102,0.3301083,0.6141675,0.00009438212,0.00005400443,0.0001784926,0.006808699],"genre_candidate":"editorial","genre_consensus":"editorial","teacher_disagreement_score":0.03803993,"threshold_uncertainty_score":0.2011768,"prediction_status":"machine_predicted_unvalidated"},"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":"W2149125425","doi":"10.1136/eb-2013-101231","title":"EBM, CME and the EMR","year":2013,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Health Sciences Research and Education","field":"Health Professions","cited_by":16,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; College of Family Physicians of Canada; McGill University","funders":"","keywords":"Table (database); Computer science; Data mining","authors":[{"name":"Roland Grad","is_ca":true},{"name":"Pierre Pluye","is_ca":true},{"name":"Michael Shulha","is_ca":false},{"name":"David Li Tang","is_ca":false},{"name":"Karen Tu","is_ca":true},{"name":"Ken Goodman","is_ca":false},{"name":"Jamie Meuser","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2912953454851579,"gpt":0.524028586675324,"spread":0.2327332411901661,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.006571583,0.0001390035,0.0002828284,0.001760074,0.001965879,0.006637726,0.0006721862,0.004075048,0.03685458],"category_scores_gemma":[0.02149808,0.0001943995,0.0001942839,0.002375689,0.004641295,0.005396218,0.003484539,0.002908241,0.002276113],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.004348959,"about_ca_system_score_gemma":0.005514907,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007420695,"about_ca_topic_score_gemma":0.009797819,"domain_scores_codex":[0.9935546,0.003815673,0.0002678001,0.0002739036,0.001461441,0.0006266682],"domain_scores_gemma":[0.9826215,0.00920293,0.003210725,0.0003418007,0.001007107,0.003615895],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001958786,0.0003844833,0.01983159,0.001617183,0.00003114346,0.0009633614,0.01213704,0.0002115933,0.0002173223,0.3790806,0.1882799,0.3970499],"study_design_scores_gemma":[0.00004847565,0.0001522105,0.04039995,0.005284529,0.00001507034,0.001847188,0.01518408,0.0001886659,0.0001009296,0.08975989,0.8469791,0.00004002059],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0231142,0.1507847,0.000813348,0.6105759,0.0025954,0.00003278609,0.0001130744,0.00003297742,0.2119376],"genre_scores_gemma":[0.6724288,0.1385285,0.003517844,0.1040547,0.01477772,0.0001062115,0.0001346438,0.00005377875,0.06639777],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.03685458,"threshold_uncertainty_score":0.1232908,"prediction_status":"machine_predicted_unvalidated"},"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":"W2110458718","doi":"10.1136/ebm.13.1.3","title":"Less is more: where do the abstracts in the EBM journal come from?","year":2008,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":14,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Information retrieval; Psychology; History; Medicine; Computer science","authors":[{"name":"A.J. Robin Eady","is_ca":true},{"name":"Paul Glasziou","is_ca":false},{"name":"Brian Haynes","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8225130300919945,"gpt":0.533793031563296,"spread":0.2887199985286985,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.07537835,0.001203953,0.003058493,0.01807721,0.006968107,0.0358723,0.001913367,0.006269053,0.02901883],"category_scores_gemma":[0.3704084,0.001934456,0.002003956,0.01208123,0.008029978,0.03614252,0.004806535,0.007901391,0.02677817],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006856527,"about_ca_system_score_gemma":0.01244512,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005149812,"about_ca_topic_score_gemma":0.007536516,"domain_scores_codex":[0.9378356,0.02681838,0.0115861,0.002811732,0.01997356,0.0009745935],"domain_scores_gemma":[0.6525142,0.1493743,0.05821607,0.01726297,0.1039549,0.01867755],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.0008997631,0.0002053226,0.01296472,0.007990668,0.0002657921,0.0003303835,0.004963787,0.00005657012,0.001418377,0.007985023,0.4038602,0.5590595],"study_design_scores_gemma":[0.0002513413,0.0004072299,0.02643497,0.01989554,0.000517805,0.002133017,0.01634171,0.0001687279,0.001184996,0.02546695,0.9066464,0.0005513762],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.008465569,0.286389,0.006778223,0.6117553,0.06189113,0.0002827252,0.0006647043,0.0007207354,0.02305267],"genre_scores_gemma":[0.116849,0.3649314,0.05561083,0.3011415,0.1075835,0.0009779849,0.001787306,0.003480801,0.04763773],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.9246216,"threshold_uncertainty_score":0.3986436,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch","bibliometrics"],"domain":"reporting","study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["scholarly_communication"],"domain":null,"study_design":"observational","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2102619816","doi":"10.1136/ebm.14.3.68","title":"What does it take to put an ugly fact through the heart of a beautiful hypothesis?","year":2009,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Obesity and Health Practices","field":"Health Professions","cited_by":13,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"","keywords":"Tragedy (event); Medicine; Clinical trial; Intensive care medicine; Law and economics; Psychiatry; Economics; Pathology","authors":[{"name":"R. Brian Haynes","is_ca":true},{"name":"Graeme A. Haynes","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2493291391995027,"gpt":0.5017600065165541,"spread":0.2524308673170513,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1110211,0.001619556,0.005000109,0.004488926,0.005295496,0.01595836,0.006269186,0.01938627,0.01020383],"category_scores_gemma":[0.2879488,0.0009038204,0.002465273,0.00279269,0.04832578,0.04245076,0.00536413,0.02550292,0.00519645],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.005947161,"about_ca_system_score_gemma":0.01124049,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00513721,"about_ca_topic_score_gemma":0.006310949,"domain_scores_codex":[0.9551881,0.02874284,0.002800678,0.002637937,0.009432052,0.001198384],"domain_scores_gemma":[0.662293,0.271217,0.01373387,0.01117787,0.03030867,0.01126959],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0008935421,0.0003161197,0.003702814,0.007882549,0.001353379,0.001183295,0.003538572,0.0005759925,0.0004030592,0.1517779,0.5791866,0.2491861],"study_design_scores_gemma":[0.0002059832,0.0002980544,0.001726958,0.007627972,0.0003271814,0.001088476,0.006442953,0.00059474,0.0002837382,0.6240528,0.3570661,0.0002851999],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003355106,0.0877334,0.001909743,0.8874094,0.0212216,0.00001335347,0.00002926192,0.0000370673,0.001310603],"genre_scores_gemma":[0.03012497,0.1878523,0.01203716,0.6667874,0.09967633,0.0001330564,0.0001293544,0.000177985,0.003081484],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.8889789,"threshold_uncertainty_score":0.5871427,"prediction_status":"machine_predicted_unvalidated"},"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":"W2599817023","doi":"10.1136/ebmed-2017-110672","title":"Caesarean section is associated with offspring obesity in childhood and young adulthood","year":2017,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Birth, Development, and Health","field":"Medicine","cited_by":13,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Dalhousie University; Izaak Walton Killam Health Centre","funders":"","keywords":"Offspring; Section (typography); Caesarean section; Obesity; Early adulthood; Childhood obesity; Medicine; Young adult; Developmental psychology; Psychology; Pregnancy; Overweight; Biology; Endocrinology; Genetics","authors":[{"name":"Stefan Kuhle","is_ca":true},{"name":"Christy Woolcott","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03713527622359718,"gpt":0.2886793665991125,"spread":0.2515440903755153,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002834997,0.0006029251,0.001585071,0.001064793,0.001137957,0.001208919,0.002025226,0.01336561,0.01079812],"category_scores_gemma":[0.03349128,0.0004429122,0.001116259,0.001023115,0.0007802968,0.001504868,0.0007317435,0.01062338,0.007215948],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001995582,"about_ca_system_score_gemma":0.002857466,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008261067,"about_ca_topic_score_gemma":0.009544215,"domain_scores_codex":[0.9979371,0.0006961226,0.0003153982,0.0002335705,0.0006671977,0.0001507594],"domain_scores_gemma":[0.9870182,0.007468792,0.001041382,0.0002414069,0.003621274,0.0006089102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.000109515,0.00001290896,0.001047477,0.0005066975,0.00004802651,0.0003964014,0.00005901821,0.00001924211,0.00005040301,0.0004640285,0.9868264,0.01045984],"study_design_scores_gemma":[0.0006749486,0.000221745,0.01045507,0.01155926,0.0006043612,0.003804547,0.0007695203,0.0005662996,0.0004923966,0.005870479,0.9648212,0.0001601573],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"editorial","genre_scores_codex":[0.0006265406,0.01122141,0.00008421351,0.9222191,0.06284197,0.00002167086,0.0005980012,0.0000367149,0.002350494],"genre_scores_gemma":[0.007923444,0.01517889,0.0001759817,0.7897531,0.1817599,0.00008736359,0.0002948443,0.00003458661,0.00479192],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.01336561,"threshold_uncertainty_score":0.03612334,"prediction_status":"machine_predicted_unvalidated"},"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":"low"}],"label_agreement":"agree"},{"id":"W2056758832","doi":"10.1136/ebm.6.3.79","title":"Glyburide was as safe and effective as insulin in gestational diabetes","year":2001,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Gestational Diabetes Research and Management","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Gestational diabetes; Medicine; Diabetes mellitus; Insulin; Web of science; Gestation; Pregnancy; Internal medicine; Obstetrics; Gynecology; Endocrinology; Gastroenterology; Pediatrics","authors":[{"name":"EM Ryan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02656802855407683,"gpt":0.3333169266604353,"spread":0.3067488981063585,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002498984,0.00042151,0.001014095,0.0005035145,0.000310705,0.0008294963,0.0002481887,0.001788496,0.008622841],"category_scores_gemma":[0.006976519,0.0002745109,0.001233886,0.0004871039,0.0005082447,0.0005631778,0.0003474418,0.002193683,0.00076116],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007092607,"about_ca_system_score_gemma":0.0006479822,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001322337,"about_ca_topic_score_gemma":0.001508821,"domain_scores_codex":[0.999163,0.0004509607,0.0001160813,0.00008208126,0.0001236577,0.00006417607],"domain_scores_gemma":[0.9981311,0.00116793,0.0003899169,0.00007391471,0.00008579557,0.0001513838],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"nonrandomized_trial","study_design_scores_codex":[0.373534,0.0008586783,0.006335912,0.009218593,0.002928191,0.002220142,0.0002602888,0.0002854001,0.006852954,0.003501066,0.05751343,0.5364914],"study_design_scores_gemma":[0.4456731,0.07290837,0.07704704,0.007343179,0.0123625,0.003775605,0.000406711,0.0008376001,0.01139326,0.004792408,0.3632699,0.0001902758],"study_design_candidate":"nonrandomized_trial","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.2337853,0.6519133,0.002184551,0.05321173,0.01699497,0.001109043,0.005652194,0.0002470961,0.03490182],"genre_scores_gemma":[0.7272977,0.2184282,0.002404107,0.02538127,0.009476068,0.001037023,0.002995347,0.0001069195,0.01287351],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.008622841,"threshold_uncertainty_score":0.02884632,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"nonrandomized_trial","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":[],"domain":null,"study_design":"randomized_trial","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2101003324","doi":"10.1136/ebm1189","title":"Treatment with the probiotic VSL#3 as an adjunctive therapy in relapsing mild-to-moderate ulcerative colitis significantly reduces ulcerative colitis disease activity","year":2011,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Probiotics and Fermented Foods","field":"Agricultural and Biological Sciences","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; Stollery Children's Hospital","funders":"","keywords":"Ulcerative colitis; Probiotic; Medicine; Placebo; Internal medicine; Gastroenterology; Colitis; Adjunctive treatment; Disease; Randomized controlled trial; Therapeutic effect; Pathology; Biology","authors":[{"name":"Jean-François Turcotte","is_ca":true},{"name":"H Huynh","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07352506901589025,"gpt":0.2800703998675619,"spread":0.2065453308516717,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003013378,0.000772314,0.002573127,0.0006416448,0.001570693,0.001259248,0.001526695,0.02574458,0.006384345],"category_scores_gemma":[0.0174838,0.0004640845,0.001396631,0.0009550606,0.001077362,0.001518763,0.0004647297,0.01758485,0.006777785],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003564917,"about_ca_system_score_gemma":0.002198836,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005314239,"about_ca_topic_score_gemma":0.01042838,"domain_scores_codex":[0.9978344,0.0007986412,0.0003281979,0.0001995761,0.0006458241,0.0001933915],"domain_scores_gemma":[0.9919021,0.00525598,0.000521719,0.0001203721,0.001667652,0.0005322805],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00153605,0.00006262019,0.000334372,0.0004822633,0.0000600822,0.0007209497,0.00004218326,0.00004674519,0.0003826651,0.0005358946,0.9727737,0.02302247],"study_design_scores_gemma":[0.003743966,0.0009286627,0.004496764,0.001940003,0.0003876336,0.001744421,0.000277751,0.0005095178,0.0009817654,0.004229515,0.9806423,0.000117769],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"editorial","genre_scores_codex":[0.001058096,0.01471645,0.0001233705,0.9361899,0.04469841,0.00004008681,0.0003562979,0.00004794832,0.002769395],"genre_scores_gemma":[0.007524719,0.006326854,0.0002532023,0.9083525,0.07095262,0.00009962474,0.0001603711,0.00002354042,0.006306628],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.02574458,"threshold_uncertainty_score":0.02586544,"prediction_status":"machine_predicted_unvalidated"},"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":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W1969167985","doi":"10.1136/ebmed-2014-110090","title":"Physician practice and PECARN rule outperform CATCH and CHALICE rules based on the detection of traumatic brain injury as defined by PECARN","year":2014,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Traumatic Brain Injury and Neurovascular Disturbances","field":"Medicine","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Computed tomography; Pediatrics; Gold standard (test); Cohort; Surgery; Internal medicine","authors":[{"name":"Franz E Babl","is_ca":false},{"name":"Silvia Bressan","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04154083209929348,"gpt":0.2961649819021624,"spread":0.254624149802869,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01815209,0.0004981456,0.001542125,0.001770209,0.0008466964,0.002921141,0.002499272,0.006118192,0.01387286],"category_scores_gemma":[0.2902958,0.000416591,0.0020966,0.001920991,0.001155233,0.002541902,0.001303551,0.007781445,0.005547681],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003075999,"about_ca_system_score_gemma":0.004472937,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03276905,"about_ca_topic_score_gemma":0.04134151,"domain_scores_codex":[0.9820732,0.008171713,0.002395346,0.001875872,0.004841843,0.0006420909],"domain_scores_gemma":[0.7655528,0.180776,0.0136758,0.002492062,0.03444115,0.003062178],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0007642013,0.00005073228,0.02261847,0.00117579,0.000416728,0.000144565,0.0003476056,0.0002616743,0.00003236948,0.001077907,0.9226591,0.05045087],"study_design_scores_gemma":[0.004071175,0.001186011,0.1843798,0.04109889,0.0037762,0.001728339,0.00389543,0.009539392,0.001263991,0.02343666,0.724482,0.001142244],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01202328,0.02138572,0.001964996,0.8999106,0.02762524,0.000339531,0.01653058,0.0002475574,0.01997249],"genre_scores_gemma":[0.2787138,0.03004302,0.009244848,0.5868314,0.06984198,0.001240317,0.01262118,0.0005031768,0.01096025],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.03276905,"threshold_uncertainty_score":0.09599859,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"editorial","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":"split"},{"id":"W2270048893","doi":"10.1136/ebmed-2015-110302","title":"‘Cognitive biases plus’: covert subverters of healthcare evidence","year":2015,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Patient-Provider Communication in Healthcare","field":"Health Professions","cited_by":12,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Western University; South Bruce Grey Health Centre; University of Saskatchewan","funders":"","keywords":"Covert; Health care; Cognition; Cognitive bias; Psychology; Confirmation bias; Quality (philosophy); Causal inference; Social psychology; Cognitive psychology; Medicine; Political science; Epistemology; Law","authors":[{"name":"Shashi S. Seshia","is_ca":true},{"name":"Michael Makhinson","is_ca":false},{"name":"G. Bryan Young","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6880623042370246,"gpt":0.522648864000794,"spread":0.1654134402362306,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1361476,0.001293074,0.001545817,0.006405069,0.004326816,0.016939,0.003169301,0.006134846,0.004448874],"category_scores_gemma":[0.359581,0.001360392,0.001740174,0.002288941,0.05170465,0.02126847,0.01584607,0.01147581,0.0007243865],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.006769222,"about_ca_system_score_gemma":0.007330609,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00155754,"about_ca_topic_score_gemma":0.001099281,"domain_scores_codex":[0.8031616,0.1200158,0.01238968,0.01434915,0.04501636,0.005067409],"domain_scores_gemma":[0.4937609,0.3882819,0.0360192,0.05759261,0.01893968,0.005405773],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0006014244,0.0001973105,0.0424655,0.002370568,0.001363454,0.003303844,0.1387466,0.002102224,0.004078347,0.6097886,0.01066499,0.1843172],"study_design_scores_gemma":[0.0001860197,0.0003225408,0.008614449,0.002850763,0.0003707953,0.004378837,0.01556578,0.004860465,0.003159745,0.9005732,0.05888051,0.0002369162],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.2803678,0.02199179,0.2335543,0.2950023,0.003897457,0.00114807,0.0002682821,0.001118184,0.1626518],"genre_scores_gemma":[0.9463458,0.001800612,0.02840669,0.01944333,0.001328607,0.0002958099,0.00004771458,0.0001807222,0.002150658],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.8638524,"threshold_uncertainty_score":0.7200259,"prediction_status":"machine_predicted_unvalidated"},"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":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2233830625","doi":"10.1136/ebmed-2015-110354","title":"Maternal postpartum high-dose vitamin D3 supplementation (6400 IU/day) or conventional infant vitamin D3 supplementation (400 IU/day) lead to similar vitamin D status of healthy exclusively/fully breastfeeding infants by 7 months of age","year":2016,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Vitamin D Research Studies","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hospital for Sick Children; University of Toronto; SickKids Foundation","funders":"","keywords":"Breastfeeding; Breast milk; Vitamin D and neurology; Medicine; Rickets; Vitamin; Lactation; Breast feeding; Pediatrics; vitamin D deficiency; Cholecalciferol; Gastroenterology; Internal medicine; Pregnancy; Chemistry; Biology","authors":[{"name":"Daniel Roth","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04154429869868904,"gpt":0.354947721441828,"spread":0.313403422743139,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.005697481,0.0005911768,0.002243968,0.000714011,0.0007715163,0.001489813,0.001810245,0.007286873,0.03720424],"category_scores_gemma":[0.04270904,0.0002943231,0.003107374,0.0007489292,0.0009020056,0.001245445,0.0005309092,0.005226836,0.002695192],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001659388,"about_ca_system_score_gemma":0.002245382,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01012512,"about_ca_topic_score_gemma":0.01311765,"domain_scores_codex":[0.9974843,0.001378586,0.0002545892,0.0003547151,0.0003659064,0.0001618194],"domain_scores_gemma":[0.9801013,0.01411285,0.002527261,0.000390458,0.001868279,0.0009998424],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.192068,0.0009871253,0.005844998,0.01970783,0.00616658,0.0005832819,0.0003545553,0.0001982163,0.001050759,0.002323496,0.6225838,0.1481314],"study_design_scores_gemma":[0.3213344,0.03991912,0.05624727,0.07062602,0.05231624,0.001611866,0.001314627,0.001443957,0.004447163,0.009645385,0.4405888,0.0005051295],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"editorial","genre_scores_codex":[0.04958621,0.3200584,0.001107404,0.4821208,0.0912308,0.001529716,0.01657414,0.0003654113,0.03742717],"genre_scores_gemma":[0.2897282,0.122811,0.002186726,0.4984471,0.05056261,0.001836864,0.004944101,0.000219605,0.02926384],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.03720424,"threshold_uncertainty_score":0.1244606,"prediction_status":"machine_predicted_unvalidated"},"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":"randomized_trial","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2317501191","doi":"10.1136/eb-2014-101798","title":"Fifty years of age-based screening: time for a new risk-based screening approach","year":2014,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Global Cancer Incidence and Screening","field":"Medicine","cited_by":11,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Mammography; Breast cancer; Breast cancer screening; Incidence (geometry); Gynecology; Pediatrics; Cancer; Internal medicine","authors":[{"name":"Karla Kerlikowske","is_ca":false},{"name":"Margaret E O'Kane","is_ca":false},{"name":"Laura J. Esserman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.1768554982965745,"gpt":0.3499018134933837,"spread":0.1730463151968092,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02241888,0.001559501,0.003710366,0.002207051,0.003741385,0.005761511,0.005929107,0.03426359,0.02983595],"category_scores_gemma":[0.1258791,0.001011474,0.003785206,0.001143154,0.00888463,0.01863494,0.003973115,0.05346116,0.01037357],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009212038,"about_ca_system_score_gemma":0.01676483,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02304831,"about_ca_topic_score_gemma":0.03385798,"domain_scores_codex":[0.9836459,0.007671384,0.001421381,0.001604988,0.004651002,0.001005268],"domain_scores_gemma":[0.8845243,0.08458615,0.003578613,0.002868082,0.01467892,0.009763893],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001519988,0.00002976712,0.00009479152,0.0007457474,0.00005638505,0.00009241942,0.0001821827,0.0000402521,0.00004665435,0.007688918,0.9629513,0.02791951],"study_design_scores_gemma":[0.000420063,0.0001803808,0.0009082841,0.008649597,0.0002276783,0.0003071106,0.0006425545,0.0001406005,0.00009610534,0.02497209,0.9633385,0.0001171154],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.00008214784,0.02338764,0.000245726,0.916972,0.05738565,0.00001992687,0.0001280321,0.00003762025,0.001741304],"genre_scores_gemma":[0.002147019,0.017954,0.0004438132,0.8996486,0.07653961,0.00009935434,0.00008916722,0.00004533105,0.003033052],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.03426359,"threshold_uncertainty_score":0.1185638,"prediction_status":"machine_predicted_unvalidated"},"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":"W2065926725","doi":"10.1136/ebm.14.2.42","title":"Review: inhaled anticholinergics increase risk of major cardiovascular events in COPD","year":2009,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University Health Network; University of Toronto","funders":"Pfizer","keywords":"COPD; Medicine; Intensive care medicine; Anesthesia; Internal medicine","authors":[{"name":"Matthew B. Stanbrook","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.03404952557897534,"gpt":0.3180702947815399,"spread":0.2840207692025645,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003517357,0.0008650059,0.005127997,0.001255711,0.0006466503,0.002061537,0.001482222,0.00682191,0.004503544],"category_scores_gemma":[0.01495418,0.0006771991,0.002243004,0.001679568,0.0006895475,0.00162359,0.0007660332,0.004171554,0.001883626],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001534431,"about_ca_system_score_gemma":0.003596752,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002933024,"about_ca_topic_score_gemma":0.0109053,"domain_scores_codex":[0.9976473,0.0006223698,0.0006314301,0.0001893656,0.0008217119,0.00008780944],"domain_scores_gemma":[0.9877628,0.008315503,0.001909052,0.0001207728,0.001632779,0.0002590878],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0008444004,0.00004980058,0.001145804,0.1113807,0.003011701,0.0008932422,0.0001051715,0.0001399337,0.0004015368,0.001016436,0.6626562,0.218355],"study_design_scores_gemma":[0.001981975,0.0004769744,0.01217496,0.140238,0.0134212,0.002439342,0.0002146176,0.0002389991,0.0003049723,0.003153832,0.8252592,0.00009584609],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"commentary","genre_scores_codex":[0.0003588967,0.9557514,0.0001103227,0.02763482,0.01390062,0.00006752922,0.0002611791,0.00002732802,0.001887887],"genre_scores_gemma":[0.005336683,0.905099,0.0003832112,0.06491818,0.02138335,0.0001047762,0.0001507338,0.00001121432,0.002612878],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.00682191,"threshold_uncertainty_score":0.01860178,"prediction_status":"machine_predicted_unvalidated"},"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":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2129733011","doi":"10.1136/ebm.2011.100252","title":"Early supplemental parenteral nutrition in critically ill adults increased infections, ICU length of stay and cost","year":2011,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Clinical Nutrition and Gastroenterology","field":"Nursing","cited_by":10,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Kingston General Hospital","funders":"","keywords":"Medicine; Web of science; Parenteral nutrition; Pediatrics; Internal medicine; Adverse effect; Gastroenterology; Gynecology; Meta-analysis","authors":[{"name":"Daren K. Heyland","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0438096655080412,"gpt":0.3242809925175503,"spread":0.2804713270095091,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003121615,0.000941642,0.001634025,0.001510978,0.001665386,0.001834772,0.002887853,0.01538128,0.02588469],"category_scores_gemma":[0.03866107,0.0004446403,0.001603753,0.001563102,0.001136501,0.001884006,0.0009215657,0.0111119,0.007239848],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003271706,"about_ca_system_score_gemma":0.006090235,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03039918,"about_ca_topic_score_gemma":0.03494129,"domain_scores_codex":[0.9973277,0.000765785,0.0002698527,0.0003372481,0.001076943,0.0002224351],"domain_scores_gemma":[0.9774874,0.0115888,0.001612023,0.0002785866,0.007718083,0.001315108],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.00004523788,0.000006748847,0.0001065463,0.000309279,0.00001674846,0.00004949124,0.00002980746,0.000009931588,0.00001820155,0.0002756131,0.9946411,0.004491333],"study_design_scores_gemma":[0.0005455108,0.0002328641,0.005340587,0.01508084,0.0002899381,0.0005236045,0.0004933808,0.0002075373,0.000287115,0.003142728,0.97374,0.0001158345],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0003666139,0.01395033,0.00007039912,0.8354,0.1442727,0.00003515942,0.001022482,0.00005217957,0.004830117],"genre_scores_gemma":[0.007343901,0.02563092,0.0002435746,0.7448832,0.2099625,0.0001179827,0.0005017571,0.00007617929,0.01124],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.03039918,"threshold_uncertainty_score":0.08659291,"prediction_status":"machine_predicted_unvalidated"},"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":"design_other","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2131342734","doi":"10.1136/ebm.2011.000013","title":"Older women who use bisphosphonate for longer than 5 years may have increased odds of a subtrochanteric or femoral shaft fracture, but absolute risk is low","year":2011,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Bone health and osteoporosis research","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Medicine; Osteoporosis; Web of science; Bisphosphonate; Surgery; Gynecology; Internal medicine; Meta-analysis","authors":[{"name":"Bo Abrahamsen","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.0839805015196584,"gpt":0.3456652572770419,"spread":0.2616847557573835,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003536318,0.0009320653,0.002178653,0.001438804,0.001005019,0.001462724,0.003163769,0.01049242,0.0154128],"category_scores_gemma":[0.03428322,0.0005943562,0.001578889,0.001572396,0.001077235,0.001994059,0.0004802567,0.008704444,0.008024468],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001578693,"about_ca_system_score_gemma":0.002422847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.03280112,"about_ca_topic_score_gemma":0.03808739,"domain_scores_codex":[0.9973938,0.0007188449,0.0003422075,0.0004328341,0.0009489169,0.0001633626],"domain_scores_gemma":[0.9797699,0.01062876,0.001326361,0.0003031779,0.007369886,0.0006017876],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"observational","study_design_scores_codex":[0.0003453581,0.00001386986,0.001569542,0.001488996,0.0001224915,0.0001945101,0.00008153199,0.00003280134,0.0001072567,0.0006534081,0.9817695,0.0136206],"study_design_scores_gemma":[0.001418298,0.0004360896,0.0390542,0.01750658,0.001590596,0.00163032,0.0009660769,0.0009457556,0.0007683786,0.007113036,0.9280715,0.0004992659],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.001662646,0.05445051,0.0005757432,0.7444646,0.1848882,0.0001021392,0.003801348,0.0001658637,0.009889017],"genre_scores_gemma":[0.024194,0.05719026,0.000718613,0.5540161,0.3432218,0.0002185497,0.001328384,0.0001290418,0.01898335],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.03280112,"threshold_uncertainty_score":0.06522042,"prediction_status":"machine_predicted_unvalidated"},"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":"observational","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2323588829","doi":"10.1136/ebmed-2016-110381","title":"High prevalence of depression in medical residents: the sad reality of medical training","year":2016,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Healthcare professionals’ stress and burnout","field":"Health Professions","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Depression (economics); Medicine; Web of science; Population; Meta-analysis; Depressive symptoms; Psychiatry; Internal medicine","authors":[{"name":"Shirley Schipper","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.2053745751891052,"gpt":0.4931592620581465,"spread":0.2877846868690413,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00608117,0.0009787803,0.003613029,0.00235625,0.001041369,0.002263766,0.002982367,0.007744309,0.01200429],"category_scores_gemma":[0.05758479,0.0007077914,0.002159691,0.002919007,0.001484461,0.00308845,0.001101596,0.006096799,0.002808307],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002406467,"about_ca_system_score_gemma":0.004072154,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01570863,"about_ca_topic_score_gemma":0.02311905,"domain_scores_codex":[0.993183,0.002477069,0.001223534,0.0008300209,0.002068734,0.0002177527],"domain_scores_gemma":[0.9510527,0.02893074,0.008276445,0.000613771,0.01009459,0.001031758],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002753132,0.00001412021,0.001846169,0.02520956,0.00106792,0.0002767326,0.0003213375,0.00004372796,0.0001254807,0.0007560487,0.9477173,0.02234631],"study_design_scores_gemma":[0.001412935,0.0005410983,0.04532573,0.2916326,0.009833897,0.002819752,0.003451039,0.0005879672,0.0008353305,0.009094738,0.6339815,0.000483488],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.001415944,0.3270888,0.0003185655,0.5777833,0.08781043,0.00008889849,0.002884983,0.00005670117,0.002552428],"genre_scores_gemma":[0.04019901,0.2901817,0.000836265,0.5005747,0.160804,0.0003252556,0.001495849,0.0001000299,0.00548322],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.01570863,"threshold_uncertainty_score":0.04015833,"prediction_status":"machine_predicted_unvalidated"},"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":"W2119126532","doi":"10.1136/ebm.5.2.64","title":"Review: interactive, but not didactic, continuing medical education is effective in changing physician performance","year":2000,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Innovations in Medical Education","field":"Medicine","cited_by":9,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Web of science; CINAHL; Medicine; Psychological intervention; PsycINFO; MEDLINE; Family medicine; Nursing; Meta-analysis; Internal medicine","authors":[{"name":"H. A. Holm","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.01635019418769604,"gpt":0.3557592179699208,"spread":0.3394090237822247,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004902937,0.001237662,0.007360772,0.004374808,0.000665913,0.003395249,0.001936451,0.002586451,0.01812686],"category_scores_gemma":[0.0342392,0.0008273579,0.003378874,0.009081703,0.0008485795,0.001786215,0.0009781156,0.001845176,0.002278388],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003823501,"about_ca_system_score_gemma":0.006725748,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.009816447,"about_ca_topic_score_gemma":0.02186775,"domain_scores_codex":[0.9954555,0.001534041,0.001135375,0.0002946895,0.00146003,0.0001202706],"domain_scores_gemma":[0.9818295,0.0129166,0.003052982,0.0001799185,0.001700707,0.0003202083],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"observational","study_design_scores_codex":[0.000639843,0.00003084749,0.0003802174,0.7779772,0.003738375,0.0001235062,0.0001783594,0.0001039274,0.0001627161,0.0003450924,0.03912156,0.1771984],"study_design_scores_gemma":[0.001284271,0.0005825923,0.008569872,0.8280114,0.02985541,0.0005366237,0.0002860667,0.0001292051,0.0003491137,0.0007513461,0.1295838,0.00006043387],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.0002490229,0.9971862,0.00006304685,0.0007387737,0.0004849685,0.0001796211,0.0002138744,0.00001368722,0.0008707276],"genre_scores_gemma":[0.004578256,0.9932503,0.0001772606,0.001019999,0.000235243,0.000232195,0.0001393335,0.00000715722,0.0003603224],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.01812686,"threshold_uncertainty_score":0.06064045,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":[],"domain":null,"study_design":"observational","genre":"review","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":"W2127381235","doi":"10.1136/ebm.11.2.61","title":"The Canadian CT Head Rule reduced the need for CT scans more than the New Orleans Criteria in minor head injury","year":2006,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Traumatic Brain Injury and Neurovascular Disturbances","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"Western University","funders":"","keywords":"Head (geology); Head injury; Minor (academic); Nuclear medicine; Medicine; Surgery; Geology; Humanities; Art","authors":[{"name":"M. Edmonds","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07000077531802613,"gpt":0.3414905236998589,"spread":0.2714897483818328,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01034393,0.0005254261,0.0007535998,0.001311905,0.001317988,0.001319121,0.001243685,0.0009986007,0.006799093],"category_scores_gemma":[0.06709167,0.0003101081,0.001241569,0.002054955,0.0006698731,0.0007651059,0.001226046,0.001189124,0.001278108],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008411447,"about_ca_system_score_gemma":0.02960643,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.78714,"about_ca_topic_score_gemma":0.8272395,"domain_scores_codex":[0.993038,0.001887753,0.000523588,0.0007261615,0.003150208,0.0006743145],"domain_scores_gemma":[0.9738391,0.008051242,0.002592727,0.001260947,0.01249807,0.001757922],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.007991089,0.0004091498,0.4337411,0.0008714605,0.0005158011,0.0001966241,0.0007315361,0.003629808,0.0005785729,0.001459468,0.2824498,0.2674255],"study_design_scores_gemma":[0.001662328,0.001518318,0.9228201,0.0008928068,0.001056849,0.0002901604,0.0005852315,0.005958307,0.001328023,0.000943732,0.06275246,0.0001915798],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7812119,0.00561478,0.008126765,0.02932849,0.002693354,0.003407692,0.05354765,0.001476923,0.1145925],"genre_scores_gemma":[0.9547384,0.001401674,0.01138793,0.006656216,0.0003976445,0.0006850223,0.01653351,0.0002192377,0.007980298],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.78714,"threshold_uncertainty_score":0.4282271,"prediction_status":"machine_predicted_unvalidated"},"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":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"split"},{"id":"W2192978482","doi":"10.1136/ebmed-2015-110342","title":"Improving reporting and utility of evaluations of complex interventions","year":2015,"lang":"en","type":"editorial","venue":"Evidence-Based Medicine","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":7,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta","funders":"","keywords":"Psychological intervention; Computer science; Psychology; Psychiatry","authors":[{"name":"Denise Campbell‐Scherer","is_ca":true},{"name":"Richard Saitz","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8886211946752411,"gpt":0.7497801860907314,"spread":0.1388410085845098,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7578909,0.005038891,0.009841763,0.02357375,0.003126807,0.02138597,0.0106005,0.007664463,0.01647322],"category_scores_gemma":[0.9214269,0.00291692,0.009717034,0.01884457,0.009188483,0.02418896,0.01785258,0.00914659,0.003672683],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01813356,"about_ca_system_score_gemma":0.02352851,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007410207,"about_ca_topic_score_gemma":0.006681892,"domain_scores_codex":[0.0818383,0.7062206,0.11961,0.01701086,0.07356816,0.001752079],"domain_scores_gemma":[0.03031524,0.7339755,0.07304443,0.0912085,0.06981925,0.001637127],"domain_codex":"methods","domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.007275926,0.001746124,0.09657317,0.05429226,0.01296122,0.0003301526,0.01029405,0.009321925,0.0009038256,0.06460989,0.07266612,0.6690254],"study_design_scores_gemma":[0.007687572,0.009070758,0.1704037,0.1720614,0.007934696,0.001236656,0.009117822,0.03655001,0.009164741,0.2958047,0.2791024,0.001865601],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"editorial","genre_scores_codex":[0.06652972,0.0886027,0.5064128,0.05046403,0.01423705,0.1112917,0.04597807,0.005013025,0.111471],"genre_scores_gemma":[0.4366729,0.01378692,0.3665646,0.01245436,0.003622006,0.1464059,0.01489051,0.001871222,0.003731548],"genre_candidate":"editorial","genre_consensus":null,"teacher_disagreement_score":0.2421091,"threshold_uncertainty_score":0.2985637,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"editorial","about_ca_system":false,"about_ca_topic":false,"confidence":"high"}],"label_agreement":"agree"},{"id":"W1965501053","doi":"10.1136/ebm.14.5.130-b","title":"\"Pragmatic\" clinical trials: from whose perspective?","year":2009,"lang":"en","type":"article","venue":"Evidence-Based Medicine","topic":"Smoking Behavior and Cessation","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University; Western University","funders":"","keywords":"Perspective (graphical); Clinical trial; Psychology; Medicine; Computer science; Artificial intelligence; Internal medicine","authors":[{"name":"P. J Karanicolas","is_ca":true},{"name":"Víctor M. Montori","is_ca":false},{"name":"Holger J. Schünemann","is_ca":true},{"name":"Gordon Guyatt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.3327434213927938,"gpt":0.5194585994828483,"spread":0.1867151780900545,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5884989,0.00430057,0.01897369,0.00778209,0.007459566,0.04213277,0.01331543,0.06701656,0.005784739],"category_scores_gemma":[0.7493704,0.004440546,0.005736648,0.008151555,0.06634427,0.07869471,0.01224766,0.07945019,0.003134055],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01767007,"about_ca_system_score_gemma":0.04907704,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0055282,"about_ca_topic_score_gemma":0.00552596,"domain_scores_codex":[0.2504431,0.6708577,0.03637306,0.008738242,0.03071137,0.002876427],"domain_scores_gemma":[0.170881,0.740939,0.01639243,0.02360698,0.03880186,0.009378627],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.001364897,0.0002684077,0.00091567,0.02715625,0.002365825,0.0006009218,0.0101174,0.001404364,0.0002544798,0.4643225,0.3577501,0.1334792],"study_design_scores_gemma":[0.002027093,0.0003278881,0.0003008498,0.04309336,0.0005743162,0.0004477632,0.004146522,0.001489845,0.0002103936,0.6920171,0.2550302,0.0003346735],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.0001452523,0.1133168,0.008574272,0.8517789,0.02372946,0.0002372571,0.00006082049,0.00007096057,0.002086233],"genre_scores_gemma":[0.01863238,0.0857904,0.03990864,0.7762682,0.07506728,0.00275466,0.0000971689,0.0004095284,0.001071664],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.4115011,"threshold_uncertainty_score":0.5074542,"prediction_status":"machine_predicted_unvalidated"},"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":["metaresearch"],"domain":"methods","study_design":"not_applicable","genre":"commentary","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"split"},{"id":"W2004100133","doi":"10.1136/ebm.11.4.115","title":"Glucosamine and chondroitin sulphate did not improve pain in osteoarthritis of the knee","year":2006,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Osteoarthritis Treatment and Mechanisms","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"","keywords":"Osteoarthritis; Glucosamine; Chondroitin sulfate; Chondroitin; Medicine; Knee pain; Chemistry; Biochemistry; Pathology; Glycosaminoglycan; Anatomy; Alternative medicine","authors":[{"name":"Michael J. Baime","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.02055674383385181,"gpt":0.2450992088961687,"spread":0.2245424650623169,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.001450227,0.0004415146,0.001884254,0.0006642295,0.0003187611,0.000557779,0.0004897416,0.001572445,0.01063873],"category_scores_gemma":[0.003036631,0.0004130327,0.002117355,0.0007904431,0.0006172462,0.0006841058,0.0003576124,0.00131176,0.0008266581],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004418588,"about_ca_system_score_gemma":0.001035522,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002745525,"about_ca_topic_score_gemma":0.006624439,"domain_scores_codex":[0.9989993,0.0004810852,0.0001890748,0.00006236284,0.0001679153,0.000100255],"domain_scores_gemma":[0.9989508,0.0003811245,0.0002389908,0.00008342056,0.0001135476,0.0002320135],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"not_applicable","study_design_scores_codex":[0.7396688,0.006182512,0.002623959,0.01479689,0.00668923,0.0003087631,0.0001995544,0.0003269724,0.01202792,0.0005538243,0.007923376,0.2086982],"study_design_scores_gemma":[0.6434831,0.295852,0.03609525,0.002010792,0.007452523,0.000332486,0.0002032053,0.0003894268,0.002659047,0.000804605,0.01063739,0.000080224],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"commentary","genre_scores_codex":[0.8194309,0.1491211,0.0005656623,0.005421962,0.00437728,0.001645011,0.002326838,0.0001860364,0.01692519],"genre_scores_gemma":[0.9612175,0.02854107,0.0008020536,0.002423189,0.001235479,0.0004695542,0.0007798445,0.00002482468,0.004506588],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.01063873,"threshold_uncertainty_score":0.03559017,"prediction_status":"machine_predicted_unvalidated"},"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":"design_other","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"split"},{"id":"W2025044100","doi":"10.1136/eb-2013-101497","title":"The effectiveness of pressure ulcer risk assessment instruments and associated intervention protocols remains uncertain","year":2014,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Pressure Ulcer Prevention and Management","field":"Health Professions","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"","keywords":"Medicine; Web of science; Psychological intervention; Incidence (geometry); Internal medicine; Surgery; Psychiatry; Meta-analysis","authors":[{"name":"Madhuri Reddy","is_ca":false},{"name":"Sudeep S. Gill","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.07947037456200354,"gpt":0.4506448367897557,"spread":0.3711744622277522,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.09210963,0.00199363,0.009834846,0.005481863,0.002033398,0.006195952,0.009254346,0.01604468,0.03118123],"category_scores_gemma":[0.5871204,0.001600964,0.008562244,0.005069227,0.004160517,0.010119,0.003136553,0.0144362,0.009357826],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009308835,"about_ca_system_score_gemma":0.02681684,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02724971,"about_ca_topic_score_gemma":0.03227431,"domain_scores_codex":[0.905561,0.05203037,0.01811201,0.004273758,0.01792463,0.002098138],"domain_scores_gemma":[0.4148754,0.462095,0.0306557,0.006524754,0.08186369,0.00398544],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.002318228,0.0001701428,0.0006916658,0.104689,0.002179232,0.0001491314,0.0006199726,0.0003211233,0.0001274205,0.009668209,0.7061837,0.1728821],"study_design_scores_gemma":[0.005153758,0.0008244459,0.004104541,0.4226288,0.007420037,0.0002789244,0.00141744,0.0007432197,0.0007535003,0.01742562,0.538849,0.0004006301],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0008184507,0.3096144,0.001418706,0.5714779,0.102656,0.0008205342,0.003336094,0.0001256231,0.009732314],"genre_scores_gemma":[0.02559292,0.3245507,0.005360695,0.5500087,0.08403308,0.002999622,0.001748196,0.0002735604,0.005432613],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.09210963,"threshold_uncertainty_score":0.4871281,"prediction_status":"machine_predicted_unvalidated"},"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":"W2024756877","doi":"10.1136/ebm.12.4.116","title":"Review: drugs for mild to moderate hypertension in pregnancy reduce the risk of severe hypertension but not pre-eclampsia","year":2007,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Pregnancy and preeclampsia studies","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of British Columbia","funders":"","keywords":"Eclampsia; Medicine; Pregnancy; Hypertension in Pregnancy; Obstetrics; Preeclampsia; Biology","authors":[{"name":"Laura A. Magee","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.09058092051650665,"gpt":0.3245942603351467,"spread":0.2340133398186401,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.00229474,0.001522805,0.006939325,0.004511381,0.0005850312,0.002178981,0.001581127,0.002502782,0.01317484],"category_scores_gemma":[0.01421314,0.0005741937,0.00426989,0.005968196,0.0006511928,0.001711198,0.0008739133,0.001709561,0.001300017],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002019123,"about_ca_system_score_gemma":0.003992397,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.005641876,"about_ca_topic_score_gemma":0.01055358,"domain_scores_codex":[0.9975381,0.001005458,0.0006066438,0.0001891276,0.0005753473,0.00008548211],"domain_scores_gemma":[0.9936507,0.004503922,0.001143553,0.00006760126,0.0005092677,0.0001249648],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.0006876401,0.00002655397,0.0002322374,0.8192798,0.003907927,0.0001309512,0.0001316708,0.0001150826,0.000228554,0.0005596625,0.03794362,0.1367563],"study_design_scores_gemma":[0.001858286,0.0006135958,0.0052891,0.7572697,0.03334678,0.0009729962,0.000229153,0.0001483165,0.0004488957,0.001204778,0.1985466,0.00007178743],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"commentary","genre_scores_codex":[0.0001972476,0.9976291,0.00005798693,0.0006492491,0.0005307185,0.0001591713,0.0001838019,0.00001544716,0.0005773447],"genre_scores_gemma":[0.004641898,0.9922825,0.000291105,0.001622142,0.00030883,0.0002607538,0.0001303335,0.000007555609,0.0004550008],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.01317484,"threshold_uncertainty_score":0.04407418,"prediction_status":"machine_predicted_unvalidated"},"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":"systematic_review","genre":"review","about_ca_system":false,"about_ca_topic":false,"confidence":"low"}],"label_agreement":"agree"},{"id":"W2158987550","doi":"10.1136/eb-2014-101789","title":"Benzodiazepines and opioids need to be prescribed with caution in advanced COPD","year":2014,"lang":"en","type":"letter","venue":"Evidence-Based Medicine","topic":"Chronic Obstructive Pulmonary Disease (COPD) Research","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Toronto; St. Michael's Hospital","funders":"","keywords":"Medicine; COPD; Prospective cohort study; Pulmonary disease; Internal medicine; Cohort","authors":[{"name":"Nicholas T. Vozoris","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.04541813758142248,"gpt":0.3187812964703936,"spread":0.2733631588889711,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003300099,0.0007599366,0.00178019,0.0011357,0.002164601,0.002653464,0.003423226,0.02002004,0.02122952],"category_scores_gemma":[0.04214667,0.0005943774,0.00151392,0.001212866,0.002552766,0.004817818,0.001191051,0.01789368,0.01369847],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002856654,"about_ca_system_score_gemma":0.005475736,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02326018,"about_ca_topic_score_gemma":0.03737667,"domain_scores_codex":[0.9952625,0.001644716,0.0008009335,0.0004710843,0.001493173,0.0003275996],"domain_scores_gemma":[0.9751657,0.01396302,0.002136061,0.0005097329,0.006649785,0.001575783],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00004345915,0.00001363059,0.0002600493,0.0003758404,0.00001472448,0.0002027984,0.00007101839,0.00001364888,0.00003685974,0.00028262,0.9928167,0.005868745],"study_design_scores_gemma":[0.0002748947,0.0001545371,0.005443549,0.009627839,0.0001096041,0.001637891,0.001457315,0.000168909,0.0002643795,0.003566938,0.977164,0.0001301607],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"commentary","genre_gemma":"commentary","genre_scores_codex":[0.0006048902,0.01394078,0.0001590678,0.8732466,0.1048932,0.00004647473,0.0007110566,0.00007399098,0.006323825],"genre_scores_gemma":[0.008604475,0.01889109,0.0004317093,0.7748447,0.1857445,0.00009009655,0.0003257178,0.00007210769,0.01099558],"genre_candidate":"commentary","genre_consensus":"commentary","teacher_disagreement_score":0.02326018,"threshold_uncertainty_score":0.07101989,"prediction_status":"machine_predicted_unvalidated"},"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"}]}