{"meta":{"page":1,"per_page":50,"max_per_page":100,"total":504,"total_is_capped":false,"direct_labels_cover":25,"predictions_cover":504,"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":"0259130d8859","filters":{"venue":"BMC Medical Research Methodology"}},"results":[{"id":"W2119790985","doi":"10.1186/1471-2288-7-10","title":"Development of AMSTAR: a measurement tool to assess the methodological quality of systematic reviews","year":2007,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":4413,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; Institute of Population and Public Health","funders":"","keywords":"Quality (philosophy); Computer science; Data science; Systematic review; MEDLINE; Medicine; Medical physics; Management science; Political science; Engineering","authors":[{"name":"Beverley Shea","is_ca":false},{"name":"Jeremy Grimshaw","is_ca":true},{"name":"George A. Wells","is_ca":true},{"name":"Maarten Boers","is_ca":false},{"name":"Neil Andersson","is_ca":false},{"name":"Candyce Hamel","is_ca":false},{"name":"Ashley C Porter","is_ca":false},{"name":"Peter Tugwell","is_ca":true},{"name":"David Moher","is_ca":true},{"name":"L.M. Bouter","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9972372497208365,"gpt":0.7946501814039316,"spread":0.2025870683169049,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.9831024,0.0003631593,0.01156464,0.0009208298,0.0002358765,0.0001044282,0.006537694,0.0003652178,0.007331577],"category_scores_gemma":[0.9800288,0.0001298421,0.002067931,0.003973815,0.0006883178,0.00006606116,0.001184445,0.0008388506,0.0008237884],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001846127,"about_ca_system_score_gemma":0.001768605,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005456656,"about_ca_topic_score_gemma":0.001265656,"domain_scores_codex":[0.02236548,0.8683776,0.04720718,0.001963682,0.0586547,0.001431375],"domain_scores_gemma":[0.02987111,0.934719,0.01164705,0.009319601,0.01315278,0.00129049],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001084479,0.001211233,0.01437703,0.09481447,0.00186895,0.00003570355,0.01437782,0.00003007017,0.04862365,0.1347882,0.03693414,0.6518543],"study_design_scores_gemma":[0.003355857,0.001841166,0.2419862,0.02625168,0.001358438,0.0001844548,0.05963894,0.002202972,0.03767365,0.08862063,0.5344858,0.002400221],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05770337,0.003414743,0.9320829,0.000660737,0.0002818461,0.004320981,0.00000231663,0.00000435282,0.001528764],"genre_scores_gemma":[0.04606115,0.0001079321,0.9521921,0.0003619511,0.0001009053,0.0004905964,0.000001171342,0.00001258076,0.0006715908],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.6494541,"threshold_uncertainty_score":0.9999542,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2170182466","doi":"10.1186/1471-2288-10-1","title":"A tutorial on pilot studies: the what, why and how","year":2010,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3439,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"GlaxoSmithKline (Canada); St. Joseph’s Healthcare Hamilton; University of Waterloo; McMaster University","funders":"Canadian Institutes of Health Research","keywords":"Clinical trial; Medicine; Pilot trial; Psychological intervention; Intervention (counseling); Research design; Clinical study design; Medical physics; Randomized controlled trial; Nursing; Surgery","authors":[{"name":"Lehana Thabane","is_ca":true},{"name":"Jinhui Ma","is_ca":true},{"name":"Rong Chu","is_ca":true},{"name":"Ji Cheng","is_ca":true},{"name":"Afisi Ismaila","is_ca":true},{"name":"Lorena P Rios","is_ca":true},{"name":"Reid Robson","is_ca":true},{"name":"Marroon Thabane","is_ca":true},{"name":"Lora Giangregorio","is_ca":true},{"name":"Charles H. Goldsmith","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9876457482859581,"gpt":0.7589303971843917,"spread":0.2287153511015664,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","scholarly_communication","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8012421,0.0002380548,0.003086618,0.000493243,0.0004799431,0.001250541,0.00350099,0.0002551821,0.01150193],"category_scores_gemma":[0.9487209,0.00008060721,0.0005846202,0.00146357,0.00207095,0.0002074963,0.000894886,0.001923607,0.001228608],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002162896,"about_ca_system_score_gemma":0.0005655257,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003513182,"about_ca_topic_score_gemma":0.001710221,"domain_scores_codex":[0.5185942,0.4432422,0.005348859,0.002221257,0.02940331,0.001190188],"domain_scores_gemma":[0.07611386,0.9117685,0.00144746,0.0060079,0.003592475,0.001069792],"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":[0.0001753934,0.00009438108,0.001215308,0.00009053646,0.0002861148,0.0000352605,0.001500127,7.308905e-7,0.0006600718,0.09040518,0.6395152,0.2660216],"study_design_scores_gemma":[0.0004820123,0.0006071592,0.001090166,0.0000617468,0.00005090133,0.00005450187,0.007814477,0.001069928,0.00005627709,0.09073941,0.8978522,0.000121207],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4419507,0.01682964,0.1764257,0.3303159,0.02127584,0.004235539,0.000009428359,0.00003100202,0.008926184],"genre_scores_gemma":[0.505003,0.0132151,0.3468096,0.02362221,0.0226113,0.00195187,0.000008575037,0.0001343433,0.08664399],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4685263,"threshold_uncertainty_score":0.9997863,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1501619595","doi":"10.1186/1471-2288-1-2","title":"The CONSORT statement: revised recommendations for improving the quality of reports of parallel group randomized trials","year":2001,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2754,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Canadian Wood Council; University of Ottawa","funders":"U.S. National Library of Medicine; Wellcome Trust; Abbott Laboratories","keywords":"Consolidated Standards of Reporting Trials; Checklist; Randomized controlled trial; Statement (logic); Transparency (behavior); Quality (philosophy); Reliability (semiconductor); Medicine; Computer science; Medical education; Family medicine; Psychology; Surgery","authors":[{"name":"David Moher","is_ca":true},{"name":"Kenneth F. Schulz","is_ca":false},{"name":"Douglas G. Altman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9839379047069265,"gpt":0.769898861168967,"spread":0.2140390435379596,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9782478,0.0001895873,0.008886483,0.0003238675,0.0004600416,0.0001645276,0.002474597,0.0001817385,0.008089536],"category_scores_gemma":[0.9880641,0.00005999133,0.003389474,0.001324228,0.001783862,0.00007769968,0.0003918142,0.0004440421,0.00002406232],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003073634,"about_ca_system_score_gemma":0.0009830255,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003250282,"about_ca_topic_score_gemma":0.0004650566,"domain_scores_codex":[0.0351998,0.9057769,0.04270213,0.001184715,0.01436607,0.0007703762],"domain_scores_gemma":[0.02177028,0.9534449,0.01583074,0.004206349,0.004446931,0.0003007964],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.02979126,0.0002891961,0.003082211,0.0008202022,0.001742704,0.000009194719,0.001397804,0.00001977885,0.0009844962,0.205679,0.1591701,0.597014],"study_design_scores_gemma":[0.0367435,0.0002979038,0.0009219382,0.0001710109,0.0006968952,0.00004449593,0.007478937,0.01022496,0.0001773399,0.4736927,0.4693251,0.0002252467],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.008455303,0.003787341,0.9641985,0.01666974,0.0004485205,0.005545162,0.00001611295,0.00000308332,0.0008762655],"genre_scores_gemma":[0.1395776,0.008250766,0.8325743,0.0007242201,0.0005403594,0.004914592,0.00005509214,0.00003882534,0.01332427],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5967888,"threshold_uncertainty_score":0.9928172,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2166025830","doi":"10.1186/1471-2288-14-45","title":"Newcastle-Ottawa Scale: comparing reviewers’ to authors’ assessments","year":2014,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":2472,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"McMaster University; University of Toronto","funders":"Hamilton Health Sciences Foundation; McMaster University; Hamilton Health Sciences","keywords":"Interquartile range; Confidence interval; Medicine; Observational study; Systematic review; MEDLINE; Reliability (semiconductor); Cohort study; Family medicine; Internal medicine","authors":[{"name":"Carson K. L. Lo","is_ca":true},{"name":"Dominik Mertz","is_ca":true},{"name":"Mark Loeb","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.984244127367729,"gpt":0.7712171098175411,"spread":0.2130270175501879,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8487659,0.0003897195,0.006733761,0.001382158,0.0003945083,0.0006685706,0.006748289,0.0004245199,0.04904075],"category_scores_gemma":[0.8986087,0.0001971871,0.001465566,0.004284746,0.0004945977,0.0001624495,0.001643817,0.001436197,0.02191382],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001090894,"about_ca_system_score_gemma":0.0007836511,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001419544,"about_ca_topic_score_gemma":0.001211377,"domain_scores_codex":[0.2896103,0.6438476,0.01675395,0.003595646,0.0439964,0.002196053],"domain_scores_gemma":[0.2656914,0.6959743,0.004771413,0.0166615,0.009566881,0.007334445],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00003392752,0.00008682355,0.05671593,0.0002103694,0.00009196789,0.00001305062,0.0003040227,0.00002856986,0.0001374581,0.01550737,0.5285709,0.3982996],"study_design_scores_gemma":[0.0003462101,0.0002385864,0.0222247,0.0002181114,0.00005794781,0.00003920779,0.0005078345,0.02558428,0.00003401432,0.01765667,0.9328642,0.0002281973],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02990034,0.0006254403,0.9165124,0.005103908,0.0013103,0.001120741,0.000002660965,0.00001583626,0.04540832],"genre_scores_gemma":[0.1036107,0.000120022,0.8489254,0.0027463,0.001094203,0.0003150711,0.00001045939,0.00004577285,0.043132],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4042933,"threshold_uncertainty_score":0.9986257,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2261406464","doi":"10.1186/s12874-016-0116-4","title":"A scoping review on the conduct and reporting of scoping reviews","year":2016,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":2221,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Canada Research Chairs; University of Ottawa; University of Toronto; University of Calgary; St. Michael's Hospital","funders":"Canadian Institutes of Health Research; Alberta Innovates","keywords":"Systematic review; Inclusion (mineral); MEDLINE; Protocol (science); Grey literature; Medicine; Psychology; Alternative medicine; Political science","authors":[{"name":"Andrea C. Tricco","is_ca":true},{"name":"Erin Lillie","is_ca":true},{"name":"Wasifa Zarin","is_ca":true},{"name":"Kelly K. O’Brien","is_ca":true},{"name":"Heather Colquhoun","is_ca":true},{"name":"Monika Kastner","is_ca":true},{"name":"Danielle Levac","is_ca":true},{"name":"Carmen Ng","is_ca":true},{"name":"Jane Pearson Sharpe","is_ca":true},{"name":"Katherine A. Wilson","is_ca":true},{"name":"Meghan Kenny","is_ca":true},{"name":"Rachel Warren","is_ca":true},{"name":"Charlotte Wilson","is_ca":true},{"name":"Henry T. Stelfox","is_ca":true},{"name":"Sharon E. Straus","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9964033037350372,"gpt":0.8939342699261918,"spread":0.1024690338088453,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.5448813,0.0003725159,0.006587156,0.0005345605,0.001014321,0.000009207275,0.001278057,0.000712808,0.006837802],"category_scores_gemma":[0.9262681,0.0001684605,0.0003527406,0.001391304,0.001562202,0.00007123606,0.001218388,0.003830394,0.0006446354],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002781005,"about_ca_system_score_gemma":0.03206723,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000147568,"about_ca_topic_score_gemma":0.0004193106,"domain_scores_codex":[0.6710017,0.3094722,0.01336077,0.001136045,0.002960646,0.002068574],"domain_scores_gemma":[0.1493023,0.8320417,0.01401281,0.002164019,0.001077277,0.001401895],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000003518342,0.000003135232,0.000006215859,0.485946,0.000007546409,0.00000468953,0.0001548088,3.319545e-10,4.111964e-7,0.001469491,0.007933888,0.5044703],"study_design_scores_gemma":[0.00008807025,0.00006526556,0.000001071421,0.5482076,0.00002365738,0.00001923417,0.0000615064,2.283199e-7,3.578109e-7,0.0001594431,0.4512994,0.00007416779],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000002029032,0.9693027,0.002722281,0.0103427,0.0004803989,0.01549983,0.0000150259,0.00002329352,0.001611761],"genre_scores_gemma":[5.268388e-8,0.974946,0.008829672,0.01018085,0.0006008005,0.004963511,0.000004458007,0.00004229253,0.0004324118],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5225694,"threshold_uncertainty_score":0.9984678,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2117460543","doi":"10.1186/1471-2288-13-48","title":"Enhancing the scoping study methodology: a large, inter-professional team’s experience with Arksey and O’Malley’s framework","year":2013,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":1930,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Royal Inland Hospital; University of Victoria; Interior Health; BC Cancer Agency","funders":"Canadian Institutes of Health Research","keywords":"Psychology; Knowledge management; Applied psychology; Computer science","authors":[{"name":"Helena ML Daudt","is_ca":true},{"name":"Catherine van Mossel","is_ca":true},{"name":"Samantha Scott","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9073939510957548,"gpt":0.7921934918806057,"spread":0.1152004592151491,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1655313,0.0002738498,0.0008531199,0.0004476966,0.00313671,0.0000469404,0.001389332,0.0005688752,0.01050453],"category_scores_gemma":[0.4078687,0.0001532272,0.000046799,0.001237975,0.001962468,0.0002889234,0.002097035,0.005140403,0.0004349164],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002026835,"about_ca_system_score_gemma":0.006866187,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003583445,"about_ca_topic_score_gemma":0.01563079,"domain_scores_codex":[0.8433324,0.1486631,0.001609229,0.001022693,0.002725059,0.002647539],"domain_scores_gemma":[0.5633159,0.4327731,0.0004064211,0.001021126,0.001033111,0.001450358],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0007432734,0.0003671758,0.4661534,0.001627152,0.00009650836,0.00005054507,0.472397,0.000002079824,0.00145717,0.02319125,0.01382786,0.02008659],"study_design_scores_gemma":[0.00335416,0.001574926,0.1709522,0.003738327,0.00002071045,0.0000603088,0.7966134,0.000730365,0.0002181861,0.01564071,0.00665933,0.0004373685],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.5213194,0.0000897161,0.453915,0.01979226,0.0007143077,0.003668499,0.000003198898,0.00006018837,0.0004374648],"genre_scores_gemma":[0.4700383,0.000149755,0.4895842,0.02887387,0.0009204802,0.008797432,0.000002535972,0.00005759289,0.001575838],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3242164,"threshold_uncertainty_score":0.998764,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2156741466","doi":"10.1186/1471-2288-10-22","title":"The COSMIN checklist for evaluating the methodological quality of studies on measurement properties: A clarification of its content","year":2010,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":1844,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"Vrije Universiteit Amsterdam","keywords":"Checklist; Delphi method; Content validity; Relevance (law); Construct validity; Psychology; Quality (philosophy); Criterion validity; Delphi; Applied psychology; Formative assessment; Management science; Computer science; Psychometrics; Clinical psychology; Cognitive psychology; Engineering; Artificial intelligence","authors":[{"name":"Lidwine B. Mokkink","is_ca":false},{"name":"Caroline B. Terwee","is_ca":false},{"name":"Dirk L. Knol","is_ca":false},{"name":"Paul W. Stratford","is_ca":true},{"name":"Jordi Alonso","is_ca":false},{"name":"Donald L. Patrick","is_ca":false},{"name":"L.M. Bouter","is_ca":false},{"name":"Henrica C. W. de Vet","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9906184913252002,"gpt":0.7661254438681674,"spread":0.2244930474570328,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6651222,0.0001304431,0.0006330361,0.0001420026,0.001204179,0.00003392896,0.002024672,0.0004450908,0.00006287645],"category_scores_gemma":[0.9190369,0.0000620433,0.0002000934,0.0006024768,0.008473732,0.00004303613,0.0004081533,0.001745567,0.000004652288],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002066778,"about_ca_system_score_gemma":0.003381269,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001246318,"about_ca_topic_score_gemma":0.009481233,"domain_scores_codex":[0.7449377,0.234286,0.00268034,0.0009599031,0.01542789,0.001708102],"domain_scores_gemma":[0.4556108,0.5265525,0.000948579,0.001510851,0.01488066,0.0004966722],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.002485144,0.0002678237,0.0003040307,0.0006604718,0.0001168815,6.53098e-7,0.008258435,0.000004279612,0.3803614,0.5084951,0.001112321,0.09793344],"study_design_scores_gemma":[0.00294542,0.006691005,0.01040142,0.00118019,0.00008348981,0.000005279678,0.2185767,0.003485611,0.5130468,0.2234435,0.01961529,0.0005253004],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7289357,0.007512385,0.1478105,0.09889955,0.001855725,0.01184869,0.00001634076,0.0001188472,0.003002283],"genre_scores_gemma":[0.9127864,0.001158873,0.08253819,0.0001406852,0.0003113193,0.002481819,7.982301e-7,0.00001691813,0.0005649896],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2922665,"threshold_uncertainty_score":0.9942246,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2089198100","doi":"10.1186/1471-2288-8-70","title":"Reducing bias through directed acyclic graphs","year":2008,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":1410,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University; Jewish General Hospital","funders":"McGill University Health Centre; McGill University","keywords":"Directed acyclic graph; Confounding; Causal inference; Selection bias; Computer science; Outcome (game theory); Econometrics; Simple (philosophy); Causality (physics); Statistics; Mathematics; Algorithm","authors":[{"name":"Ian Shrier","is_ca":true},{"name":"Robert W. Platt","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9122827611924192,"gpt":0.6583154278917019,"spread":0.2539673333007173,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.02556201,0.0002524107,0.0008368503,0.0004186552,0.0003921919,0.00001390556,0.0009846976,0.0006212333,0.001962332],"category_scores_gemma":[0.4471436,0.0002013016,0.0001693329,0.001285538,0.002179991,0.00018524,0.0005874144,0.002021387,0.0001116958],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001392263,"about_ca_system_score_gemma":0.00110775,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005361075,"about_ca_topic_score_gemma":0.0002615557,"domain_scores_codex":[0.9764681,0.01825425,0.0008217107,0.0007357324,0.00238709,0.00133309],"domain_scores_gemma":[0.8679871,0.1298494,0.0001531977,0.0009571065,0.000520641,0.0005325641],"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":[0.0003792783,0.0005769926,0.003275793,0.0004784398,0.0001194597,0.0008872838,0.004181597,0.000002167448,0.01398497,0.8552887,0.07480704,0.04601822],"study_design_scores_gemma":[0.0004231975,0.0002794605,0.000663108,0.0001834521,0.000009799307,0.0003634564,0.000269678,0.000240079,0.023843,0.9677824,0.00571103,0.0002313476],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.2453303,0.0004719148,0.7387174,0.0009721924,0.0002521286,0.0006140576,0.000003739343,0.001254038,0.01238432],"genre_scores_gemma":[0.05889649,0.001936481,0.9370844,0.0002279868,0.0002520871,0.0002443778,0.000007158642,0.00006499829,0.001285967],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4215816,"threshold_uncertainty_score":0.99895,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1967747089","doi":"10.1186/s12874-015-0016-z","title":"Exploring physician specialist response rates to web-based surveys","year":2015,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":1003,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Calgary","funders":"Canadian Institutes of Health Research; Alberta Innovates; Alberta Innovates - Health Solutions","keywords":"Family medicine; Medicine; Health care; Specialty; Survey data collection; Descriptive statistics; Payment; Medical education; Business","authors":[{"name":"Ceara Cunningham","is_ca":true},{"name":"Hude Quan","is_ca":true},{"name":"Brenda R. Hemmelgarn","is_ca":true},{"name":"Tom Noseworthy","is_ca":true},{"name":"Cynthia A Beck","is_ca":true},{"name":"Elijah Dixon","is_ca":true},{"name":"Susan Samuel","is_ca":true},{"name":"William A. Ghali","is_ca":true},{"name":"Lindsay Sykes","is_ca":true},{"name":"Nathalie Jetté","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9375642504164767,"gpt":0.6565429842045365,"spread":0.2810212662119402,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.7980005,0.0002449702,0.0007847864,0.000968045,0.0008466959,0.00008077112,0.001517892,0.0005417675,0.001206888],"category_scores_gemma":[0.8998657,0.0002178366,0.0001672702,0.002661135,0.002942242,0.0002436241,0.0003966234,0.001317416,0.00116384],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004228243,"about_ca_system_score_gemma":0.01449273,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003935582,"about_ca_topic_score_gemma":0.03016128,"domain_scores_codex":[0.2386784,0.7545015,0.0005917035,0.0008495168,0.00353311,0.001845809],"domain_scores_gemma":[0.5405221,0.4557052,0.00007210961,0.0005488195,0.0009627598,0.002188991],"domain_codex":"methods","domain_gemma":null,"domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.2619004,0.001601682,0.08562551,0.0001632912,0.0002147649,0.001763389,0.06847332,0.0002224822,0.01402736,0.04173947,0.1160446,0.4082237],"study_design_scores_gemma":[0.00354267,0.001554862,0.3147812,0.0001583832,0.00001948203,0.000006135834,0.02378244,0.0002429022,0.007812456,0.008577905,0.6387432,0.0007782923],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8400149,0.0001831244,0.1325699,0.01665297,0.002544319,0.0006179377,0.0000141932,0.000193579,0.007209104],"genre_scores_gemma":[0.6539276,0.0003216762,0.3186797,0.007193448,0.008239619,0.001351137,0.00003524436,0.0001714158,0.01008017],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5226986,"threshold_uncertainty_score":0.9997712,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2095437242","doi":"10.1186/1471-2288-13-92","title":"A tutorial on sensitivity analyses in clinical trials: the what, why, when and how","year":2013,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Statistical Methods in Clinical Trials","field":"Mathematics","cited_by":901,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Simon Fraser University; University of Waterloo; Toronto General Hospital; Hamilton Health Sciences; University of Ottawa; GlaxoSmithKline (Canada); St. Joseph’s Healthcare Hamilton; McMaster University; Population Health Research Institute","funders":"","keywords":"Sensitivity (control systems); Clinical trial; Missing data; Computer science; Protocol (science); Outlier; Data science; Robustness (evolution); Research design; Data mining; Medical physics; Medicine; Management science; Alternative medicine; Machine learning; Artificial intelligence; Statistics; Mathematics","authors":[{"name":"Lehana Thabane","is_ca":true},{"name":"Lawrence Mbuagbaw","is_ca":true},{"name":"Shiyuan Zhang","is_ca":true},{"name":"Zainab Samaan","is_ca":true},{"name":"Maura Marcucci","is_ca":true},{"name":"Chenglin Ye","is_ca":true},{"name":"Marroon Thabane","is_ca":true},{"name":"Lora Giangregorio","is_ca":true},{"name":"Brittany B. Dennis","is_ca":true},{"name":"Sarah Daisy Kosa","is_ca":true},{"name":"Victoria Borg Debono","is_ca":true},{"name":"Rejane Dillenburg","is_ca":true},{"name":"Vincent Fruci","is_ca":true},{"name":"Monica Bawor","is_ca":true},{"name":"Juneyoung Lee","is_ca":false},{"name":"George A. Wells","is_ca":true},{"name":"Charles H. Goldsmith","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9873875398748938,"gpt":0.8105356203029017,"spread":0.1768519195719921,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6132049,0.0003171879,0.003057324,0.0003498132,0.0001902024,0.0003357084,0.0007733087,0.001216768,0.002259343],"category_scores_gemma":[0.9914974,0.0001743535,0.000452038,0.000586948,0.003619401,0.0001995688,0.0008197997,0.00412223,0.0001003354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008199604,"about_ca_system_score_gemma":0.0008632043,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005576465,"about_ca_topic_score_gemma":0.0009740432,"domain_scores_codex":[0.4428463,0.5486633,0.002810439,0.001200512,0.003222475,0.001257042],"domain_scores_gemma":[0.01351297,0.9832157,0.0004922445,0.001176087,0.0005657029,0.001037283],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002007049,0.0006758656,0.001062465,0.0002378381,0.0002595608,0.0002008088,0.0004191184,6.944042e-7,0.000443726,0.1533479,0.09015042,0.7511945],"study_design_scores_gemma":[0.002795205,0.0007395013,0.003342575,0.0003010958,0.00007178306,0.00002331985,0.001146674,0.002818521,0.0001748656,0.9853399,0.003051525,0.0001950714],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.08634765,0.0008052104,0.8229362,0.07917384,0.005781418,0.004025973,0.00002153574,0.0001135563,0.0007946452],"genre_scores_gemma":[0.01127589,0.001418719,0.9789947,0.002280421,0.005199804,0.0004293915,0.000001545456,0.00005411499,0.0003453895],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8319919,"threshold_uncertainty_score":0.9990922,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1602160603","doi":"10.1186/1471-2288-14-137","title":"Modern modelling techniques are data hungry: a simulation study for predicting dichotomous endpoints","year":2014,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Traumatic Brain Injury and Neurovascular Disturbances","field":"Medicine","cited_by":796,"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":"Support vector machine; Logistic regression; Statistics; Random forest; Sample size determination; Artificial intelligence; Receiver operating characteristic; Regression; Machine learning; Computer science; Medicine; Mathematics","authors":[{"name":"Tjeerd van der Ploeg","is_ca":false},{"name":"Peter C. Austin","is_ca":true},{"name":"Ewout W. Steyerberg","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7132462483034406,"gpt":0.5762959983442477,"spread":0.1369502499591929,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05033034,0.0002117726,0.0008361284,0.0003191176,0.0003105988,0.00003474902,0.0008635205,0.0003412737,0.00009315325],"category_scores_gemma":[0.1315216,0.0001680368,0.0001290141,0.0003367706,0.0003834299,0.0001439085,0.000508283,0.001044426,0.000008594636],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005162588,"about_ca_system_score_gemma":0.0004131757,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009384075,"about_ca_topic_score_gemma":0.0002157569,"domain_scores_codex":[0.986476,0.008905104,0.0007256467,0.001070129,0.002063147,0.0007599543],"domain_scores_gemma":[0.9360024,0.06087084,0.0001916165,0.00185032,0.0004950863,0.0005896917],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.007146961,0.003970191,0.05061894,0.003897286,0.000752394,0.0001600463,0.005250841,0.004778277,0.001196188,0.001731724,0.001587665,0.9189095],"study_design_scores_gemma":[0.002194774,0.001576476,0.002516148,0.0003666143,0.0001263259,0.00004128459,0.0009037263,0.9807326,0.000739442,0.007844724,0.002802164,0.0001557004],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1651945,0.0003182104,0.8311675,0.0004132254,0.00015155,0.002180387,0.00001876243,0.0001900678,0.000365739],"genre_scores_gemma":[0.830259,0.00004966712,0.1680567,0.0002576195,0.0007598153,0.0003779563,0.00008288941,0.00004800345,0.0001083922],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.9759544,"threshold_uncertainty_score":0.9778848,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3200920720","doi":"10.1186/s12874-021-01381-z","title":"Conducting proportional meta-analysis in different types of systematic reviews: a guide for synthesisers of evidence","year":2021,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":724,"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":"Meta-analysis; Pairwise comparison; Systematic review; Computer science; Data science; Management science; Psychology; MEDLINE; Medicine; Artificial intelligence; Engineering; Political science","authors":[{"name":"Timothy Hugh Barker","is_ca":false},{"name":"Celina Borges Migliavaca","is_ca":false},{"name":"Cinara Stein","is_ca":false},{"name":"Verônica Colpani","is_ca":false},{"name":"Maicon Falavigna","is_ca":true},{"name":"Edoardo Aromataris","is_ca":false},{"name":"Zachary Munn","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9978106427928172,"gpt":0.7994635274499049,"spread":0.1983471153429123,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad"],"category_scores_codex":[0.8533682,0.0008094867,0.08917446,0.004460998,0.00009497116,0.0001471527,0.006250582,0.0008293644,0.02531232],"category_scores_gemma":[0.9775119,0.0002822109,0.03550541,0.00956363,0.0007681514,0.0001225456,0.0008921716,0.001071168,0.0001335394],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001728641,"about_ca_system_score_gemma":0.003886023,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009117454,"about_ca_topic_score_gemma":0.0005776405,"domain_scores_codex":[0.1836095,0.7258913,0.06087806,0.003220403,0.02534799,0.001052742],"domain_scores_gemma":[0.0408083,0.9149777,0.02761121,0.008050778,0.00792683,0.000625237],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"meta_analysis","study_design_scores_codex":[0.00001374581,0.0001974814,0.00007194315,0.7864833,0.09223808,0.00001927777,0.0001685551,0.00001417874,0.000002994083,0.003277746,0.004730992,0.1127817],"study_design_scores_gemma":[0.0001664566,0.0001851488,0.000009221102,0.1311314,0.5808495,0.00006354242,0.001132489,0.001281791,0.00001285488,0.002366411,0.2822391,0.0005620873],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000004974898,0.8643436,0.1278754,0.0001980248,0.0001351037,0.00730839,0.00004482745,0.00000202593,0.00008765139],"genre_scores_gemma":[0.00001403766,0.889511,0.1021099,0.00002149985,0.00009419247,0.004620028,0.00002720051,0.00002945091,0.003572725],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.6553519,"threshold_uncertainty_score":0.999963,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1603573412","doi":"10.1186/1471-2288-6-57","title":"Dealing with missing data in a multi-question depression scale: a comparison of imputation methods","year":2006,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":655,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University; University of Calgary","funders":"Canadian Institutes of Health Research; Fondation pour la Recherche Médicale; Government of Canada; Heart and Stroke Foundation of Canada","keywords":"Missing data; Statistics; Imputation (statistics); Statistic; Cohen's kappa; Standard deviation; Population; Mathematics; Kappa; Regression; Psychology; Medicine","authors":[{"name":"Fiona M. Shrive","is_ca":true},{"name":"Heather Stuart","is_ca":true},{"name":"Hude Quan","is_ca":true},{"name":"William A. Ghali","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7482629066797014,"gpt":0.6934678501720167,"spread":0.05479505650768468,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0614525,0.0001637932,0.0008105234,0.0003968479,0.0001092726,0.00002513565,0.0006524046,0.0003501677,0.0001744554],"category_scores_gemma":[0.1861279,0.0001195027,0.00003412448,0.0006304182,0.0006709506,0.0001025035,0.000420532,0.0009452135,0.000002316255],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006289201,"about_ca_system_score_gemma":0.0005451152,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001733915,"about_ca_topic_score_gemma":0.002781688,"domain_scores_codex":[0.963995,0.03192572,0.001175436,0.0007200871,0.001515305,0.000668406],"domain_scores_gemma":[0.8646073,0.1338,0.0002662509,0.000711589,0.0003657909,0.0002490053],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0004629173,0.0004696494,0.05056434,0.0006848171,0.000013419,0.00003055239,0.000299217,0.00002095181,0.006642275,0.1133977,0.00008435826,0.8273298],"study_design_scores_gemma":[0.001317865,0.000220393,0.02561892,0.001033324,0.00003189185,0.00003639409,0.000371057,0.3448774,0.008535251,0.6177282,0.00005471434,0.000174603],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01445299,0.0003283146,0.984243,0.0001799353,0.0000631984,0.0003485667,0.00001044914,0.00003486113,0.0003386263],"genre_scores_gemma":[0.05135386,0.0000180745,0.9484555,0.00001181389,0.00006781305,0.00002957707,0.00002623367,0.00002443223,0.00001274977],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8271552,"threshold_uncertainty_score":0.9664322,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2924517740","doi":"10.1186/s12874-019-0666-3","title":"A review of spline function procedures in R","year":2019,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Data Analysis with R","field":"Computer Science","cited_by":577,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University Health Centre","funders":"","keywords":"Univariate; Computer science; Spline (mechanical); Observational study; Software; Data science; Regression analysis; Machine learning; Thin plate spline; Data mining; Management science; Statistics; Mathematics; Multivariate statistics; Spline interpolation; Engineering","authors":[{"name":"Aris Perperoglou","is_ca":false},{"name":"Willi Sauerbrei","is_ca":false},{"name":"Michał Abrahamowicz","is_ca":true},{"name":"Matthias Schmid","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7468500022474139,"gpt":0.6371901037000626,"spread":0.1096598985473514,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.09173721,0.0003259211,0.004242971,0.001617175,0.00003380486,0.00002583585,0.0048497,0.000734403,0.0006804765],"category_scores_gemma":[0.2315358,0.0002219882,0.0005577196,0.004455541,0.0003957652,0.0001675281,0.002303041,0.00219506,0.0005031926],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001552832,"about_ca_system_score_gemma":0.01162722,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001373875,"about_ca_topic_score_gemma":0.0004306732,"domain_scores_codex":[0.9535694,0.03928413,0.002014477,0.001297405,0.002934546,0.000900034],"domain_scores_gemma":[0.9306318,0.06472965,0.0007106696,0.002796164,0.0006402386,0.0004915264],"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":[0.000005397421,0.00005497987,0.000003091493,0.2989516,0.00004435897,0.00001637976,0.000006368865,1.076285e-7,9.702676e-8,0.004603414,0.007411087,0.6889031],"study_design_scores_gemma":[0.0001455988,0.0001498329,0.000004700388,0.1899373,0.0001193777,0.00005897457,0.00000305319,0.0003097339,2.382251e-7,0.000776206,0.8083482,0.0001468629],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[4.312882e-8,0.7334698,0.2648272,0.0002228065,0.0002411561,0.0008904077,0.000005431768,0.00002182908,0.0003213306],"genre_scores_gemma":[3.952408e-8,0.9056979,0.09307776,0.0003734938,0.0001549303,0.0002719434,0.00006048512,0.00002247025,0.0003409242],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.8009371,"threshold_uncertainty_score":0.9939759,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2966928705","doi":"10.1186/s12874-019-0815-8","title":"Consolidated criteria for strengthening reporting of health research involving indigenous peoples: the CONSIDER statement","year":2019,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Indigenous Health, Education, and Rights","field":"Social Sciences","cited_by":532,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true},"ca_institutions":"","funders":"Health Research Council of New Zealand","keywords":"Indigenous; Statement (logic); MEDLINE; Medicine; Family medicine; Environmental health; Political science; Psychology","authors":[{"name":"Tania Huria","is_ca":false},{"name":"Suetonia C. Palmer","is_ca":false},{"name":"Suzanne Pitama","is_ca":false},{"name":"Lutz Beckert","is_ca":false},{"name":"Cameron Lacey","is_ca":false},{"name":"Shaun Ewen","is_ca":false},{"name":"Linda Tuhiwai Smith","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6896195728371073,"gpt":0.6408542495531911,"spread":0.04876532328391614,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.4242062,0.0001363852,0.0007747364,0.0005124859,0.008128214,0.00007840006,0.0008829001,0.0003581066,0.001385876],"category_scores_gemma":[0.09743724,0.00009065497,0.0001066136,0.0009807331,0.002583469,0.0001083944,0.00004002639,0.001239721,0.00003474554],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005774674,"about_ca_system_score_gemma":0.1710617,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.2367953,"about_ca_topic_score_gemma":0.6835849,"domain_scores_codex":[0.895233,0.09329699,0.003390939,0.0007680084,0.00424508,0.003066014],"domain_scores_gemma":[0.7459977,0.2452569,0.002023221,0.001015592,0.004669779,0.001036891],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001289997,0.000165837,0.004605585,0.001020326,0.00004592077,0.000005171004,0.8422758,0.000002324499,0.00008204166,0.1457519,0.001542574,0.004373524],"study_design_scores_gemma":[0.0009220032,0.001009073,0.002039333,0.0006434524,0.000009681687,0.00001755053,0.6451066,0.0001931398,0.0006727931,0.02432382,0.3248432,0.0002192865],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9760277,0.002187011,0.006331261,0.004580087,0.001157669,0.004250479,0.00001398906,0.00004377967,0.005408001],"genre_scores_gemma":[0.9059165,0.01845969,0.06705478,0.0003208594,0.001296035,0.0001930099,0.0000674847,0.00006586416,0.006625783],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4467896,"threshold_uncertainty_score":0.999527,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2120673540","doi":"10.1186/1471-2288-3-28","title":"Pragmatic controlled clinical trials in primary care: the struggle between external and internal validity","year":2003,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":506,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Queen's University","funders":"Heart and Stroke Foundation of Canada","keywords":"External validity; Internal validity; Generalizability theory; Blinding; Face validity; Psychological intervention; Clinical trial; Medicine; Randomized controlled trial; Research design; Psychology; Clinical psychology; Social psychology; Psychometrics; Statistics; Nursing; Surgery","authors":[{"name":"Marshall Godwin","is_ca":true},{"name":"Lucia Rühland","is_ca":true},{"name":"Ian Casson","is_ca":true},{"name":"Susan M. MacDonald","is_ca":true},{"name":"Dianne Delva","is_ca":true},{"name":"Richard Birtwhistle","is_ca":true},{"name":"Miu Lam","is_ca":true},{"name":"Rachelle Seguin","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9759954308528304,"gpt":0.8363125390958985,"spread":0.1396828917569318,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.6864873,0.000164221,0.001984611,0.0003595988,0.0006996758,0.0000251801,0.0008204114,0.0005460328,0.002559784],"category_scores_gemma":[0.8694932,0.00009537255,0.0001618682,0.0005541526,0.00142946,0.0000957708,0.0004961639,0.004305752,0.0001345285],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003759357,"about_ca_system_score_gemma":0.01157581,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000986093,"about_ca_topic_score_gemma":0.002096973,"domain_scores_codex":[0.2989429,0.690536,0.005283813,0.0006787447,0.002785433,0.001773119],"domain_scores_gemma":[0.1389934,0.85769,0.0008612971,0.0005973944,0.0005147687,0.001343111],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009615776,0.0000601262,0.9114558,0.0009853925,0.0000370938,0.00002744771,0.00827485,3.851589e-7,0.00008030856,0.007381232,0.01008057,0.06065522],"study_design_scores_gemma":[0.03494706,0.0009535663,0.859389,0.0008407686,0.0000502418,0.00002579016,0.01344947,0.0002384014,0.00003973404,0.03096277,0.05884892,0.0002542589],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8045934,0.0004939417,0.1683726,0.01372248,0.001394377,0.00505317,0.00003417111,0.00003276069,0.006303099],"genre_scores_gemma":[0.8196382,0.001345941,0.143746,0.02631456,0.003869401,0.003157442,0.00001727279,0.00006433,0.001846806],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1830059,"threshold_uncertainty_score":0.998352,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1964304172","doi":"10.1186/1471-2288-8-1","title":"An administrative data merging solution for dealing with missing data in a clinical registry: adaptation from ICD-9 to ICD-10","year":2008,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Medical Coding and Health Information","field":"Health Professions","cited_by":503,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Provincial Health Services Authority; University of British Columbia; University of Alberta; University of Calgary","funders":"Canadian Institutes of Health Research; Eli Lilly Canada; Fondation pour la Recherche Médicale; W. Garfield Weston Foundation; Eli Lilly and Company; University of Alberta; Government of Canada; Boston Scientific Corporation","keywords":"Missing data; Statistic; Medicine; Data mining; ICD-10; Computer science; Statistics; Mathematics; Machine learning","authors":[{"name":"Danielle A. Southern","is_ca":true},{"name":"Colleen M. Norris","is_ca":true},{"name":"Hude Quan","is_ca":true},{"name":"Fiona M. Shrive","is_ca":true},{"name":"P. Diane Galbraith","is_ca":true},{"name":"Karin H. Humphries","is_ca":true},{"name":"Min Gao","is_ca":true},{"name":"Merril L. Knudtson","is_ca":true},{"name":"William A. Ghali","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9723130920307268,"gpt":0.759938338111273,"spread":0.2123747539194538,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.0977209,0.0001852499,0.000752144,0.0002916022,0.00137579,0.00001693701,0.001882079,0.0007870065,0.0008114364],"category_scores_gemma":[0.3154645,0.0001488872,0.00003029524,0.0004497306,0.0005912038,0.0004941663,0.0008171002,0.002728362,0.00008889619],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002055471,"about_ca_system_score_gemma":0.01136106,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.007311806,"about_ca_topic_score_gemma":0.02659618,"domain_scores_codex":[0.969012,0.02432347,0.002140774,0.001173342,0.001978754,0.001371714],"domain_scores_gemma":[0.8529254,0.1400566,0.0006356674,0.00310677,0.0008730555,0.002402494],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01884386,0.0006287436,0.03813229,0.002187704,0.0001147547,0.0002048792,0.02728314,0.0001032798,0.0002652163,0.006980167,0.06581361,0.8394424],"study_design_scores_gemma":[0.004616961,0.001775381,0.03351653,0.002074253,0.00003206243,0.00002108566,0.01018926,0.860641,0.00001139617,0.002047563,0.08474501,0.0003295165],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.04888191,0.0001944439,0.9358237,0.01215064,0.0004275734,0.001566561,0.0002311768,0.00007511709,0.0006488629],"genre_scores_gemma":[0.1115964,0.0004251165,0.8795316,0.002413274,0.001792592,0.0002665606,0.003752016,0.00003747765,0.0001850169],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8605377,"threshold_uncertainty_score":0.9999243,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2609185034","doi":"10.1186/s12874-017-0338-0","title":"Double-adjustment in propensity score matching analysis: choosing a threshold for considering residual imbalance","year":2017,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":491,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Hamilton Health Sciences; McMaster University; Population Health Research Institute","funders":"University of Oxford; Cancer Research UK","keywords":"Covariate; Propensity score matching; Confounding; Statistics; Residual; Matching (statistics); Logistic regression; Regression; Mathematics; Regression analysis; Econometrics; Medicine","authors":[{"name":"Tri‐Long Nguyen","is_ca":true},{"name":"Gary S. Collins","is_ca":false},{"name":"Jessica Spence","is_ca":true},{"name":"Jean‐Pierre Daurès","is_ca":false},{"name":"P.J. Devereaux","is_ca":true},{"name":"Paul Landais","is_ca":false},{"name":"Yannick Le Manach","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8896096151371321,"gpt":0.6436425630183503,"spread":0.2459670521187818,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04284102,0.0002378853,0.001141047,0.0005774823,0.000538384,0.0001185988,0.001216027,0.0004211147,0.0001274519],"category_scores_gemma":[0.1008581,0.0001967548,0.000162395,0.0003440469,0.0009582182,0.0002617346,0.001080059,0.00141144,0.000005351801],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003179928,"about_ca_system_score_gemma":0.0008763894,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0007958629,"about_ca_topic_score_gemma":0.01271188,"domain_scores_codex":[0.9930576,0.002807129,0.000785624,0.000777531,0.001409492,0.001162588],"domain_scores_gemma":[0.9699904,0.02756566,0.0003175496,0.00136684,0.0003975896,0.0003619612],"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":[0.003787212,0.0005952665,0.1492941,0.002255698,0.0007084965,0.0005073908,0.002677052,0.000159596,0.01300951,0.8007784,0.002010536,0.02421677],"study_design_scores_gemma":[0.00201424,0.0002091713,0.03084219,0.0006806706,0.0001047635,0.00002907747,0.0004129297,0.001847048,0.01263102,0.9508296,0.00009583295,0.0003034397],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.3953219,0.0001911004,0.6014486,0.0008138763,0.0001031759,0.001165744,0.000004344708,0.0001086268,0.0008426167],"genre_scores_gemma":[0.3769965,0.0001254387,0.6218768,0.00008545961,0.0002085438,0.0005257433,0.000005235831,0.00003314762,0.0001431428],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.1500512,"threshold_uncertainty_score":0.9855966,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2895713118","doi":"10.1186/s12874-018-0600-0","title":"Improving reporting of meta-ethnography: the eMERGe reporting guidance","year":2019,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":483,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"Norwegian Institute of Public Health; Medical Research Council; Economic and Social Research Council; University of North Carolina at Chapel Hill; University of Stirling; South African Medical Research Council; University of Nottingham; United Kingdom Clinical Research Collaboration; University of Southampton; National Institute for Health and Care Research; Cancer Research UK; British Heart Foundation; Wellcome Trust; McMaster University","keywords":"CLARITY; Ethnography; Audit; Bespoke; Best practice; Medical education; Medicine; Psychology; Management science; Sociology; Political science; Business; Engineering","authors":[{"name":"Emma F. France","is_ca":false},{"name":"Nicola Ring","is_ca":false},{"name":"Isabelle Uny","is_ca":false},{"name":"Edward Duncan","is_ca":false},{"name":"Ruth Jepson","is_ca":false},{"name":"Margaret Maxwell","is_ca":false},{"name":"Rachel J Roberts","is_ca":false},{"name":"Ruth Turley","is_ca":false},{"name":"Andrew Booth","is_ca":false},{"name":"Nicky Britten","is_ca":false},{"name":"Kate Flemming","is_ca":false},{"name":"Ian Gallagher","is_ca":false},{"name":"Ruth Garside","is_ca":false},{"name":"Karin Hannes","is_ca":false},{"name":"Simon Lewin","is_ca":false},{"name":"George W. Noblit","is_ca":false},{"name":"Catherine Pope","is_ca":false},{"name":"James Thomas","is_ca":false},{"name":"Meredith Vanstone","is_ca":true},{"name":"Gina Higginbottom","is_ca":false},{"name":"Jane Noyes","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9822940030074859,"gpt":0.7359007192625816,"spread":0.2463932837449043,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9437912,0.0003238856,0.008714945,0.0008985921,0.0003202945,0.000215892,0.00492158,0.0003388004,0.04164615],"category_scores_gemma":[0.9851087,0.000120492,0.005032954,0.004457667,0.0008250461,0.0001673288,0.001319606,0.001426987,0.0007499391],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002865936,"about_ca_system_score_gemma":0.001182967,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0006997962,"about_ca_topic_score_gemma":0.0004807955,"domain_scores_codex":[0.2027829,0.5915134,0.1356745,0.004954237,0.06229942,0.002775542],"domain_scores_gemma":[0.05309585,0.7513867,0.1594871,0.02012286,0.01462582,0.001281727],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0001265439,0.0001890913,0.5457725,0.001689771,0.004550548,0.0002157018,0.003225127,0.0002235774,0.01613643,0.02363802,0.05746113,0.3467716],"study_design_scores_gemma":[0.001991001,0.001340202,0.16033,0.0008724097,0.005192359,0.001275419,0.04091699,0.1655204,0.01102166,0.1806766,0.428959,0.00190406],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5345457,0.00926759,0.4337196,0.003645109,0.001176758,0.001867194,0.000003998811,0.00001702552,0.01575702],"genre_scores_gemma":[0.6052027,0.0001017417,0.3783942,0.0004598859,0.0003765712,0.0002704635,0.000003495724,0.00003868463,0.01515231],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3854425,"threshold_uncertainty_score":0.96392,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2123074327","doi":"10.1186/1471-2288-7-5","title":"Inclusion of zero total event trials in meta-analyses maintains analytic consistency and incorporates all available data","year":2007,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":472,"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; Hospital for Sick Children; Health Sciences Centre; Sunnybrook Health Science Centre; Muscular Dystrophy Canada; University of Toronto","funders":"","keywords":"Meta-analysis; Confidence interval; Random effects model; Statistics; Relative risk; Medicine; Odds ratio; Study heterogeneity; Absolute risk reduction; Randomized controlled trial; Event (particle physics); Clinical trial; Econometrics; Mathematics; Internal medicine","authors":[{"name":"Jan O. Friedrich","is_ca":true},{"name":"Neill K. J. Adhikari","is_ca":true},{"name":"Joseph Beyene","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9930425907223072,"gpt":0.7755257982900672,"spread":0.21751679243224,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9500071,0.000376785,0.01371203,0.002716155,0.000235234,0.0001748156,0.004637614,0.0004352777,0.04509151],"category_scores_gemma":[0.9448981,0.0001669393,0.001937838,0.003986129,0.001213217,0.0002516519,0.007190143,0.0008739422,0.0005012299],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008284482,"about_ca_system_score_gemma":0.001603615,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001267342,"about_ca_topic_score_gemma":0.005610886,"domain_scores_codex":[0.1013862,0.8085623,0.04648129,0.00472929,0.03675938,0.002081569],"domain_scores_gemma":[0.06045432,0.9159569,0.008687522,0.009934805,0.003464182,0.001502317],"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":[0.002984136,0.00302108,0.1508139,0.002780138,0.03651752,0.001770173,0.005010774,0.0004912521,0.02615376,0.08734281,0.4002405,0.282874],"study_design_scores_gemma":[0.005103305,0.001598482,0.02891817,0.0004310652,0.01763089,0.0004815657,0.009459401,0.1293938,0.003204909,0.7073147,0.09513179,0.001331862],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1321279,0.02164278,0.8353031,0.003502586,0.0002937292,0.002162159,0.00009259218,0.000007938281,0.004867213],"genre_scores_gemma":[0.6805575,0.0009533025,0.3104088,0.0004795082,0.0001582263,0.00006434876,0.0000789438,0.00002723853,0.007272139],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.6199719,"threshold_uncertainty_score":0.9557814,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2400420884","doi":"10.1186/s12874-016-0165-8","title":"Using the modified Delphi method to establish clinical consensus for the diagnosis and treatment of patients with rotator cuff pathology","year":2016,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":432,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Alberta; University of Calgary; Mount Royal University","funders":"","keywords":"Rotator cuff; Medicine; Delphi method; Guideline; Delphi; Health care; MEDLINE; Physical therapy; Medical physics; Pathology; Surgery; Computer science; Artificial intelligence","authors":[{"name":"Breda Eubank","is_ca":true},{"name":"Nicholas G. Mohtadi","is_ca":true},{"name":"Mark R. Lafave","is_ca":true},{"name":"J. Preston Wiley","is_ca":true},{"name":"Aaron J. Bois","is_ca":true},{"name":"Richard S. Boorman","is_ca":true},{"name":"David M Sheps","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.895516367431202,"gpt":0.7062424009973984,"spread":0.1892739664338036,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.05199214,0.0001764793,0.000825354,0.0001593512,0.000200608,0.00001251565,0.0002700492,0.0002806848,0.0001933221],"category_scores_gemma":[0.4101235,0.00006789191,0.0001505055,0.0003677942,0.001561736,0.00003845598,0.0002300368,0.000360149,0.000003500057],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001133267,"about_ca_system_score_gemma":0.001446721,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005788815,"about_ca_topic_score_gemma":0.0007895272,"domain_scores_codex":[0.977356,0.01869687,0.00150011,0.0006299434,0.001173224,0.0006439117],"domain_scores_gemma":[0.4536524,0.5431934,0.0003380071,0.0007281034,0.001466354,0.0006217304],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01213195,0.0007276636,0.1721254,0.00005748968,0.000284548,0.00003301507,0.0003449535,0.000008807569,0.0002715035,0.001530718,0.001757966,0.810726],"study_design_scores_gemma":[0.07609741,0.05009208,0.7603635,0.0006458074,0.002812667,0.0004587786,0.006928164,0.01364043,0.004273948,0.01071046,0.07326419,0.0007125289],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.482544,0.0001367784,0.4940093,0.02101498,0.0001470468,0.002090891,0.00004131539,0.000008724276,0.000007012386],"genre_scores_gemma":[0.2442636,0.0007543285,0.7512977,0.001989867,0.0005768029,0.0009991141,0.00001276311,0.00004434194,0.00006150173],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8100135,"threshold_uncertainty_score":0.9761736,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2134792407","doi":"10.1186/1471-2288-12-159","title":"Should researchers use single indicators, best indicators, or multiple indicators in structural equation models?","year":2012,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Psychometric Methodologies and Testing","field":"Decision Sciences","cited_by":422,"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":"Structural equation modeling; Latent variable; Path analysis (statistics); Econometrics; Computer science; Statistical model; Performance indicator; Confounding; Statistics; Risk analysis (engineering); Mathematics; Machine learning; Medicine; Economics","authors":[{"name":"Leslie A. Hayduk","is_ca":true},{"name":"Levente Littvay","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9615285550475937,"gpt":0.6606017935692375,"spread":0.3009267614783562,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","bibliometrics","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","bibliometrics","research_integrity"],"category_scores_codex":[0.2797406,0.0004901982,0.001607435,0.0191726,0.0005554737,0.0003809817,0.004226325,0.001549289,0.003893149],"category_scores_gemma":[0.9326814,0.0003258387,0.0002936981,0.02976135,0.002644595,0.001624047,0.002337086,0.004276934,0.000191823],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0006479849,"about_ca_system_score_gemma":0.002590837,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001530983,"about_ca_topic_score_gemma":0.001156232,"domain_scores_codex":[0.900068,0.07724222,0.002924924,0.001847331,0.01405228,0.003865181],"domain_scores_gemma":[0.1953667,0.7965987,0.001039673,0.002414043,0.0007218213,0.0038591],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0006298674,0.0002954572,0.6568246,0.00003779817,0.00003448917,0.00004976792,0.001497525,0.00007872343,0.0001791988,0.005745786,0.00139521,0.3332315],"study_design_scores_gemma":[0.007153624,0.002405596,0.751096,0.0004744467,0.00005748705,0.0002498536,0.02222594,0.06022914,0.002670245,0.1026196,0.04893985,0.001878299],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8312503,0.001724994,0.1612789,0.001047739,0.001303451,0.0009707824,0.00002700877,0.0001106884,0.002286045],"genre_scores_gemma":[0.7845387,0.0003049903,0.2131133,0.0002796279,0.0006282906,0.0001609766,0.00002340782,0.00006083049,0.0008898939],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7193565,"threshold_uncertainty_score":0.9999194,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2544651403","doi":"10.1186/s12874-016-0247-7","title":"Why sex and gender matter in implementation research","year":2016,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Sex and Gender in Healthcare","field":"Medicine","cited_by":408,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"British Columbia Centre of Excellence for Women's Health; Institute of Gender and Health; University of Ottawa; Canadian Institutes of Health Research","funders":"Canadian Institutes of Health Research; Bangor University; Institute of Gender and Health; Ottawa Hospital Research Institute","keywords":"Psychology; Medicine; Computer science","authors":[{"name":"Cara Tannenbaum","is_ca":true},{"name":"Lorraine Greaves","is_ca":true},{"name":"Ian D. Graham","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8680714952997681,"gpt":0.7037639602905658,"spread":0.1643075350092024,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06868602,0.0001120827,0.000395217,0.0008643519,0.0001331255,0.00001268698,0.0002178283,0.0003780824,0.01032978],"category_scores_gemma":[0.02890233,0.00006993891,0.0000396432,0.0005906885,0.0009210501,0.00005779787,0.0003456673,0.001143008,0.0002398858],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002003519,"about_ca_system_score_gemma":0.001310533,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001750237,"about_ca_topic_score_gemma":0.002522404,"domain_scores_codex":[0.9723811,0.02266443,0.0005047311,0.0007313234,0.002426223,0.00129223],"domain_scores_gemma":[0.958399,0.0399002,0.00002414598,0.0004711034,0.000418158,0.0007873877],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0004865006,0.00007104925,0.6524847,0.0005500157,0.00002477187,0.00019022,0.002069142,1.46414e-8,0.003239315,0.002099134,0.05000952,0.2887757],"study_design_scores_gemma":[0.005976217,0.0009141307,0.8380067,0.0002616346,0.000009474399,0.0004714748,0.01325668,0.0000468961,0.003577761,0.05769408,0.07959314,0.0001917869],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8925747,0.001622374,0.01624966,0.08418978,0.0002347002,0.00121373,0.000007481041,0.00003373888,0.003873788],"genre_scores_gemma":[0.9467005,0.002256104,0.03776179,0.009110031,0.001044799,0.0004892785,0.00002054261,0.00006266178,0.002554325],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2885839,"threshold_uncertainty_score":0.9905749,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2006031679","doi":"10.1186/1471-2288-7-34","title":"A simulation study of sample size for multilevel logistic regression models","year":2007,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":400,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences; St. Michael's Hospital; University of Toronto","funders":"Ontario Ministry of Health and Long-Term Care","keywords":"Statistics; Sample size determination; Multilevel model; Logistic regression; Random effects model; Variance (accounting); Covariance; Mathematics; Hierarchical database model; Sample (material); Econometrics; Regression; Medicine; Computer science; Meta-analysis; Data mining","authors":[{"name":"Rahim Moineddin","is_ca":true},{"name":"Flora I. Matheson","is_ca":true},{"name":"Richard H. Glazier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8745461031859402,"gpt":0.6879301196957895,"spread":0.1866159834901506,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06386741,0.0001654242,0.0007595498,0.0002405285,0.000158251,0.00001108761,0.0004762927,0.0003602869,0.0005399524],"category_scores_gemma":[0.8397517,0.0001169479,0.00009135276,0.0003499299,0.0005103262,0.00004634779,0.000244262,0.0005821232,0.000002250609],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006427249,"about_ca_system_score_gemma":0.0004071246,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0003463426,"about_ca_topic_score_gemma":0.0004366847,"domain_scores_codex":[0.9860284,0.009764019,0.001056278,0.0005305533,0.001806435,0.0008143009],"domain_scores_gemma":[0.2488983,0.7491971,0.0001885849,0.0004714881,0.0008430945,0.0004015161],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.002205864,0.001806573,0.00114456,0.0008837701,0.00005611546,0.0000297799,0.001936532,0.0002782462,0.0003765333,0.5951161,0.0002335729,0.3959323],"study_design_scores_gemma":[0.001276607,0.0009494951,0.001201363,0.0001025231,0.00002155092,0.000001782602,0.0009770017,0.2449096,0.000194568,0.7502479,0.00002880489,0.00008880534],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.03042248,0.00002575758,0.9678813,0.00005189094,0.0001558438,0.001222795,0.00002837555,0.00003411518,0.0001774542],"genre_scores_gemma":[0.3474437,0.000004620937,0.6522952,0.00001902112,0.0001078447,0.00007018898,0.000001601707,0.00001782322,0.00004000866],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7758843,"threshold_uncertainty_score":0.9639455,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2104471097","doi":"10.1186/1471-2288-6-2","title":"Protocol of the COSMIN study: COnsensus-based Standards for the selection of health Measurement INstruments","year":2006,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":398,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"Vrije Universiteit Amsterdam","keywords":"Checklist; Protocol (science); Delphi method; Interpretability; Delphi; Health care; Medicine; Criterion validity; Consistency (knowledge bases); Computer science; Medical physics; Construct validity; Applied psychology; Management science; Psychology; Patient satisfaction; Alternative medicine; Nursing; Engineering; Artificial intelligence; Pathology","authors":[{"name":"Lidwine B. Mokkink","is_ca":false},{"name":"Caroline B. Terwee","is_ca":false},{"name":"DL Knol","is_ca":false},{"name":"Paul W. Stratford","is_ca":true},{"name":"Jordi Alonso","is_ca":false},{"name":"D Lyden Patrick","is_ca":false},{"name":"L.M. Bouter","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7892792696623773,"gpt":0.6850943130718052,"spread":0.104184956590572,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.3081651,0.00009545575,0.0004057651,0.0002120583,0.0008460847,0.00001898402,0.00116221,0.0001868284,0.0002584511],"category_scores_gemma":[0.166265,0.0000545398,0.0001186743,0.001142047,0.003162908,0.00001797005,0.0002120094,0.0006513162,6.194642e-7],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007949842,"about_ca_system_score_gemma":0.02439182,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.02563809,"about_ca_topic_score_gemma":0.05807182,"domain_scores_codex":[0.9025127,0.08032576,0.001247094,0.0004597597,0.01428321,0.001171456],"domain_scores_gemma":[0.9619954,0.03420305,0.0003136944,0.0004197179,0.002907735,0.000160385],"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":[0.005967361,0.004370817,0.1933824,0.002647937,0.0001863802,0.000003447914,0.003937534,0.0001065149,0.003716674,0.05076686,0.119704,0.6152101],"study_design_scores_gemma":[0.02229526,0.01761951,0.1635665,0.001756518,0.00008940983,0.000008623771,0.04558768,0.006894135,0.08493653,0.100032,0.5565017,0.0007121222],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"protocol","genre_gemma":"empirical","genre_scores_codex":[0.04753224,0.0001301659,0.4108446,0.04756682,0.0004927361,0.4900227,0.0000765205,0.0001118461,0.003222411],"genre_scores_gemma":[0.8390889,0.000009571549,0.03110222,0.0001209359,0.0002112037,0.1291918,9.05159e-7,0.0000250888,0.0002493817],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.7915567,"threshold_uncertainty_score":0.9995499,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3020176834","doi":"10.1186/s12874-020-00975-3","title":"How are systematic reviews of prevalence conducted? A methodological study","year":2020,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":356,"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":"Systematic review; Medicine; MEDLINE; Interquartile range; Meta-analysis; Family medicine; Pathology; Surgery","authors":[{"name":"Celina Borges Migliavaca","is_ca":false},{"name":"Cinara Stein","is_ca":false},{"name":"Verônica Colpani","is_ca":false},{"name":"Timothy Hugh Barker","is_ca":false},{"name":"Zachary Munn","is_ca":true},{"name":"Maicon Falavigna","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9945339847664632,"gpt":0.7580766681875971,"spread":0.236457316578866,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","open_science","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8729903,0.0005029082,0.01759144,0.0007095059,0.0001975426,0.0004260998,0.008294229,0.000470788,0.02832844],"category_scores_gemma":[0.9899712,0.0002086562,0.002637829,0.00512251,0.001043155,0.0002125376,0.001535075,0.001572345,0.001630786],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004643364,"about_ca_system_score_gemma":0.0006703964,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00001926957,"about_ca_topic_score_gemma":0.00008694932,"domain_scores_codex":[0.07606103,0.8735411,0.01865892,0.002796048,0.02791519,0.00102778],"domain_scores_gemma":[0.05774934,0.9103556,0.01311322,0.00888063,0.007345343,0.002555835],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"qualitative","study_design_scores_codex":[0.0009586462,0.003114895,0.2391816,0.3600555,0.003069973,0.0009832621,0.02116098,0.00003477736,0.00581306,0.02808451,0.2634945,0.07404824],"study_design_scores_gemma":[0.01301577,0.02205756,0.0912033,0.03252134,0.008213393,0.001136707,0.3615171,0.08477633,0.002281191,0.1112641,0.2667493,0.005263848],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.05663936,0.01649569,0.9019256,0.01203034,0.0004177711,0.01150234,0.00001577085,0.00002218495,0.0009509121],"genre_scores_gemma":[0.4181405,0.00220705,0.563491,0.002069059,0.0008115571,0.00246243,0.000004990253,0.00006243664,0.01075104],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3615011,"threshold_uncertainty_score":0.9991466,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2127832821","doi":"10.1186/1471-2288-12-114","title":"What is the most appropriate knowledge synthesis method to conduct a review? Protocol for a scoping review","year":2012,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":347,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Ottawa; Royal College of Physicians and Surgeons of Canada; Institute of Population and Public Health; University of Toronto; St. Michael's Hospital","funders":"Canadian Institutes of Health Research","keywords":"CINAHL; Systematic review; Psychological intervention; Computer science; Knowledge translation; Protocol (science); MEDLINE; Health care; Management science; Knowledge management; Data science; Medicine; Alternative medicine; Nursing","authors":[{"name":"Monika Kastner","is_ca":true},{"name":"Andrea C. Tricco","is_ca":true},{"name":"Charlene Soobiah","is_ca":true},{"name":"Erin Lillie","is_ca":true},{"name":"Laure Perrier","is_ca":true},{"name":"Tanya Horsley","is_ca":true},{"name":"Vivian Welch","is_ca":true},{"name":"Elise Cogo","is_ca":true},{"name":"Jesmin Antony","is_ca":true},{"name":"Sharon E. Straus","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9872001315892293,"gpt":0.8771516823879036,"spread":0.1100484492013256,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","research_integrity","insufficient_payload"],"category_scores_codex":[0.5292529,0.0008213356,0.007229471,0.0008170329,0.002136219,0.0000617343,0.004289537,0.001390513,0.01606408],"category_scores_gemma":[0.6467993,0.0004533543,0.0008509154,0.004615212,0.0008299575,0.0003425933,0.002632109,0.005369028,0.004243259],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0009629554,"about_ca_system_score_gemma":0.05893253,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002582203,"about_ca_topic_score_gemma":0.0006893551,"domain_scores_codex":[0.6454698,0.3395262,0.005809346,0.001709321,0.003341654,0.0041437],"domain_scores_gemma":[0.2145015,0.7724335,0.002507134,0.003311108,0.003296035,0.003950795],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00001352758,0.0000187206,2.77239e-7,0.4429736,0.00002538064,7.913951e-7,0.000506961,1.10348e-9,9.433029e-8,0.0002994678,0.09161814,0.464543],"study_design_scores_gemma":[0.0002158863,0.00007457456,1.544374e-7,0.4654613,0.0002134536,0.00001862288,0.0001985056,0.000002461014,8.786342e-7,0.0001293499,0.5334772,0.0002075457],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[1.426432e-9,0.5312324,0.0458245,0.02781585,0.0005752246,0.3940903,0.00005859684,0.00004308603,0.0003600612],"genre_scores_gemma":[6.376843e-11,0.4435361,0.08554396,0.03578963,0.000939711,0.4332335,0.000008511363,0.0000731261,0.0008754947],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4643355,"threshold_uncertainty_score":0.9999059,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1950366691","doi":"10.1186/s12874-015-0045-7","title":"The use of the Godin-Shephard Leisure-Time Physical Activity Questionnaire in oncology research: a systematic review","year":2015,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Cancer survivorship and care","field":"Medicine","cited_by":339,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Université Laval; Concordia University","funders":"Concordia University","keywords":"Medicine; MEDLINE; Scale (ratio); Physical activity; Alternative medicine; Family medicine; Clinical psychology; Physical therapy; Pathology","authors":[{"name":"Steve Amireault","is_ca":true},{"name":"Gaston Godin","is_ca":true},{"name":"Jason Lacombe","is_ca":false},{"name":"Catherine M. Sabiston","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.8632981202567224,"gpt":0.6646766439473891,"spread":0.1986214763093332,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity"],"consensus_categories":["metaresearch","research_integrity"],"category_scores_codex":[0.1736662,0.0005153818,0.008900436,0.0006226457,0.0002923929,0.00004118086,0.002208122,0.001399016,0.0001045837],"category_scores_gemma":[0.4488688,0.0002369659,0.001309334,0.00365449,0.004166098,0.00007963181,0.00174263,0.00881413,0.0002070827],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002045786,"about_ca_system_score_gemma":0.02574543,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0018534,"about_ca_topic_score_gemma":0.005916719,"domain_scores_codex":[0.7634559,0.2260914,0.001622318,0.0009521808,0.006411162,0.001466976],"domain_scores_gemma":[0.7919949,0.2000659,0.0007823057,0.003259565,0.002720749,0.001176586],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0002026644,0.0003039808,0.00001068112,0.6094832,0.0001729421,0.0001349336,0.0002140878,6.242743e-8,0.000001311394,0.0005637639,0.005957132,0.3829552],"study_design_scores_gemma":[0.0003152767,0.0004289037,0.0000153061,0.4573495,0.0006357623,0.0003179057,0.00006604853,0.00004726278,0.000001283621,0.0002457861,0.5404457,0.0001312567],"study_design_candidate":"systematic_review","study_design_consensus":null,"genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.00008585553,0.987205,0.0001742894,0.003611952,0.0006126011,0.007966571,0.0000225125,0.00003029679,0.0002909737],"genre_scores_gemma":[0.00004492538,0.9940903,0.0002324008,0.0001298091,0.000617555,0.002848014,0.00001329146,0.00008227483,0.001941424],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.5344886,"threshold_uncertainty_score":0.9998974,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2014411687","doi":"10.1186/1471-2288-8-32","title":"The ratio of means method as an alternative to mean differences for analyzing continuous outcome variables in meta-analysis: A simulation study","year":2008,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":339,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Hospital for Sick Children; Health Sciences Centre; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre; SickKids Foundation; University of Toronto","funders":"Canadian Institutes of Health Research","keywords":"Statistics; Confidence interval; Meta-analysis; Strictly standardized mean difference; Publication bias; Pooling; Standard deviation; Study heterogeneity; Sample size determination; Pooled variance; Weighting; Mean difference; Type I and type II errors; Statistical power; Mathematics; Medicine; Econometrics; Computer science; Internal medicine","authors":[{"name":"Jan O. Friedrich","is_ca":true},{"name":"Neill K. J. Adhikari","is_ca":true},{"name":"Joseph Beyene","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.974214244217264,"gpt":0.7376015835997741,"spread":0.2366126606174899,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7407974,0.0003678635,0.01007288,0.0023993,0.0005326459,0.0002879225,0.004450638,0.0002134198,0.004008436],"category_scores_gemma":[0.7792231,0.0001485715,0.003414688,0.006552769,0.0004418008,0.0002014353,0.0004719245,0.0005920812,0.000062124],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006288167,"about_ca_system_score_gemma":0.0006275284,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00180581,"about_ca_topic_score_gemma":0.02008636,"domain_scores_codex":[0.4277253,0.5310211,0.01695005,0.002695406,0.02043251,0.001175643],"domain_scores_gemma":[0.2144418,0.7687662,0.004065583,0.005133848,0.006678965,0.0009135803],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00127766,0.00204495,0.6253496,0.0001463233,0.1253039,0.00009242287,0.04958177,0.09963482,0.0001837025,0.06144586,0.0008180459,0.03412099],"study_design_scores_gemma":[0.001167362,0.001667888,0.04606834,0.00001114977,0.02792487,0.000009995226,0.02609448,0.8183098,0.00008501181,0.07712758,0.001168631,0.0003648342],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.1427725,0.0004867298,0.8529857,0.000812789,0.00009912213,0.002614624,0.00001383048,0.000004531119,0.0002101812],"genre_scores_gemma":[0.7705151,0.00001970036,0.2267715,0.0001194116,0.00007673687,0.0006556588,0.000004118641,0.00001263335,0.001825175],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.718675,"threshold_uncertainty_score":0.9977945,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2133710899","doi":"10.1186/1471-2288-14-80","title":"Meta-ethnography 25 years on: challenges and insights for synthesising a large number of qualitative studies","year":2014,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":331,"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":"Health Services Research Programme; National Institutes of Health; Health Services and Delivery Research Programme; National Institute for Health and Care Research","keywords":"Reflexivity; Qualitative research; Ethnography; DECIPHER; Process (computing); Interpretation (philosophy); Management science; Epistemology; Sociology; Health care; Engineering ethics; Psychology; Data science; Computer science; Social science; Bioinformatics; Political science","authors":[{"name":"Francine Toyé","is_ca":false},{"name":"Kate Seers","is_ca":false},{"name":"Nick Allcock","is_ca":false},{"name":"Michelle Briggs","is_ca":false},{"name":"Eloise Carr","is_ca":true},{"name":"Karen Barker","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9910430930723466,"gpt":0.8531529516083495,"spread":0.1378901414639971,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1824203,0.000137792,0.001053479,0.0004942649,0.0005701018,0.000003196199,0.0003522976,0.0002910202,0.000398081],"category_scores_gemma":[0.4368542,0.00009621782,0.0001180328,0.0004418337,0.001331187,0.00007228752,0.0003192893,0.0008207042,0.00005102992],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005744439,"about_ca_system_score_gemma":0.000997917,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001081373,"about_ca_topic_score_gemma":0.001585746,"domain_scores_codex":[0.8823978,0.1134272,0.001011878,0.0005123229,0.001541357,0.001109409],"domain_scores_gemma":[0.530417,0.4678925,0.0002342781,0.0002950931,0.000713159,0.0004479721],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"qualitative","study_design_scores_codex":[0.0002616482,0.0000688014,0.0003489162,0.002715277,0.0005178853,0.000002021506,0.4769914,1.111489e-7,0.00007173682,0.4844393,0.003939003,0.03064392],"study_design_scores_gemma":[0.003228795,0.001076182,0.007433739,0.0006807266,0.0001816571,0.000003331429,0.482475,0.000166442,0.0002039822,0.3323139,0.1719816,0.0002546479],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7697022,0.01042444,0.1312101,0.07614697,0.001056712,0.004378786,0.0001058618,0.0001155284,0.006859409],"genre_scores_gemma":[0.5219412,0.01256931,0.4437415,0.01600044,0.0007706854,0.004254803,0.000005157374,0.00007319741,0.0006436575],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3544653,"threshold_uncertainty_score":0.8418703,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2132268984","doi":"10.1186/1471-2288-13-150","title":"Exploratory randomized controlled trial evaluating the impact of a waiting list control design","year":2013,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Substance Abuse Treatment and Outcomes","field":"Medicine","cited_by":330,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Centre for Addiction and Mental Health","funders":"Medical Research Council; Wellcome Trust","keywords":"Randomized controlled trial; Intervention (counseling); Waiting list; Research design; Exploratory research; Medicine; Psychology; Control (management); Clinical trial; Random assignment; Computer science; Psychiatry; Statistics","authors":[{"name":"John Cunningham","is_ca":true},{"name":"Kypros Kypri","is_ca":false},{"name":"Jim McCambridge","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6349216352576696,"gpt":0.5924382886032753,"spread":0.04248334665439435,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1407481,0.0002598212,0.003486448,0.0003371015,0.0001987348,0.00002725724,0.0003896365,0.000315175,0.005914731],"category_scores_gemma":[0.4946291,0.000111312,0.001200282,0.0004389416,0.001619047,0.00008946925,0.00008126901,0.0009449681,0.00008662551],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001630416,"about_ca_system_score_gemma":0.003664415,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004708856,"about_ca_topic_score_gemma":0.00003843435,"domain_scores_codex":[0.913429,0.08127724,0.001435598,0.0004042689,0.00253035,0.0009235287],"domain_scores_gemma":[0.4313613,0.5650887,0.0005211292,0.0007862983,0.0015172,0.0007253694],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"randomized_trial","study_design_gemma":"randomized_trial","study_design_scores_codex":[0.9830998,0.0002329655,0.001275721,0.00002832679,0.001174861,0.00002719952,0.001156668,0.00001633593,0.002242253,0.0005278063,0.0009151806,0.00930291],"study_design_scores_gemma":[0.9094582,0.003281047,0.0001984676,0.000144912,0.0003233379,0.00002112831,0.001453669,0.08204064,0.0002784295,0.002713476,0.000001843775,0.00008488329],"study_design_candidate":"randomized_trial","study_design_consensus":"randomized_trial","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9148375,0.002146519,0.06680942,0.00144856,0.0002552591,0.01386031,5.300628e-7,0.00004826694,0.0005936362],"genre_scores_gemma":[0.9719265,0.0002500121,0.02087096,0.0002255115,0.0006630713,0.005481181,0.00000419345,0.00003346382,0.000545184],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4838114,"threshold_uncertainty_score":0.994994,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2607509261","doi":"10.1186/s12874-017-0347-z","title":"Grey literature in systematic reviews: a cross-sectional study of the contribution of non-English reports, unpublished studies and dissertations to the results of meta-analyses in child-relevant reviews","year":2017,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":319,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Cochrane; University of Alberta","funders":"Canadian Institutes of Health Research","keywords":"Grey literature; Meta-analysis; Cross-sectional study; Systematic review; Medicine; MEDLINE; Psychology; Family medicine; Political science; Pathology","authors":[{"name":"Lisa Hartling","is_ca":true},{"name":"Robin Featherstone","is_ca":true},{"name":"Megan Nuspl","is_ca":true},{"name":"Kassi Shave","is_ca":true},{"name":"Donna M Dryden","is_ca":true},{"name":"Ben Vandermeer","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9431944957606768,"gpt":0.7244683342400873,"spread":0.2187261615205895,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.755313,0.0002968226,0.01056148,0.0009261007,0.0003863494,0.0004373554,0.003270122,0.0002049301,0.0001699904],"category_scores_gemma":[0.9816445,0.00009194949,0.001677466,0.003349427,0.000793328,0.000318722,0.0009097002,0.0008418006,0.000007334212],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004934258,"about_ca_system_score_gemma":0.0003246479,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0005566783,"about_ca_topic_score_gemma":0.01891867,"domain_scores_codex":[0.3584462,0.5684512,0.0500338,0.002738073,0.01941285,0.0009179079],"domain_scores_gemma":[0.3476224,0.5470673,0.04659528,0.02315849,0.03480441,0.0007521351],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.000594053,0.001288206,0.894954,0.02828134,0.006011262,0.00006996554,0.05259082,0.0003043597,0.0003474337,0.003003713,0.01060733,0.001947557],"study_design_scores_gemma":[0.001112208,0.000321327,0.972266,0.007419392,0.001658172,0.00004209316,0.01096758,0.0005803208,0.00005844279,0.004138508,0.001283183,0.0001527746],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8692095,0.1088604,0.0031332,0.002389226,0.0007122696,0.01534135,0.00003237477,0.000002070721,0.0003196599],"genre_scores_gemma":[0.9923389,0.002112809,0.003682334,0.00004233341,0.00006489961,0.001169474,0.000004627215,0.00000715004,0.0005774976],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2263315,"threshold_uncertainty_score":0.9989835,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2981452012","doi":"10.1186/s12874-019-0829-2","title":"BUGSnet: an R package to facilitate the conduct and reporting of Bayesian network Meta-analyses","year":2019,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":319,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Calgary; Public Health Ontario; University of Waterloo","funders":"University of Waterloo","keywords":"Computer science; Bayesian probability; Nice; Inference; Bayesian network; Gibbs sampling; Data mining; Data science; Machine learning; Operations research; Artificial intelligence; Mathematics","authors":[{"name":"Audrey Béliveau","is_ca":true},{"name":"Devon J. Boyne","is_ca":true},{"name":"Justin Slater","is_ca":false},{"name":"Darren R. Brenner","is_ca":true},{"name":"Paul Arora","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9946169984997515,"gpt":0.76499760368872,"spread":0.2296193948110314,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch","insufficient_payload"],"category_scores_codex":[0.8535877,0.0003034631,0.008746522,0.0005479185,0.0002333893,0.0002821023,0.003228605,0.0002394635,0.04568693],"category_scores_gemma":[0.8736935,0.0001059734,0.001982065,0.002832173,0.0007281927,0.0001513272,0.0008943671,0.0007870075,0.0009043295],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00001895679,"about_ca_system_score_gemma":0.000597025,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0004823291,"about_ca_topic_score_gemma":0.001402961,"domain_scores_codex":[0.3427695,0.5956274,0.03091038,0.003123395,0.02592685,0.00164248],"domain_scores_gemma":[0.3074303,0.6575924,0.01434377,0.01301906,0.005401393,0.002213095],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"design_other","study_design_gemma":"not_applicable","study_design_scores_codex":[0.0006698034,0.0003543627,0.2309749,0.001392444,0.01890395,0.0002171597,0.01488528,0.002977142,0.01388507,0.06421934,0.3188866,0.3326339],"study_design_scores_gemma":[0.001350477,0.003189131,0.08497331,0.0002267761,0.007223051,0.0004407148,0.03402036,0.06188311,0.001414442,0.3134416,0.4905662,0.001270856],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.3690816,0.005618363,0.6097715,0.006320025,0.0003733038,0.002163044,0.00001558989,0.000008703097,0.006647803],"genre_scores_gemma":[0.6383692,0.0001076775,0.338438,0.001231465,0.0003124656,0.0001947406,0.000008182651,0.0000292553,0.02130902],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.3313631,"threshold_uncertainty_score":0.9998736,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1534789916","doi":"10.1186/1471-2288-7-7","title":"Researching complementary and alternative treatments – the gatekeepers are not at home","year":2007,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Complementary and Alternative Medicine Studies","field":"Medicine","cited_by":289,"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":"Context (archaeology); Modalities; Alternative medicine; Acknowledgement; Medicine; Clinical trial; Strengths and weaknesses; Evidence-based medicine; Randomized controlled trial; Psychology; Computer science; Pathology; Sociology; Social psychology","authors":[{"name":"Vinjar Fønnebø","is_ca":false},{"name":"Sameline Grimsgaard","is_ca":false},{"name":"Harald Walach","is_ca":false},{"name":"Cheryl Ritenbaugh","is_ca":false},{"name":"Arne Johan Norheim","is_ca":false},{"name":"Hugh MacPherson","is_ca":false},{"name":"George Lewith","is_ca":false},{"name":"Laila Launsø","is_ca":false},{"name":"Mary Koithan","is_ca":false},{"name":"Torkel Falkenberg","is_ca":false},{"name":"Heather Boon","is_ca":true},{"name":"Mikel Aickin","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6942752946247227,"gpt":0.6102542469850069,"spread":0.08402104763971585,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03328614,0.0002403486,0.000692229,0.0004431776,0.0008869567,0.00001527863,0.000443392,0.00005386683,0.002142626],"category_scores_gemma":[0.02192437,0.0001395619,0.0001136912,0.0004135623,0.002715972,0.00005344528,0.001471147,0.001268509,0.00004822304],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003540866,"about_ca_system_score_gemma":0.000237831,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001773593,"about_ca_topic_score_gemma":0.004675636,"domain_scores_codex":[0.9885865,0.006037229,0.0005707502,0.000664119,0.002886725,0.001254645],"domain_scores_gemma":[0.9424444,0.05580031,0.0001110638,0.0004556999,0.0003197716,0.0008687819],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00789312,0.0005006446,0.7625061,0.0004163175,0.001396621,0.003131081,0.00664797,0.000001033522,0.0067876,0.005658385,0.00653095,0.1985302],"study_design_scores_gemma":[0.008941995,0.00313304,0.9279895,0.0004874972,0.0001342005,0.0006669192,0.02461476,0.0003696073,0.006140728,0.003828828,0.02344298,0.0002499467],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9751755,0.001309358,0.006311165,0.01378483,0.0003468356,0.0009566834,0.00002907837,0.0000328424,0.002053716],"genre_scores_gemma":[0.9815481,0.001820805,0.01094219,0.002059328,0.00133291,0.00009238285,0.00005097212,0.0000339457,0.002119387],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1982803,"threshold_uncertainty_score":0.9999981,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2036060563","doi":"10.1186/1471-2288-10-82","title":"Inter-rater agreement and reliability of the COSMIN (COnsensus-based Standards for the selection of health status Measurement Instruments) Checklist","year":2010,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":289,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McMaster University","funders":"Vrije Universiteit Amsterdam; University of Oxford; McMaster University; Goddard Space Flight Center; University of Washington","keywords":"Checklist; Kappa; Inter-rater reliability; Intraclass correlation; Cohen's kappa; Reliability (semiconductor); Medicine; Quality (philosophy); Agreement; Terminology; Psychology; Applied psychology; Physical therapy; Medical physics; Clinical psychology; Statistics; Psychometrics; Rating scale; Mathematics","authors":[{"name":"Lidwine B. Mokkink","is_ca":false},{"name":"Caroline B. Terwee","is_ca":false},{"name":"Elizabeth Gibbons","is_ca":false},{"name":"Paul W. Stratford","is_ca":true},{"name":"Jordi Alonso","is_ca":false},{"name":"Donald L. Patrick","is_ca":false},{"name":"Dirk L. Knol","is_ca":false},{"name":"L.M. Bouter","is_ca":false},{"name":"Henrica C. W. de Vet","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5948043968788056,"gpt":0.5666070490965269,"spread":0.02819734778227867,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.394015,0.0001699311,0.0006664849,0.0002072472,0.0004807004,0.000051671,0.001216097,0.0002002344,0.001058631],"category_scores_gemma":[0.3844943,0.00007941615,0.0002122274,0.0007389415,0.002842776,0.00004208873,0.0004454363,0.00094143,0.000002441472],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003325753,"about_ca_system_score_gemma":0.00629961,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001463431,"about_ca_topic_score_gemma":0.009785491,"domain_scores_codex":[0.9401533,0.03720393,0.003069788,0.001145841,0.01720643,0.001220689],"domain_scores_gemma":[0.9302254,0.06034505,0.0008108895,0.001648001,0.006530828,0.000439812],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.002453449,0.0008920706,0.3676997,0.0008563165,0.00009789977,3.107363e-7,0.001033717,0.0001358515,0.01665502,0.002569515,0.01801015,0.589596],"study_design_scores_gemma":[0.007085396,0.004468236,0.5668253,0.0006157815,0.0000784599,0.000008793487,0.005424237,0.02663061,0.07637274,0.07185221,0.2402325,0.0004056925],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.7360402,0.0002679026,0.2360404,0.02366712,0.001168196,0.00253549,0.00008794098,0.000008788287,0.0001840239],"genre_scores_gemma":[0.9704114,0.00007649354,0.02881524,0.0003273527,0.00007062288,0.0001740064,0.000001368642,0.000008853127,0.000114639],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.5891903,"threshold_uncertainty_score":0.9998709,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2136604958","doi":"10.1186/1471-2288-13-118","title":"Towards a methodology for cluster searching to provide conceptual and contextual “richness” for systematic reviews of complex interventions: case study (CLUSTER)","year":2013,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Health Policy Implementation Science","field":"Health Professions","cited_by":279,"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":"National Institute for Health and Care Research","keywords":"Context (archaeology); Systematic review; Psychological intervention; Computer science; Data science; Psychology; Intervention (counseling); Management science; MEDLINE","authors":[{"name":"Andrew Booth","is_ca":false},{"name":"Janet Harris","is_ca":false},{"name":"Elizabeth Croot","is_ca":false},{"name":"Jane Springett","is_ca":true},{"name":"Fiona Campbell","is_ca":false},{"name":"Emma Wilkins","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9784724983140504,"gpt":0.8165912940575061,"spread":0.1618812042565443,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2925355,0.0002696053,0.002345257,0.000907017,0.001035201,0.00002938293,0.0008691291,0.0003555902,0.001175314],"category_scores_gemma":[0.7194202,0.0001927752,0.0002359844,0.0007631466,0.001154735,0.0001979351,0.001220943,0.001066381,0.00008645488],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002537326,"about_ca_system_score_gemma":0.003019693,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006993514,"about_ca_topic_score_gemma":0.01252285,"domain_scores_codex":[0.7959505,0.1958864,0.004132556,0.0008707083,0.001322017,0.001837874],"domain_scores_gemma":[0.4313544,0.5625533,0.0009662858,0.0008083994,0.002709439,0.00160813],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"qualitative","study_design_scores_codex":[0.002049092,0.0007818696,0.006887091,0.4259932,0.0003484589,0.00004938372,0.3584399,0.000006640632,0.001398205,0.02639562,0.08705821,0.09059229],"study_design_scores_gemma":[0.01917712,0.01220494,0.002659865,0.01457058,0.000210984,0.0006323652,0.9009613,0.01000938,0.00006222179,0.01005558,0.02875894,0.0006967523],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1746537,0.000162172,0.7644108,0.0105848,0.0003136554,0.04971859,0.00006410097,0.000032082,0.00006010099],"genre_scores_gemma":[0.1543279,0.00004776637,0.7607127,0.01156385,0.0004446708,0.07210167,0.00001357321,0.00005754461,0.0007303343],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5425214,"threshold_uncertainty_score":0.9997377,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2112200626","doi":"10.1186/1471-2288-12-82","title":"Interpreting the concordance statistic of a logistic regression model: relation to the variance and odds ratio of a continuous explanatory variable","year":2012,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":271,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Institute for Clinical Evaluative Sciences","funders":"Canadian Institutes of Health Research; Ontario Ministry of Health and Long-Term Care; Institute for Clinical Evaluative Sciences; Heart and Stroke Foundation of Canada","keywords":"Statistics; Mathematics; Statistic; Logistic regression; Sample size determination; Normal distribution; Standard deviation; Sampling distribution; PRESS statistic; Monte Carlo method; Variable (mathematics); Cumulative distribution function; Regression analysis; Econometrics; Ancillary statistic; F-test; Probability density function","authors":[{"name":"Peter C. Austin","is_ca":true},{"name":"Ewout W. Steyerberg","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.5031041758697814,"gpt":0.5613378965208847,"spread":0.05823372065110333,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.04448416,0.0001270187,0.000541539,0.00009257286,0.0001541563,0.00001290939,0.0004636037,0.0001732981,0.0002117715],"category_scores_gemma":[0.4205032,0.00006629209,0.00003387572,0.0003535995,0.001094917,0.00005645094,0.0003310838,0.0006908388,0.000003281677],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003675538,"about_ca_system_score_gemma":0.0004785519,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001570599,"about_ca_topic_score_gemma":0.00004367694,"domain_scores_codex":[0.9826901,0.01475744,0.0007447417,0.0002650896,0.001018946,0.0005237159],"domain_scores_gemma":[0.8122408,0.1864792,0.0002477326,0.0004345629,0.0003563781,0.0002412646],"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":[0.0002388986,0.0000548915,0.0006699442,0.0003294735,0.00001931947,0.000002092122,0.001986678,0.00001638171,0.00145908,0.9713193,0.001030092,0.02287387],"study_design_scores_gemma":[0.0003365251,0.0002693946,0.002442401,0.0006926664,0.00004027951,0.00002734845,0.0009263704,0.1354298,0.0004255315,0.8591267,0.0001813765,0.0001015939],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.004900067,0.0004523951,0.992688,0.0005532404,0.0001655631,0.0004677585,0.00002850211,0.00001160747,0.0007328856],"genre_scores_gemma":[0.3056449,0.00004722084,0.6939827,0.0001065162,0.00006340764,0.00007649522,0.00000113376,0.00001118517,0.00006639726],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.376019,"threshold_uncertainty_score":0.9839046,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1791587479","doi":"10.1186/1471-2288-5-1","title":"Structural equation model testing and the quality of natural killer cell activity measurements","year":2005,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Immune Cell Function and Interaction","field":"Immunology and Microbiology","cited_by":258,"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":"Covariance; Structural equation modeling; Econometrics; Natural killer cell; Natural (archaeology); Statistics; Cell; Immunology; Biology; Mathematics; Cytotoxic T cell; Genetics","authors":[{"name":"Leslie A. Hayduk","is_ca":true},{"name":"Hannah Pazderka","is_ca":true},{"name":"Greta G. Cummings","is_ca":true},{"name":"Merry-Jo D Levers","is_ca":true},{"name":"Melanie A. Beres","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.6722927013247167,"gpt":0.5234907777996085,"spread":0.1488019235251082,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.02664762,0.0001158978,0.0003797479,0.0001323179,0.0002886644,0.000009880466,0.0002564112,0.0002962893,0.0006599888],"category_scores_gemma":[0.06981695,0.00007176852,0.00008019385,0.0001786636,0.001191888,0.0001178401,0.000213513,0.001186107,0.00005100489],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00006845035,"about_ca_system_score_gemma":0.0003214679,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0008069844,"about_ca_topic_score_gemma":0.0002212434,"domain_scores_codex":[0.9818569,0.01661591,0.000470299,0.0002883188,0.000365869,0.0004026787],"domain_scores_gemma":[0.9698097,0.02927405,0.0001949818,0.0002657713,0.0004076247,0.00004788554],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"bench_or_experimental","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.005345464,0.0001143337,0.00378446,0.0001215153,0.0001024477,2.92956e-7,0.0009327443,0.0002214415,0.8675091,0.002889812,0.001950379,0.1170281],"study_design_scores_gemma":[0.02449879,0.0006959492,0.07177069,0.0001665209,0.00009414231,0.0001283781,0.001434016,0.346514,0.5423534,0.007925832,0.003807937,0.0006102984],"study_design_candidate":"bench_or_experimental","study_design_consensus":"bench_or_experimental","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9554499,0.001736233,0.03710837,0.001430687,0.001621467,0.0003681597,8.588009e-7,0.00002874731,0.002255565],"genre_scores_gemma":[0.9849671,0.00005394988,0.01078014,0.0001657803,0.00006652174,0.0000249991,0.000008102988,0.000008250101,0.003925117],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3462926,"threshold_uncertainty_score":0.9380184,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3121736026","doi":"10.1186/s12874-021-01269-y","title":"Managing overlap of primary study results across systematic reviews: practical considerations for authors of overviews of reviews","year":2021,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":254,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Montfort Hospital; Cochrane; Ottawa Hospital; University of British Columbia","funders":"","keywords":"Computer science; Data extraction; Usability; Systematic review; Data science; Information retrieval; Data mining; Resource (disambiguation); MEDLINE; Biology","authors":[{"name":"Carole Lunny","is_ca":true},{"name":"Dawid Pieper","is_ca":false},{"name":"Pierre Thabet","is_ca":true},{"name":"Salmaan Kanji","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9844134215215561,"gpt":0.7615619304483037,"spread":0.2228514910732524,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.8596493,0.0003687097,0.01653252,0.0006103158,0.0002069097,0.0001303655,0.00163232,0.0003381071,0.003417465],"category_scores_gemma":[0.9803037,0.0001854319,0.003089384,0.003078158,0.0007863953,0.0002002764,0.0009940596,0.0007605164,0.0001659051],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007446847,"about_ca_system_score_gemma":0.002175669,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005698229,"about_ca_topic_score_gemma":0.0006480095,"domain_scores_codex":[0.1850978,0.7381988,0.05216662,0.00258619,0.02076977,0.001180797],"domain_scores_gemma":[0.08179337,0.8765656,0.01937375,0.009625222,0.01181066,0.0008314144],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001038512,0.005299697,0.007354485,0.4385677,0.002813823,0.0002660565,0.02587352,0.0000524461,0.004042008,0.07921266,0.3514602,0.08401892],"study_design_scores_gemma":[0.01477064,0.006521545,0.0130558,0.08185232,0.007149188,0.001177988,0.1165351,0.02191572,0.004031414,0.3193175,0.411468,0.002204707],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01456451,0.1208891,0.8305911,0.004574183,0.001041031,0.02408017,0.0001771815,0.000008777578,0.004074018],"genre_scores_gemma":[0.02690693,0.01508829,0.9491822,0.0004829457,0.0002234332,0.002090532,0.00003005158,0.00004444695,0.005951183],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.3567154,"threshold_uncertainty_score":0.9980681,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2056264205","doi":"10.1186/1471-2288-11-126","title":"A randomised trial and economic evaluation of the effect of response mode on response rate, response bias, and item non-response in a survey of doctors","year":2011,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Survey Methodology and Nonresponse","field":"Social Sciences","cited_by":252,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"University of Manitoba","funders":"National Health and Medical Research Council; Medical Research Council; University of Melbourne; Monash University","keywords":"Non-response bias; Response bias; Medicine; Family medicine; Response time; Interactive voice response; Mode (computer interface); Computer-assisted web interviewing; Psychology; Statistics; Social psychology; Computer science","authors":[{"name":"Anthony Scott","is_ca":false},{"name":"Sung‐Hee Jeon","is_ca":false},{"name":"Catherine Joyce","is_ca":false},{"name":"John Humphreys","is_ca":false},{"name":"Guyonne Kalb","is_ca":false},{"name":"Julia Witt","is_ca":true},{"name":"Anne Leahy","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7886407742558061,"gpt":0.6233967687334692,"spread":0.165244005522337,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9323868,0.0002898459,0.001583203,0.001450598,0.000264142,0.00001264315,0.0008835383,0.0009388429,0.0004133624],"category_scores_gemma":[0.9640813,0.0002015506,0.000215689,0.001218757,0.005872172,0.0001086586,0.0003626741,0.0009679274,0.000003748583],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002685351,"about_ca_system_score_gemma":0.00968721,"about_ca_topic_candidate":true,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01326669,"about_ca_topic_score_gemma":0.01331928,"domain_scores_codex":[0.03025343,0.9646453,0.001563008,0.000760184,0.001975515,0.0008025475],"domain_scores_gemma":[0.2459978,0.7519389,0.0005079758,0.0006796115,0.000511557,0.0003641399],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"randomized_trial","study_design_gemma":"observational","study_design_scores_codex":[0.9391641,0.0001399504,0.0317282,0.00004601945,0.0000691755,0.00001084495,0.009527367,0.00001754308,0.01698071,0.00005540626,0.00005598933,0.002204712],"study_design_scores_gemma":[0.0849214,0.00675922,0.8841314,0.0001650635,0.00004938393,0.000004349996,0.0009540566,0.0007105719,0.02175693,0.0003625292,0.00003529641,0.0001497778],"study_design_candidate":"randomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9953374,0.0002740576,0.0002962533,0.0004439089,0.0004972285,0.003025277,0.00005409806,0.00001116183,0.00006064522],"genre_scores_gemma":[0.9981073,0.0001449997,0.00122166,0.00002961133,0.00003960355,0.0002743321,0.000001954343,0.00002532132,0.0001551779],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8542427,"threshold_uncertainty_score":0.9968333,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1775522749","doi":"10.1186/1471-2288-5-5","title":"Comparison of nested case-control and survival analysis methodologies for analysis of time-dependent exposure","year":2005,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":235,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"McGill University Health Centre; McGill University","funders":"Canadian Institutes of Health Research; McGill University","keywords":"Nested case-control study; Covariate; Proportional hazards model; Statistics; Regression analysis; Cohort; Nested set model; Cohort study; Regression; Medicine; Survival analysis; Econometrics; Mathematics; Computer science; Data mining","authors":[{"name":"Vidal Essebag","is_ca":false},{"name":"Robert W. Platt","is_ca":true},{"name":"Michał Abrahamowicz","is_ca":true},{"name":"Louise Pilote","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.7964481715181267,"gpt":0.6724582142401484,"spread":0.1239899572779783,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06685697,0.0002266384,0.00296008,0.001947952,0.00009645441,0.0000113509,0.0005493661,0.0005598657,0.0006436655],"category_scores_gemma":[0.2464014,0.0001796553,0.0004829081,0.002379477,0.001117776,0.00008355456,0.0002923502,0.0006170315,0.000001269913],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000071276,"about_ca_system_score_gemma":0.0002892041,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002898198,"about_ca_topic_score_gemma":0.005247874,"domain_scores_codex":[0.9734638,0.02233014,0.001508349,0.0005846072,0.001446263,0.0006668249],"domain_scores_gemma":[0.7551475,0.2427253,0.0004237086,0.0006047116,0.0008549841,0.000243853],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.005934169,0.002121109,0.2859776,0.002736607,0.04593874,0.0002398241,0.007089669,0.007105907,0.07011959,0.1779649,0.001221342,0.3935505],"study_design_scores_gemma":[0.005872359,0.004499071,0.02263741,0.0001682918,0.0326031,0.0001506328,0.01144496,0.4992222,0.09363995,0.328326,0.0002974469,0.001138468],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.1680755,0.0004932604,0.8303967,0.0002299639,0.00002264885,0.0005329034,0.00007442266,0.00007387649,0.0001007758],"genre_scores_gemma":[0.4017955,0.00006916381,0.5978634,0.00001346464,0.00004009245,0.0001038967,0.00001057021,0.00001537254,0.00008848887],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.4921164,"threshold_uncertainty_score":0.9608671,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1990368073","doi":"10.1186/1471-2288-12-111","title":"Investigating clinical heterogeneity in systematic reviews: a methodologic review of guidance in the literature","year":2012,"lang":"en","type":"review","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":226,"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; SickKids Foundation; University of Toronto; Hospital for Sick Children; University of Ottawa","funders":"","keywords":"Systematic review; MEDLINE; CINAHL; Study heterogeneity; Protocol (science); Clinical trial; Medicine; Meta-analysis; Psychology; Alternative medicine; Psychological intervention; Pathology; Nursing; Political science","authors":[{"name":"Joel Gagnier","is_ca":false},{"name":"David Moher","is_ca":true},{"name":"Heather Boon","is_ca":true},{"name":"Joseph Beyene","is_ca":true},{"name":"Claire Bombardier","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9933136472504177,"gpt":0.804038130845693,"spread":0.1892755164047247,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_broad","research_integrity","insufficient_payload"],"category_scores_codex":[0.9805543,0.001137536,0.05856532,0.002429079,0.0001297432,0.0002896066,0.01611515,0.002104706,0.003255606],"category_scores_gemma":[0.9872727,0.0003982818,0.01060323,0.01450854,0.001482716,0.0002110444,0.001742409,0.00710677,0.00154568],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001832127,"about_ca_system_score_gemma":0.00269739,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00003917351,"about_ca_topic_score_gemma":0.0003809578,"domain_scores_codex":[0.01887619,0.8924771,0.06483344,0.002753131,0.01953466,0.00152551],"domain_scores_gemma":[0.01609568,0.939724,0.02673195,0.0135806,0.002829575,0.00103818],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.000002065211,0.00009646996,0.0005615427,0.6425956,0.00009825917,0.00003136013,0.0002232492,6.192201e-8,5.302788e-8,0.001139843,0.004141504,0.35111],"study_design_scores_gemma":[0.0001018834,0.00006492716,0.00009341751,0.5393328,0.0006723785,0.0002085859,0.0001138221,0.00003363115,4.889717e-8,0.002819546,0.4563399,0.0002191088],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"review","genre_scores_codex":[0.000008515306,0.9725053,0.01314172,0.0004745347,0.0005764195,0.01271078,0.00001727229,0.000004462138,0.0005609646],"genre_scores_gemma":[0.000001493392,0.8602348,0.1345865,0.001185456,0.0004138548,0.00324169,0.00002675012,0.00003526211,0.0002742149],"genre_candidate":"review","genre_consensus":"review","teacher_disagreement_score":0.4521984,"threshold_uncertainty_score":0.9998469,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2098756124","doi":"10.1186/1471-2288-10-42","title":"Surname lists to identify South Asian and Chinese ethnicity from secondary data in Ontario, Canada: a validation study","year":2010,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Migration, Health and Trauma","field":"Psychology","cited_by":224,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Health Sciences Centre; University of Toronto; Institute for Clinical Evaluative Sciences; Sunnybrook Health Science Centre","funders":"Canadian Institutes of Health Research; University of Toronto; Ontario Ministry of Health and Long-Term Care; Banting and Best Diabetes Centre, University of Toronto; Heart and Stroke Foundation of Canada; Institute for Clinical Evaluative Sciences; Canadian Diabetes Association","keywords":"Ethnic group; Demography; Medicine; Population; Geography; Gerontology; Family medicine; Environmental health; Political science; Sociology","authors":[{"name":"Baiju R. Shah","is_ca":true},{"name":"Maria Chiu","is_ca":true},{"name":"Meera Ramani","is_ca":true},{"name":"Sharon Sadry","is_ca":true},{"name":"Jack V. Tu","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.4588016625415811,"gpt":0.581583771404613,"spread":0.1227821088630319,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03739393,0.0001788085,0.0005477363,0.0003756974,0.0001712588,0.00003938394,0.001001709,0.0004242607,0.01428737],"category_scores_gemma":[0.05532162,0.0001511829,0.00002241858,0.0004965256,0.0002289119,0.0001053163,0.0006491108,0.003478583,0.00005605097],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001639183,"about_ca_system_score_gemma":0.007996006,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.9961928,"about_ca_topic_score_gemma":0.9999747,"domain_scores_codex":[0.9800643,0.01600151,0.0007529147,0.001092051,0.001205428,0.0008838167],"domain_scores_gemma":[0.9809315,0.01628056,0.00009440604,0.001255665,0.0001838921,0.00125403],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005660341,0.000279357,0.9305989,0.00003141496,0.00004029921,0.0002210498,0.0321146,1.66383e-7,0.0001275647,0.0002155149,0.004845103,0.03096002],"study_design_scores_gemma":[0.001559351,0.0001941644,0.9903327,0.00001438702,0.0000077096,0.00002018652,0.004756568,0.00005337143,0.000005967705,0.001361158,0.001573317,0.0001210935],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9897231,0.00009854235,0.004810698,0.002100015,0.001423521,0.0009200715,0.0001060062,0.00001849443,0.0007994764],"genre_scores_gemma":[0.991194,0.000004912596,0.007051492,0.0005435413,0.0003565168,0.0001158448,0.0002538594,0.00002035204,0.0004595351],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05973385,"threshold_uncertainty_score":0.9988204,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2078863577","doi":"10.1186/1471-2288-9-47","title":"Interviewee Transcript Review: assessing the impact on qualitative research","year":2009,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Qualitative Research Methods and Ethics","field":"Social Sciences","cited_by":193,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"University of Toronto; Cancer Care Ontario","funders":"Canadian Institutes of Health Research","keywords":"Rigour; Qualitative research; Set (abstract data type); Data collection; Psychology; Process (computing); Interview; Computer science; Data science; Applied psychology; Medical education; Medicine; Sociology; Epistemology","authors":[{"name":"Victoria Hagens","is_ca":true},{"name":"Mark Dobrow","is_ca":true},{"name":"Roger Chafe","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9386232119935493,"gpt":0.8327565621737846,"spread":0.1058666498197647,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","sts"],"category_scores_codex":[0.7624618,0.0002861018,0.001013988,0.0007493636,0.003071805,0.0002702974,0.002939679,0.0006859303,0.003987025],"category_scores_gemma":[0.7239269,0.0001598256,0.0004936185,0.003964765,0.01259797,0.0003227265,0.0002644909,0.008804344,0.0003728919],"about_ca_system_candidate":true,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007176773,"about_ca_system_score_gemma":0.008495869,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006601081,"about_ca_topic_score_gemma":0.001672512,"domain_scores_codex":[0.3335936,0.6495872,0.001011894,0.0009305247,0.01186827,0.003008471],"domain_scores_gemma":[0.3471239,0.6446934,0.0001878286,0.001266456,0.004595082,0.002133345],"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":[0.000264321,0.0003517908,0.0000346158,0.0004546533,0.00006555908,0.00006600736,0.1533892,0.000001383591,0.0004501097,0.4155637,0.08612606,0.3432326],"study_design_scores_gemma":[0.0005936954,0.002304284,0.001615379,0.003867009,0.00001341575,0.000009208014,0.1966867,0.0001614809,0.0002091673,0.6240726,0.1701468,0.0003202751],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"commentary","genre_gemma":"empirical","genre_scores_codex":[0.01545477,0.04931316,0.1630811,0.6090173,0.0006546987,0.003534245,0.00002117777,0.0001655778,0.158758],"genre_scores_gemma":[0.3804096,0.3439052,0.219524,0.02868441,0.007014131,0.001389102,0.00004954285,0.0002063428,0.01881778],"genre_candidate":"commentary","genre_consensus":null,"teacher_disagreement_score":0.5803329,"threshold_uncertainty_score":0.998226,"prediction_status":"machine_predicted_unvalidated"},"labels":[{"model":"gemma","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"low"},{"model":"gpt","categories":["metaresearch"],"domain":"methods","study_design":"qualitative","genre":"empirical","about_ca_system":false,"about_ca_topic":false,"confidence":"medium"}],"label_agreement":"agree"},{"id":"W3037840142","doi":"10.1186/s12874-020-01050-7","title":"A conceptual framework for prognostic research","year":2020,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":192,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false},"ca_institutions":"Ontario Tech University; Public Health Ontario; University of Toronto; Canadian Memorial Chiropractic College","funders":"","keywords":"Computer science; Conceptual framework; Data science; MEDLINE; Medicine; Management science; Sociology; Political science","authors":[{"name":"Peter Kent","is_ca":false},{"name":"Carol Cancelliere","is_ca":true},{"name":"Eleanor Boyle","is_ca":false},{"name":"J. David Cassidy","is_ca":true},{"name":"Alice Kongsted","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9537009219251084,"gpt":0.7282200696729361,"spread":0.2254808522521723,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.06224936,0.0002119293,0.0007753365,0.0002979134,0.0003322586,0.00004074993,0.001463801,0.0008808991,0.00193765],"category_scores_gemma":[0.8790135,0.0001741527,0.0001458706,0.001211276,0.003546461,0.00009338793,0.0009364445,0.003833684,0.0001617276],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000130223,"about_ca_system_score_gemma":0.002040245,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002632113,"about_ca_topic_score_gemma":0.00005883664,"domain_scores_codex":[0.9695448,0.02327936,0.0007715838,0.0009381897,0.003534509,0.00193152],"domain_scores_gemma":[0.4421055,0.5537502,0.0001080394,0.0007215974,0.001814165,0.001500465],"domain_codex":null,"domain_gemma":"methods","domain_candidate":null,"domain_consensus":null,"study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0003870403,0.00009540031,0.0002424929,0.0003411472,0.00002338206,0.00004404636,0.001722559,3.55427e-7,0.0007656477,0.9640669,0.01990509,0.0124059],"study_design_scores_gemma":[0.0004387698,0.001658209,0.00002893783,0.0002232677,0.000008778,0.00001293857,0.00283604,0.0008136174,0.004310206,0.970135,0.01936833,0.0001658704],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.007186115,0.0002985679,0.978814,0.01000669,0.0001093218,0.002027963,0.00001328231,0.0003780432,0.001165985],"genre_scores_gemma":[0.04223992,0.0001289467,0.9538759,0.0007487549,0.001038944,0.001693425,0.000007775503,0.00007305294,0.0001932886],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.8167641,"threshold_uncertainty_score":0.9991653,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3035611753","doi":"10.1186/s12874-020-01029-4","title":"Improving design choices in Delphi studies in medicine: the case of an exemplary physician multi-round panel study with 100% response","year":2020,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Delphi Technique in Research","field":"Social Sciences","cited_by":187,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Newfoundland and Labrador; Achmea","keywords":"Delphi method; Delphi; Operations research; Computer science; Psychology; Medical education; Medicine; Management science; Data science; Artificial intelligence","authors":[{"name":"Rebekka Veugelers","is_ca":false},{"name":"Menno I. Gaakeer","is_ca":false},{"name":"P. Patka","is_ca":false},{"name":"Robbert Huijsman","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9143798360928609,"gpt":0.6898360383718415,"spread":0.2245437977210194,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","sts"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2143165,0.0002176599,0.0008887201,0.0007210181,0.0004950828,0.00003463596,0.001885548,0.0003047557,0.00008591428],"category_scores_gemma":[0.3555852,0.0001356154,0.00004545366,0.00325823,0.007526118,0.0002566231,0.0007180095,0.002170261,0.000005345134],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002895926,"about_ca_system_score_gemma":0.003290585,"about_ca_topic_candidate":true,"about_ca_topic_consensus":true,"about_ca_topic_score_codex":0.06959035,"about_ca_topic_score_gemma":0.2055721,"domain_scores_codex":[0.8358136,0.1577104,0.0009275078,0.000900183,0.003298139,0.001350091],"domain_scores_gemma":[0.8231097,0.1748009,0.0001776098,0.0006003765,0.0006844323,0.0006269762],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"qualitative","study_design_gemma":"qualitative","study_design_scores_codex":[0.03073626,0.003068877,0.01855449,0.0009053997,0.0002169143,0.04987479,0.6852881,0.0002372734,0.008375018,0.002041542,0.001184975,0.1995164],"study_design_scores_gemma":[0.00344424,0.007264465,0.007879219,0.0003238507,0.00002386674,0.0001065058,0.9717151,0.006551702,0.0004310289,0.001751766,0.0002445074,0.0002637613],"study_design_candidate":"qualitative","study_design_consensus":"qualitative","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9518112,0.001038548,0.03190736,0.01194258,0.00008315082,0.00303516,0.000002460672,0.00006747597,0.0001121035],"genre_scores_gemma":[0.9458477,0.0002825252,0.05261456,0.0004491828,0.0003012331,0.0004185949,7.78009e-7,0.00003322677,0.00005223908],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2864271,"threshold_uncertainty_score":0.9951748,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3170269751","doi":"10.1186/s12874-021-01310-0","title":"Most published meta-regression analyses based on aggregate data suffer from methodological pitfalls: a meta-epidemiological study","year":2021,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":179,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Public Health Ontario; University of Toronto; St. Michael's Hospital","funders":"Canada Research Chairs","keywords":"Meta-regression; Meta-analysis; Logistic regression; Overfitting; Regression; Regression analysis; Medicine; Confounding; Statistics; Internal medicine; Computer science; Mathematics; Machine learning","authors":[{"name":"Michael Geissbühler","is_ca":false},{"name":"Cesar A. Hincapié","is_ca":true},{"name":"Soheila Aghlmandi","is_ca":false},{"name":"Marcel Zwahlen","is_ca":false},{"name":"Peter Jüni","is_ca":true},{"name":"Bruno R. da Costa","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9961418339580116,"gpt":0.7873935714980298,"spread":0.2087482624599818,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_narrow","metaepi_broad","scholarly_communication","open_science","research_integrity","insufficient_payload"],"consensus_categories":["metaresearch","metaepi_narrow","research_integrity","insufficient_payload"],"category_scores_codex":[0.8850442,0.001409187,0.02894812,0.001804096,0.0007117591,0.001607921,0.01728588,0.001370454,0.3570942],"category_scores_gemma":[0.9872183,0.0004808972,0.007503142,0.009396645,0.001496604,0.0005580562,0.007045828,0.004090142,0.002309508],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001122699,"about_ca_system_score_gemma":0.002949456,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0009274093,"about_ca_topic_score_gemma":0.002271944,"domain_scores_codex":[0.1333469,0.8042513,0.0171131,0.008181944,0.03512034,0.001986516],"domain_scores_gemma":[0.02224213,0.9345885,0.005338127,0.02681954,0.008123185,0.00288854],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"not_applicable","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.00193797,0.008138814,0.05178978,0.000240546,0.0772541,0.005474147,0.0005365816,0.001691679,0.00167288,0.006444358,0.745556,0.09926317],"study_design_scores_gemma":[0.00567313,0.00329757,0.05110065,0.0002441626,0.06140316,0.0001792285,0.008385799,0.4506955,0.001302294,0.185412,0.2298335,0.002473007],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02407368,0.01520603,0.9311513,0.02314473,0.0008747491,0.002303202,0.0003203685,0.00006160203,0.002864338],"genre_scores_gemma":[0.09220821,0.0004834803,0.8897399,0.008670182,0.0009238249,0.001175404,0.0008457802,0.00007992026,0.005873332],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5157225,"threshold_uncertainty_score":0.999926,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W1989853314","doi":"10.1186/1471-2288-9-5","title":"Measuring agreement of administrative data with chart data using prevalence unadjusted and adjusted kappa","year":2009,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":177,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true},"ca_institutions":"Calgary General Hospital; Alberta Bone and Joint Health Institute; University of Calgary","funders":"Canadian Institutes of Health Research; Fondation pour la Recherche Médicale","keywords":"Kappa; Cohen's kappa; Statistics; Medicine; Sampling (signal processing); Sample size determination; Sample (material); Mathematics; Computer science","authors":[{"name":"Guanmin Chen","is_ca":true},{"name":"Peter Faris","is_ca":true},{"name":"Brenda R. Hemmelgarn","is_ca":true},{"name":"Robin L. Walker","is_ca":true},{"name":"Hude Quan","is_ca":true}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9684346225908497,"gpt":0.6570201568897845,"spread":0.3114144657010651,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","open_science","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1828558,0.0002265619,0.0008053505,0.000405469,0.00026327,0.00009626042,0.005753323,0.0002279883,0.001188401],"category_scores_gemma":[0.2484596,0.0001428505,0.00003819934,0.001287191,0.001629139,0.0005030331,0.00259774,0.0008296284,0.00001981652],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007229082,"about_ca_system_score_gemma":0.002306484,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002220004,"about_ca_topic_score_gemma":0.001191019,"domain_scores_codex":[0.9596533,0.02368828,0.001591617,0.002099676,0.01202403,0.0009430707],"domain_scores_gemma":[0.9358465,0.05494029,0.0005086852,0.005941864,0.001830552,0.0009320845],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.005504302,0.002212081,0.1304403,0.001494355,0.000405046,0.0003023226,0.002959083,0.0002245794,0.02797928,0.007033973,0.01093176,0.8105129],"study_design_scores_gemma":[0.009283043,0.009673474,0.6395493,0.004378966,0.0004193304,0.000432117,0.0224212,0.2441572,0.01166218,0.04065389,0.01580456,0.001564746],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.5973017,0.002872126,0.3907014,0.005393614,0.0002711269,0.001746694,0.0002083691,0.00003211892,0.001472812],"genre_scores_gemma":[0.7302495,0.0007399202,0.2682155,0.0002326822,0.0002154479,0.00001454449,0.00005388672,0.00001269744,0.0002658691],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8089482,"threshold_uncertainty_score":0.9997246,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W3157785707","doi":"10.1186/s12874-021-01306-w","title":"Comparison of six statistical methods for interrupted time series studies: empirical evaluation of 190 published series","year":2021,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Advanced Causal Inference Techniques","field":"Mathematics","cited_by":159,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"National Health and Medical Research Council; Medical Research Council; Canadian Institutes of Health Research; Monash University","keywords":"Statistics; Series (stratigraphy); Confidence interval; Autocorrelation; Pairwise comparison; Statistical significance; Sample size determination; Point estimation; Statistical hypothesis testing; Mathematics; Computer science; Econometrics","authors":[{"name":"Simon Turner","is_ca":false},{"name":"Amalia Karahalios","is_ca":false},{"name":"Andrew Forbes","is_ca":false},{"name":"Monica Taljaard","is_ca":true},{"name":"Jeremy Grimshaw","is_ca":true},{"name":"Joanne E. McKenzie","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9326180323716993,"gpt":0.7848555400066978,"spread":0.1477624923650015,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1003847,0.0002281455,0.001893788,0.0003368388,0.0001006562,0.00001977992,0.0005755817,0.000495734,0.002912541],"category_scores_gemma":[0.8162003,0.0001853095,0.0001589553,0.0007868676,0.002471722,0.0002452173,0.0007530133,0.0008640803,0.000004528746],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002149571,"about_ca_system_score_gemma":0.002838004,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0000143714,"about_ca_topic_score_gemma":0.0002437142,"domain_scores_codex":[0.9433912,0.05032922,0.001961401,0.0007085279,0.002831216,0.0007784084],"domain_scores_gemma":[0.7298309,0.2601709,0.0004520231,0.0008463223,0.008353094,0.000346819],"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":[0.002833332,0.001503358,0.002050007,0.005197097,0.000897756,0.00001882266,0.007609781,0.00001578762,0.05075987,0.6059377,0.04562907,0.2775475],"study_design_scores_gemma":[0.0007372991,0.001242457,0.0001572587,0.00025626,0.0001357244,0.00002618773,0.005296077,0.007481461,0.163482,0.8194193,0.001615727,0.0001503334],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.009407206,0.001016476,0.9861456,0.001602622,0.0001879305,0.0008941239,0.00004204737,0.00008873657,0.0006153174],"genre_scores_gemma":[0.008021544,0.0001248586,0.9907952,0.00003865873,0.00009199267,0.0004961076,0.00006191841,0.00003661716,0.000333129],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.7158156,"threshold_uncertainty_score":0.997999,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null},{"id":"W2111680408","doi":"10.1186/1471-2288-11-14","title":"Methodological criteria for the assessment of moderators in systematic reviews of randomised controlled trials: a consensus study","year":2011,"lang":"en","type":"article","venue":"BMC Medical Research Methodology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":158,"is_retracted":false,"has_abstract":true,"routes":{"ca_aff":false,"ca_fund":true,"ca_venue":false,"about_ca":false},"ca_institutions":"","funders":"Keele University; National Institute for Health and Care Research; Advantage West Midlands; Programme Grants for Applied Research; Queen Mary University of London; Université Laval; University of Washington; Arizona State University","keywords":"Systematic review; Delphi method; Randomized controlled trial; MEDLINE; Grading (engineering); Meta-analysis; Psychology; Medicine; Psychological intervention; Applied psychology; Family medicine; Management science; Computer science; Psychiatry","authors":[{"name":"Tamar Pincus","is_ca":false},{"name":"Clare Miles","is_ca":false},{"name":"Robert Froud","is_ca":false},{"name":"Martin Underwood","is_ca":false},{"name":"Dawn Carnes","is_ca":false},{"name":"Stephanie Taylor","is_ca":false}],"retraction":null,"screen_n_in":null,"score":{"opus":0.9949133789799366,"gpt":0.7958935538530266,"spread":0.19901982512691,"validation_status":"score_only:v0-immature-baseline"},"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","metaepi_broad","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.9825663,0.0003908623,0.03844984,0.001189158,0.0001376963,0.00009242968,0.003936392,0.0003569344,0.01563005],"category_scores_gemma":[0.9929546,0.0001215572,0.005670871,0.001949272,0.0009058471,0.00004473475,0.0004177524,0.0007040919,0.00002731114],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00004768868,"about_ca_system_score_gemma":0.001231905,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001353764,"about_ca_topic_score_gemma":0.0003488994,"domain_scores_codex":[0.01841133,0.9271398,0.04147986,0.001223766,0.0110785,0.0006667072],"domain_scores_gemma":[0.006738426,0.9718366,0.01298056,0.00408018,0.00398252,0.00038169],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.1706231,0.02086201,0.08789729,0.168385,0.03585796,0.0003198444,0.06302314,0.0003127648,0.009044252,0.2370648,0.08146366,0.1251461],"study_design_scores_gemma":[0.216147,0.006539559,0.01425028,0.006050466,0.00863631,0.00008299281,0.07506029,0.5035818,0.0004213515,0.1668118,0.001509459,0.0009086414],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.02861499,0.005849766,0.9251194,0.0004240387,0.0004516634,0.03849073,0.000009906174,0.000003220301,0.001036313],"genre_scores_gemma":[0.3972508,0.0005281152,0.5895296,0.0001049471,0.0001062141,0.01199819,0.00000129285,0.00001887818,0.0004619662],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.5032691,"threshold_uncertainty_score":0.9852698,"prediction_status":"machine_predicted_unvalidated"},"labels":[],"label_agreement":null}]}