{"id":"W3020992444","doi":"10.1136/bmj.h5527","title":"STARD 2015: an updated list of essential items for reporting diagnostic accuracy studies","year":2015,"lang":"en","type":"article","venue":"BMJ","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":3338,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Public Health; Ottawa Hospital; University of Ottawa","funders":"Medical Research Council; Universiteit van Amsterdam; Cancer Research UK","keywords":"Diagnostic accuracy; Medical physics; Computer science; Transparency (behavior); Medicine; Information retrieval; Data science; Data mining; Radiology; Computer security","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2565274,0.002865887,0.01056784,0.03161434,0.002177505,0.008197298,0.007572194,0.007785378,0.0324133],"category_scores_gemma":[0.6478366,0.004156496,0.01770508,0.02963432,0.002599191,0.006321938,0.008642969,0.01199568,0.01536664],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01006504,"about_ca_system_score_gemma":0.03514027,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.007564794,"about_ca_topic_score_gemma":0.009846183,"domain_scores_codex":[0.5433458,0.1658789,0.2497357,0.004079354,0.03419725,0.002763009],"domain_scores_gemma":[0.2345867,0.4250545,0.1307322,0.0253583,0.1805761,0.003692241],"domain_codex":null,"domain_gemma":"reporting","domain_candidate":"reporting","domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.001173421,0.0001207337,0.004107505,0.195233,0.004548113,0.000189508,0.001030675,0.0009419183,0.0007694674,0.007523753,0.6184051,0.1659569],"study_design_scores_gemma":[0.003109179,0.0004491212,0.01715721,0.1935441,0.00614579,0.0007755656,0.000614468,0.001581197,0.001683646,0.02305851,0.7510846,0.0007965435],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"methods","genre_scores_codex":[0.003870465,0.1581781,0.170669,0.06342743,0.02913742,0.077897,0.4737704,0.006568395,0.01648165],"genre_scores_gemma":[0.01459114,0.09574214,0.4636414,0.02084887,0.003901499,0.194378,0.199105,0.00229534,0.005496496],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.7434726,"threshold_uncertainty_score":0.9168341,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8654628643617654,"score_gpt":0.6438387366617384,"score_spread":0.221624127700027,"validation_status":"score_only:v0-immature-baseline","note":"Baseline scores from an immature model (maturity gate not passed). Scores rank; they never assert a category."}}