{"id":"W2188214922","doi":"10.1373/clinchem.2015.246280","title":"STARD 2015: An Updated List of Essential Items for Reporting Diagnostic Accuracy Studies","year":2015,"lang":"en","type":"article","venue":"Clinical Chemistry","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":974,"is_retracted":false,"has_abstract":true,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"Medical Research Council; Universiteit van Amsterdam; Cancer Research UK","keywords":"Diagnostic accuracy; Medical physics; Transparency (behavior); Computer science; Medicine; Radiology","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[{"model":"gemma","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"},{"model":"gpt","categories":["metaresearch"],"domain":"reporting","study_design":"not_applicable","genre":"methods","about_ca_system":false,"about_ca_topic":false,"confidence":"high","status":"direct model label, unvalidated"}],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch","insufficient_payload"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1496173,0.0002375983,0.004014397,0.00003636079,0.00007252159,0.0002406097,0.001295738,0.0001709313,0.001199895],"category_scores_gemma":[0.7869868,0.000126219,0.001703045,0.0004407183,0.0002363478,0.0002053803,0.0002316152,0.0001941444,0.0001226329],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0000214899,"about_ca_system_score_gemma":0.0002007265,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000004732775,"about_ca_topic_score_gemma":0.000005760887,"domain_scores_codex":[0.9702978,0.001776653,0.02422482,0.001031855,0.002424081,0.0002447733],"domain_scores_gemma":[0.8930175,0.05067478,0.04204229,0.005904796,0.007624142,0.0007365029],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"not_applicable","study_design_gemma":"not_applicable","study_design_scores_codex":[0.00005311092,0.0003546254,0.2769082,0.000247385,0.0006468375,0.00001657628,0.0002793766,0.00003951171,0.0007436201,0.00003943172,0.7158048,0.004866516],"study_design_scores_gemma":[0.004987849,0.0007590501,0.01793949,0.0005856334,0.004383276,0.00008272377,0.02037124,0.01907947,0.03031301,0.03510684,0.864455,0.001936387],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9948934,0.00129239,0.001178758,0.0006574469,0.0005774946,0.0004354421,0.0000394251,0.00001035751,0.0009152963],"genre_scores_gemma":[0.9912025,0.00003329765,0.002695317,0.00008734556,0.0006501959,0.00003884608,0.00007958349,0.00001327045,0.005199596],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.6373695,"threshold_uncertainty_score":0.9997131,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.91801681553035,"score_gpt":0.6911491831444544,"score_spread":0.2268676323858956,"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."}}