{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.313121,0.003095539,0.01133026,0.03200334,0.002475356,0.009705089,0.008378131,0.008604603,0.02655604],"category_scores_gemma":[0.6664891,0.004644138,0.01919458,0.03068112,0.003002899,0.007042093,0.009319081,0.01271277,0.01437459],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01046516,"about_ca_system_score_gemma":0.03741789,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.006916833,"about_ca_topic_score_gemma":0.009124779,"domain_scores_codex":[0.4680649,0.1809464,0.3089098,0.00424505,0.03485601,0.002977778],"domain_scores_gemma":[0.1917789,0.4359993,0.1405132,0.03073116,0.1971063,0.003871063],"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":[0.001354627,0.0001573167,0.005236527,0.1837336,0.004756498,0.0002313685,0.001248512,0.001212469,0.0009260357,0.009375197,0.6174568,0.1743111],"study_design_scores_gemma":[0.00303663,0.0004920908,0.01805414,0.1663788,0.005572698,0.0008203996,0.0007938356,0.001880089,0.002146833,0.02408654,0.7758834,0.0008545122],"study_design_candidate":"not_applicable","study_design_consensus":"not_applicable","genre_codex":"dataset","genre_gemma":"methods","genre_scores_codex":[0.004604215,0.1320061,0.214905,0.06668297,0.03325991,0.09257013,0.4312622,0.007456454,0.01725298],"genre_scores_gemma":[0.01401999,0.07065465,0.5170226,0.02045311,0.003638732,0.1991377,0.1680894,0.002100537,0.004883212],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.686879,"threshold_uncertainty_score":0.8470442,"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."}}