{"id":"W2041996150","doi":"10.1148/radiol.2483072067","title":"Quality of Reporting of Diagnostic Accuracy Studies: No Change Since STARD Statement Publication—Before-and-after Study","year":2008,"lang":"en","type":"article","venue":"Radiology","topic":"Meta-analysis and systematic reviews","field":"Decision Sciences","cited_by":62,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"U.S. National Library of Medicine","keywords":"Checklist; Diagnostic accuracy; Medicine; Medical physics; Statement (logic); Significant difference; Psychology; Internal medicine; Political science","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.1069164,0.0001904332,0.004113163,0.0002361652,0.00008190205,0.00002837219,0.0006028882,0.00004978281,0.0007845399],"category_scores_gemma":[0.4599957,0.00009488128,0.0004805142,0.0006133928,0.0002271423,0.0002186164,0.000214694,0.00008200677,0.0001444413],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003536245,"about_ca_system_score_gemma":0.00005781427,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00007928379,"about_ca_topic_score_gemma":0.0001628122,"domain_scores_codex":[0.9619524,0.0100226,0.02462234,0.0009210645,0.002215888,0.0002656814],"domain_scores_gemma":[0.9481145,0.01946938,0.02641437,0.002607777,0.003301297,0.00009269606],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.00001297262,0.0001457897,0.9846187,0.000107743,0.000383133,0.00001029779,0.007425939,0.000002108244,0.00002841209,0.0002502378,0.003230004,0.00378462],"study_design_scores_gemma":[0.0002357792,0.0002699905,0.9893191,0.00002084876,0.0000967956,0.00002102957,0.002458765,0.00007006732,0.000007936899,0.0004740279,0.006935981,0.00008964001],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9945578,0.002532023,0.000623411,0.0006565074,0.000172594,0.001255046,0.00002473467,0.000003566947,0.0001743159],"genre_scores_gemma":[0.9974618,0.0003555689,0.001223247,0.000157554,0.00005989879,0.0002730519,0.000005701022,0.000005502907,0.0004576644],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3530793,"threshold_uncertainty_score":0.9196175,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.8500857908818524,"score_gpt":0.5980544114885218,"score_spread":0.2520313793933306,"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."}}