{"id":"W2785817299","doi":"10.1136/bmj.j5779","title":"Concerns about composite reference standards in diagnostic research","year":2018,"lang":"en","type":"article","venue":"BMJ","topic":"Statistical Methods and Bayesian Inference","field":"Mathematics","cited_by":50,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University Health Centre","funders":"Canadian Institutes of Health Research","keywords":"Composite number; Imperfect; Reference model; Reference data; Computer science; Test (biology); Statistics; Class (philosophy); Reference values; Econometrics; Data mining; Artificial intelligence; Mathematics; Algorithm; Medicine","routes":{"ca_aff":true,"ca_fund":true,"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":[],"category_scores_codex":[0.003442553,0.000102427,0.0002344998,0.00009491509,0.0001106391,0.00006108388,0.0002438243,0.00008510786,0.0009123153],"category_scores_gemma":[0.02747883,0.00008587415,0.00002006655,0.0003096601,0.000413684,0.00004820665,0.0001163913,0.0003269701,0.0001552887],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000115339,"about_ca_system_score_gemma":0.0001786961,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001625315,"about_ca_topic_score_gemma":0.0002603311,"domain_scores_codex":[0.9977909,0.0005282476,0.0003907999,0.0002519551,0.000641834,0.0003962259],"domain_scores_gemma":[0.9885413,0.01039011,0.00006417189,0.0003660686,0.0005353831,0.0001029723],"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":[0.00006046275,0.00008019423,0.005497652,0.0000765853,0.000008248358,0.00006427859,0.0005545627,5.195135e-8,0.000317973,0.9188441,0.03739753,0.03709836],"study_design_scores_gemma":[0.0003980453,0.0002576809,0.02670224,0.0004415553,0.000007057666,0.00000646095,0.00009786763,0.0002527354,0.001028078,0.955546,0.01509179,0.0001704958],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"empirical","genre_scores_codex":[0.09460802,0.0001435416,0.6548077,0.00172585,0.0004382254,0.001197735,0.0004274672,0.0001241252,0.2465273],"genre_scores_gemma":[0.634203,0.00002896628,0.3649726,0.00008481,0.000349788,0.0000679195,0.000002468706,0.00001718131,0.0002733458],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.5395949,"threshold_uncertainty_score":0.9989213,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2694314483759581,"score_gpt":0.5502487348604476,"score_spread":0.2808172864844895,"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."}}