{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.7773238,0.00205243,0.005962058,0.010219,0.005140701,0.01196718,0.01625126,0.01196485,0.004996222],"category_scores_gemma":[0.8733042,0.00303393,0.004749158,0.01271829,0.04097485,0.01905654,0.0130587,0.02578052,0.002123053],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.01400218,"about_ca_system_score_gemma":0.01041786,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0082678,"about_ca_topic_score_gemma":0.0064206,"domain_scores_codex":[0.1939753,0.6534689,0.03999358,0.02763188,0.08294727,0.00198304],"domain_scores_gemma":[0.06046553,0.83634,0.02372402,0.04849384,0.02915404,0.001822628],"domain_codex":"methods","domain_gemma":"methods","domain_candidate":"methods","domain_consensus":"methods","study_design_codex":"theoretical_or_conceptual","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.000795577,0.0001211485,0.01241221,0.003098497,0.0009429744,0.0003862316,0.008434718,0.003743071,0.0004032507,0.7819633,0.03173224,0.1559668],"study_design_scores_gemma":[0.0001747171,0.0002853437,0.005111328,0.004910392,0.0002644934,0.001056922,0.001034239,0.007986789,0.0009119294,0.9301335,0.04792982,0.0002005849],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":"theoretical_or_conceptual","genre_codex":"methods","genre_gemma":"methods","genre_scores_codex":[0.01146539,0.03695647,0.7435386,0.1681825,0.008644762,0.001759217,0.00113836,0.0007243094,0.02759043],"genre_scores_gemma":[0.2691359,0.0100796,0.610427,0.08557947,0.01254792,0.007234081,0.001009122,0.0006268661,0.00336005],"genre_candidate":"methods","genre_consensus":"methods","teacher_disagreement_score":0.2226762,"threshold_uncertainty_score":0.2745995,"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."}}