{"id":"W4417070771","doi":"10.1002/sim.70327","title":"Likelihood‐Based Non‐Parametric Receiver Operating Characteristic Curve Analysis in the Presence of Imperfect Reference Standard","year":2025,"lang":"en","type":"article","venue":"Statistics in Medicine","topic":"Clinical Laboratory Practices and Quality Control","field":"Medicine","cited_by":1,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Waterloo","funders":"Natural Sciences and Engineering Research Council of Canada","keywords":"Receiver operating characteristic; Imperfect; Gold standard (test); Index (typography); Standard error; Mean squared error","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":[0.004767082,0.0001751922,0.0009915204,0.0006782923,0.00004949501,0.00001866757,0.0002428043,0.0001055452,0.0002884932],"category_scores_gemma":[0.02702611,0.0001127552,0.00005240219,0.003841348,0.0002429261,0.00005668215,0.00003501547,0.0007394471,0.000003167497],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00009333881,"about_ca_system_score_gemma":0.0003999948,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001049595,"about_ca_topic_score_gemma":0.0009743704,"domain_scores_codex":[0.9969497,0.0006806646,0.001143943,0.0003563575,0.0005932215,0.000276113],"domain_scores_gemma":[0.9885052,0.01007114,0.0003338409,0.0005686787,0.0004437073,0.00007747163],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002524563,0.0008591564,0.9469422,0.001276114,0.000692957,0.0003957996,0.002141932,0.0004102513,0.002254693,0.009754292,0.003510926,0.02923715],"study_design_scores_gemma":[0.006070767,0.001512024,0.9236277,0.001144741,0.001844269,0.000001541869,0.0013281,0.06129242,0.00008882293,0.0009496787,0.001951732,0.0001882012],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8794856,0.0009858091,0.1030449,0.009271734,0.0004126521,0.001737414,0.0008517142,0.00002492216,0.004185319],"genre_scores_gemma":[0.9928173,0.0002709134,0.004591649,0.002009873,0.0000564213,0.00003326682,0.0001395563,0.000007731382,0.00007328035],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1133317,"threshold_uncertainty_score":0.9811696,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04203627574508664,"score_gpt":0.4171302863718228,"score_spread":0.3750940106267361,"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."}}