{"id":"W3191198632","doi":"10.1016/j.jclinepi.2021.07.021","title":"Evaluating tests for diagnosing COVID-19 in the absence of a reliable reference standard: pitfalls and potential solutions","year":2021,"lang":"en","type":"article","venue":"Journal of Clinical Epidemiology","topic":"SARS-CoV-2 detection and testing","field":"Medicine","cited_by":14,"is_retracted":false,"has_abstract":false,"ca_institutions":"Ottawa Hospital; University of Ottawa","funders":"","keywords":"Coronavirus disease 2019 (COVID-19); Severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2); Diagnostic test; 2019-20 coronavirus outbreak; Pandemic; Diagnostic accuracy; Medicine; Computer science; Pathology; Pediatrics; 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":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.03232184,0.00008599101,0.001110801,0.00008120155,0.00009986332,0.000006468731,0.00009801064,0.0001785853,0.00001694429],"category_scores_gemma":[0.4606438,0.0000572419,0.0002279627,0.0001817648,0.000224945,0.00005579932,0.00004585207,0.0007161482,5.910376e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005024845,"about_ca_system_score_gemma":0.001044532,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005008403,"about_ca_topic_score_gemma":0.00003835935,"domain_scores_codex":[0.9948926,0.001879133,0.002575525,0.0002032635,0.000201002,0.0002484529],"domain_scores_gemma":[0.9066836,0.0897411,0.002296789,0.0002702312,0.0008224257,0.0001859125],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002608411,0.0007182879,0.9171392,0.0004987572,0.0002152263,0.000567351,0.0004856161,0.002136548,0.02830393,0.005200926,0.004392225,0.03773348],"study_design_scores_gemma":[0.02037223,0.01181746,0.7771398,0.002353936,0.0006876888,0.01319372,0.001899175,0.04401549,0.002285883,0.107197,0.01863218,0.0004054693],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9639865,0.001366128,0.02309164,0.01084556,0.0003768418,0.000176508,0.000006129485,0.000006237227,0.0001445003],"genre_scores_gemma":[0.949478,0.0003373715,0.03990345,0.009946759,0.0003116112,0.000005736772,0.000001188939,0.000006523794,0.000009393564],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.428322,"threshold_uncertainty_score":0.9964283,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.6092105048534818,"score_gpt":0.6004669986547454,"score_spread":0.008743506198736428,"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."}}