{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.100392,0.003276992,0.004454514,0.006010008,0.001769673,0.008736542,0.006716465,0.006438105,0.001725946],"category_scores_gemma":[0.2263684,0.001937002,0.001748449,0.003181682,0.003831552,0.005694546,0.002909578,0.005438194,0.001684352],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003345822,"about_ca_system_score_gemma":0.006001317,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01285156,"about_ca_topic_score_gemma":0.01626152,"domain_scores_codex":[0.907714,0.04042246,0.01823361,0.006316025,0.02359156,0.003722395],"domain_scores_gemma":[0.7549811,0.1540424,0.01834428,0.009187403,0.05992084,0.003523813],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"theoretical_or_conceptual","study_design_scores_codex":[0.0015046,0.0008072885,0.6841383,0.002956079,0.0008488484,0.005975454,0.003186044,0.004502489,0.01094693,0.008939194,0.03200345,0.2441914],"study_design_scores_gemma":[0.0007748022,0.006111324,0.4127462,0.02104291,0.003724597,0.04655071,0.02852187,0.1379121,0.08458132,0.0955675,0.1612769,0.00118978],"study_design_candidate":"theoretical_or_conceptual","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"methods","genre_scores_codex":[0.4066867,0.1364142,0.220573,0.1664022,0.01562824,0.002040799,0.002212507,0.002676622,0.04736584],"genre_scores_gemma":[0.8061445,0.01026898,0.1567705,0.01966409,0.003577643,0.0003491317,0.001032803,0.0002765805,0.001915862],"genre_candidate":"methods","genre_consensus":null,"teacher_disagreement_score":0.899608,"threshold_uncertainty_score":0.5309299,"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."}}