{"id":"W4401879307","doi":"10.2196/50935","title":"Evaluation of a Natural Language Processing Approach to Identify Diagnostic Errors and Analysis of Safety Learning System Case Review Data: Retrospective Cohort Study","year":2024,"lang":"en","type":"article","venue":"Journal of Medical Internet Research","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":8,"is_retracted":false,"has_abstract":true,"ca_institutions":"Toronto East General Hospital; University of Toronto","funders":"Agency for Healthcare Research and Quality","keywords":"Medicine; Logistic regression; Retrospective cohort study; Patient safety; Diagnosis code; Artificial intelligence; Emergency medicine; Health care; Medical emergency; Surgery; Computer science; Population; Internal medicine","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":[],"consensus_categories":[],"category_scores_codex":[0.03960512,0.000700427,0.0007872015,0.006458589,0.001169012,0.00161238,0.001589651,0.0009499359,0.001259583],"category_scores_gemma":[0.08697637,0.000767254,0.00164821,0.002781168,0.00130855,0.001953126,0.002028781,0.0009473072,0.0005517909],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002522674,"about_ca_system_score_gemma":0.004314067,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00906201,"about_ca_topic_score_gemma":0.01144332,"domain_scores_codex":[0.9732525,0.01207102,0.00475385,0.00383437,0.004945419,0.001142902],"domain_scores_gemma":[0.8909452,0.0387956,0.02811716,0.01261036,0.02703655,0.002495223],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005216329,0.0004540037,0.9896066,0.0001851657,0.0002123917,0.0004552365,0.001357801,0.0001698832,0.0003323878,0.0001041745,0.0005684419,0.006032181],"study_design_scores_gemma":[0.0002407971,0.003890105,0.9769735,0.0002643056,0.0005448865,0.002383829,0.005982045,0.004481999,0.001519666,0.0002337929,0.003380434,0.0001047575],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9926845,0.000295079,0.003118363,0.00008671998,0.00002221261,0.002147771,0.001208473,0.00002622537,0.0004104874],"genre_scores_gemma":[0.9871655,0.0002744093,0.00696692,0.000178077,0.00004552431,0.002895079,0.002179023,0.00002955407,0.0002659961],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03960512,"threshold_uncertainty_score":0.2094544,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1335321804761331,"score_gpt":0.5479332831673962,"score_spread":0.4144011026912631,"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."}}