{"id":"W4414418115","doi":"10.21203/rs.3.rs-7652676/v1","title":"AI-driven quality assurance for emergency department documentation: Pilot comparison with physician peer review of chest pain cases","year":2025,"lang":"en","type":"preprint","venue":"Research Square","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"Niagara Health System; McMaster University","funders":"","keywords":"Quality assurance; Concordance; Emergency department; Documentation; Kappa; Chest pain; Audit; Checklist","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":[0.05381588,0.0004722484,0.0008827165,0.001987054,0.002141239,0.003267521,0.002063351,0.001472274,0.006328421],"category_scores_gemma":[0.2361487,0.0006098726,0.0009263927,0.001385925,0.001172153,0.002972874,0.003378895,0.001605944,0.001472886],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003152551,"about_ca_system_score_gemma":0.01026756,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00738041,"about_ca_topic_score_gemma":0.006962107,"domain_scores_codex":[0.9429963,0.04047405,0.004874279,0.002899251,0.007487253,0.001268969],"domain_scores_gemma":[0.5975335,0.2747878,0.02423291,0.02575359,0.06587581,0.01181651],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"nonrandomized_trial","study_design_scores_codex":[0.03116998,0.04752539,0.172183,0.005271625,0.0008237428,0.0005944658,0.04481285,0.003998884,0.009473481,0.001998244,0.0102799,0.6718685],"study_design_scores_gemma":[0.02017659,0.1279125,0.677681,0.002450028,0.002042628,0.001717633,0.02956105,0.05997099,0.02600068,0.003612406,0.04794041,0.0009341131],"study_design_candidate":"nonrandomized_trial","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9804888,0.0004707728,0.00779683,0.00134991,0.0002008125,0.00440802,0.0004090155,0.0006825902,0.004193319],"genre_scores_gemma":[0.96443,0.0003285414,0.02975947,0.000616962,0.0001422661,0.002355156,0.0005220021,0.0001178617,0.001727917],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9461841,"threshold_uncertainty_score":0.284609,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5082447121263785,"score_gpt":0.6430492248690679,"score_spread":0.1348045127426895,"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."}}