{"id":"W4416901397","doi":"10.15441/ceem.25.153","title":"Automated chain-of-thought evaluation framework for large language model–generated emergency department documentation: a simulation-based study","year":2025,"lang":"en","type":"article","venue":"Clinical and Experimental Emergency Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Samsung","keywords":"Emergency department; Workload; Documentation; Emergency medical services; Expert system","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":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001492084,0.0002039151,0.0004832259,0.0001485222,0.0002040661,0.000003359192,0.00007586956,0.0001611278,0.001541098],"category_scores_gemma":[0.00164498,0.000165765,0.0001225209,0.0004702582,0.00006796312,0.00007193599,0.000026312,0.0001496649,0.000005077988],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00007375416,"about_ca_system_score_gemma":0.0002206668,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009728137,"about_ca_topic_score_gemma":0.00002645171,"domain_scores_codex":[0.9970076,0.0001921872,0.001710586,0.0004512143,0.0003878827,0.0002505447],"domain_scores_gemma":[0.9983612,0.0003824365,0.0002612746,0.0003102588,0.0005105494,0.0001743286],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.008168879,0.02482067,0.8067091,0.001627467,0.001425657,0.000008610821,0.03678659,0.0339778,0.01012028,0.01096679,0.02534781,0.0400404],"study_design_scores_gemma":[0.001669411,0.002469114,0.007501968,0.0002179926,0.0004566988,1.917623e-7,0.01217072,0.9661036,0.007074256,0.001899211,0.0002634646,0.0001733353],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9543952,0.003845552,0.03399389,0.001648176,0.002808803,0.002967398,0.00001465663,0.000109367,0.0002169381],"genre_scores_gemma":[0.996084,0.0001216153,0.001609928,0.0003543125,0.0004715589,0.0005696483,0.000306475,0.00001808149,0.0004643315],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9321258,"threshold_uncertainty_score":0.9993716,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2238047675465595,"score_gpt":0.5876967528501084,"score_spread":0.3638919853035489,"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."}}