{"id":"W4408591745","doi":"10.1093/jamia/ocaf048","title":"Utilizing large language models for detecting hospital-acquired conditions: an empirical study on pulmonary embolism","year":2025,"lang":"en","type":"article","venue":"Journal of the American Medical Informatics Association","topic":"Venous Thromboembolism Diagnosis and Management","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"Libin Cardiovascular Institute of Alberta; Concordia University; University of Calgary; Alberta Health Services","funders":"National Institute on Deafness and Other Communication Disorders; Canadian Institutes of Health Research","keywords":"Medicine; Medical record; Population; Pulmonary embolism; Computer science; Chart; Documentation; Data extraction; Artificial intelligence; Natural language processing; Machine learning; MEDLINE; Internal medicine; Statistics","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":false,"about_ca":true,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.002166427,0.0001350458,0.000609409,0.0002108237,0.0002461734,0.00005234768,0.0002702995,0.00007598083,0.00001323326],"category_scores_gemma":[0.002041763,0.00009163897,0.0002352829,0.0003921131,0.0000450397,0.0002636877,0.00009392006,0.0003153086,0.000002097474],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004563549,"about_ca_system_score_gemma":0.0002242906,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002172348,"about_ca_topic_score_gemma":0.000007882965,"domain_scores_codex":[0.9975498,0.0001407872,0.0009119733,0.00008169792,0.001033209,0.0002824792],"domain_scores_gemma":[0.9974703,0.0004170351,0.001448651,0.0002421686,0.0002650394,0.0001567985],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001694503,0.05107939,0.3275231,0.001390397,0.01064145,0.0002390933,0.181639,0.003655815,0.0007142173,0.006420999,0.1315662,0.2834359],"study_design_scores_gemma":[0.007257203,0.006140722,0.55159,0.00138634,0.002005317,0.00003274314,0.2422647,0.182723,0.0003995409,0.002548586,0.003214382,0.0004375545],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9877117,0.00003328154,0.004708071,0.005165002,0.0005151668,0.0007448216,0.00001145378,0.00002577305,0.001084716],"genre_scores_gemma":[0.9909534,0.00004604712,0.0005417649,0.00810479,0.0002269225,0.00003217024,0.000007968481,0.0000127579,0.00007411152],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.2829984,"threshold_uncertainty_score":0.3736929,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01764632910414123,"score_gpt":0.3565085571870042,"score_spread":0.338862228082863,"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."}}