{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.02309732,0.0008191898,0.0006014127,0.001573544,0.0006419861,0.001521656,0.001138929,0.0008282433,0.001011557],"category_scores_gemma":[0.09121164,0.000396972,0.001579947,0.001325254,0.0008549091,0.002792793,0.001278974,0.001798649,0.0003069607],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002107024,"about_ca_system_score_gemma":0.001869683,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02586498,"about_ca_topic_score_gemma":0.02725195,"domain_scores_codex":[0.9908827,0.006860662,0.0005608637,0.0008797838,0.000613815,0.0002022445],"domain_scores_gemma":[0.8069038,0.1805596,0.004548488,0.00348367,0.003653483,0.000850858],"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":[0.001440177,0.002867156,0.8927175,0.0006038755,0.0009348298,0.0005990195,0.002275378,0.03331523,0.001096587,0.0006867134,0.002382491,0.06108088],"study_design_scores_gemma":[0.0003351762,0.001691894,0.3371791,0.00031691,0.0008011555,0.0007505875,0.004109273,0.6481089,0.001945486,0.002317584,0.002349261,0.00009466455],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9940515,0.0003372842,0.004324663,0.000300525,0.000009934992,0.0001230701,0.000512493,0.0000538975,0.0002864844],"genre_scores_gemma":[0.9907165,0.0001760776,0.007092563,0.00008711098,0.0000146791,0.0000967975,0.001644391,0.00001817772,0.000153633],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02586498,"threshold_uncertainty_score":0.1221518,"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."}}