{"id":"W4409480249","doi":"10.1016/j.cjca.2025.04.008","title":"Evaluating Large Language Models in Cardiovascular Antithrombotic Care: Performance, Accuracy, and Implications for Clinical Practice","year":2025,"lang":"en","type":"article","venue":"Canadian Journal of Cardiology","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":4,"is_retracted":false,"has_abstract":false,"ca_institutions":"London Health Sciences Centre; Western University","funders":"Western University","keywords":"Medicine; Antithrombotic; Clinical Practice; Intensive care medicine; Cardiology; Family medicine","routes":{"ca_aff":true,"ca_fund":true,"ca_venue":true,"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":[0.01613202,0.001264117,0.001424343,0.001476569,0.0006096965,0.004035832,0.001550953,0.001852246,0.00234865],"category_scores_gemma":[0.09510776,0.0005247446,0.001230545,0.001115994,0.000544131,0.002652176,0.001340451,0.002567365,0.001179025],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001531579,"about_ca_system_score_gemma":0.002877932,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.02009995,"about_ca_topic_score_gemma":0.02237568,"domain_scores_codex":[0.9909707,0.006960085,0.0005649598,0.000750025,0.0006106721,0.0001436323],"domain_scores_gemma":[0.8804036,0.1122427,0.001983215,0.001710757,0.002677426,0.0009823198],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.01202447,0.003439022,0.2206689,0.001444749,0.003490576,0.0008410728,0.001463192,0.2992811,0.003071985,0.002315869,0.02169082,0.4302683],"study_design_scores_gemma":[0.0004746428,0.001013276,0.01090676,0.0002312807,0.0006667208,0.000233112,0.0004834133,0.9766994,0.001252461,0.006266148,0.001686674,0.00008620244],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8750373,0.008087247,0.09205417,0.01041388,0.0006371561,0.0006231845,0.006144215,0.003417589,0.003585365],"genre_scores_gemma":[0.9479277,0.001527439,0.04383946,0.001196078,0.0002607968,0.0001970094,0.004138413,0.0001819162,0.0007311975],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.02009995,"threshold_uncertainty_score":0.08531535,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2488629289314998,"score_gpt":0.5283369286599129,"score_spread":0.2794739997284131,"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."}}