{"id":"W4410293035","doi":"10.2196/68527","title":"The Advanced Reasoning Capabilities of Large Language Models for Detecting Contraindicated Options in Medical Exams","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Clinical Reasoning and Diagnostic Skills","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Computer science; Clinical Practice; Sample (material); Medical physics; Medical practice; Medicine; Medical education; Family medicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":true,"about_ca":false,"invisible_to_affiliation_only":true},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0103593,0.001247245,0.0006333331,0.001507466,0.000523876,0.003208794,0.001402405,0.001021151,0.003728141],"category_scores_gemma":[0.0632377,0.000511843,0.001424566,0.0008417993,0.0006893385,0.004882291,0.002104261,0.001899574,0.0007351419],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001009047,"about_ca_system_score_gemma":0.002422494,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01135036,"about_ca_topic_score_gemma":0.01415384,"domain_scores_codex":[0.9961283,0.002254348,0.0003458881,0.0005808701,0.0005365824,0.0001538597],"domain_scores_gemma":[0.9361519,0.05619617,0.002747169,0.002031527,0.002145068,0.000728222],"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":[0.006508448,0.002980721,0.1986218,0.002018775,0.001322358,0.001029277,0.002700784,0.1148855,0.01615992,0.006294532,0.01222162,0.6352563],"study_design_scores_gemma":[0.0006621007,0.0009893056,0.02789841,0.0006578272,0.001255603,0.0009868913,0.001042805,0.9144877,0.01252565,0.0319474,0.007375188,0.0001711764],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.68642,0.002547776,0.2877394,0.004941917,0.0002077826,0.001017014,0.004737811,0.005810506,0.00657776],"genre_scores_gemma":[0.8680111,0.0004756042,0.1276295,0.0005505558,0.00007262913,0.0002427123,0.002234494,0.0001166623,0.0006666441],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01135036,"threshold_uncertainty_score":0.05478591,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01534813934814373,"score_gpt":0.3691452342738435,"score_spread":0.3537970949256998,"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."}}