{"id":"W4415925415","doi":"10.2196/74899","title":"Assessing the Limitations of Large Language Models in Clinical Practice Guideline–Concordant Treatment Decision-Making on Real-World Data: Retrospective Study","year":2025,"lang":"en","type":"article","venue":"JMIRx Med","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Clinical Practice; Retrospective cohort study; MEDLINE; Risk assessment; Health care","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":["metaresearch"],"consensus_categories":["metaresearch"],"category_scores_codex":[0.2341895,0.0009324614,0.001334057,0.005104819,0.001074594,0.003443598,0.002353171,0.001200555,0.001330327],"category_scores_gemma":[0.5769923,0.001333787,0.002257527,0.005345094,0.002796499,0.004284719,0.004292798,0.001482568,0.0004789378],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002444301,"about_ca_system_score_gemma":0.00452867,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004844885,"about_ca_topic_score_gemma":0.004820853,"domain_scores_codex":[0.7146604,0.2157834,0.03353852,0.01387805,0.02018082,0.001958846],"domain_scores_gemma":[0.2318034,0.5570199,0.1069446,0.06681275,0.03508078,0.00233849],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009833967,0.0002340453,0.9696338,0.0006529055,0.001058749,0.00028903,0.005872671,0.002447559,0.0001249728,0.0006981621,0.00127408,0.01673082],"study_design_scores_gemma":[0.0009200406,0.004014007,0.832599,0.004139188,0.00369073,0.004163775,0.01694408,0.10097,0.003768824,0.007702289,0.02051746,0.0005705919],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.960203,0.001772919,0.02892441,0.001009428,0.00008306662,0.002123487,0.003521748,0.0001124532,0.002249464],"genre_scores_gemma":[0.9866523,0.0002088372,0.009675327,0.0003112066,0.00004702708,0.001294119,0.001711458,0.00003837587,0.00006145034],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.7658105,"threshold_uncertainty_score":0.9443806,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4364023688737394,"score_gpt":0.629305263558597,"score_spread":0.1929028946848576,"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."}}