{"id":"W4414887759","doi":"10.2196/73941","title":"Application of Large Language Models in Complex Clinical Cases: Cross-Sectional Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Boosting (machine learning); Language model; Process (computing); Modeling language","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":[0.0227495,0.0006585794,0.0007203686,0.001820169,0.0005141682,0.001190745,0.0007763969,0.0007437143,0.001580905],"category_scores_gemma":[0.07138531,0.0005388244,0.001547143,0.001268256,0.0007452733,0.00211065,0.00115652,0.001074415,0.0005631587],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001320501,"about_ca_system_score_gemma":0.001179536,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.002553098,"about_ca_topic_score_gemma":0.003200962,"domain_scores_codex":[0.9873274,0.008276516,0.001346549,0.001236499,0.001480959,0.0003319908],"domain_scores_gemma":[0.8900032,0.07379251,0.01551417,0.005311654,0.01264203,0.002736321],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001287289,0.004033003,0.9655321,0.000289385,0.0005492537,0.0004237674,0.002520704,0.002038768,0.0002591449,0.0001406975,0.001097425,0.02182854],"study_design_scores_gemma":[0.0006242212,0.01334935,0.929239,0.0003451207,0.001215468,0.002440586,0.008315713,0.03893263,0.00145044,0.0005775043,0.00334363,0.0001664984],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9980863,0.0001643704,0.0009330753,0.00005610172,0.000006312545,0.0002722431,0.0002351591,0.00001301473,0.000233271],"genre_scores_gemma":[0.9964078,0.0002096695,0.002041578,0.00006797223,0.0000162896,0.0003752767,0.0007576406,0.00001327747,0.0001104677],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0227495,"threshold_uncertainty_score":0.1203123,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.2921922517435435,"score_gpt":0.5974429379351518,"score_spread":0.3052506861916083,"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."}}