{"id":"W4414189579","doi":"10.1136/bmjdhai-2025-000014","title":"Optimising large language models for clinical information extraction: a benchmarking study in the context of ulcerative colitis research","year":2025,"lang":"en","type":"article","venue":"BMJ Digital Health & AI","topic":"Biomedical Text Mining and Ontologies","field":"Biochemistry, Genetics and Molecular Biology","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"Artificial Intelligence in Medicine (Canada)","funders":"Clinical and Translational Science Institute, University of California, Los Angeles; Clinical and Translational Science Institute, University of California, San Francisco","keywords":"Benchmarking; Context (archaeology); Adaptation (eye); Oracle; Language model; Set (abstract data type); Predictive modelling; Colonoscopy","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.003135381,0.00006779822,0.0001727392,0.00008161768,0.0001251392,0.00006812927,0.0001553395,0.0001075576,0.000001242662],"category_scores_gemma":[0.0007589741,0.00005115116,0.00005446397,0.0001860279,0.0001006369,0.00002592824,0.00007804556,0.00019462,7.020527e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00002795718,"about_ca_system_score_gemma":0.0003760414,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00008527783,"about_ca_topic_score_gemma":0.0003649863,"domain_scores_codex":[0.9986291,0.0002508255,0.0005181264,0.0001560127,0.0001907834,0.0002551466],"domain_scores_gemma":[0.9991369,0.0003529517,0.0001056733,0.0001925436,0.0001730945,0.00003884273],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"qualitative","study_design_scores_codex":[0.001223497,0.0007079646,0.004263274,0.0002702049,0.00005901367,0.00000408344,0.009693703,0.00003465832,0.00001150035,0.002614143,0.02724208,0.9538759],"study_design_scores_gemma":[0.01490548,0.01626605,0.1199334,0.001157368,0.00003137571,0.00003709017,0.4740117,0.02635313,0.0005843525,0.003878626,0.3421566,0.0006847409],"study_design_candidate":"design_other","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9427738,0.001566821,0.03648613,0.00918843,0.0003821417,0.0040428,0.000298892,0.00002197452,0.00523907],"genre_scores_gemma":[0.9976831,0.00006859194,0.0004816547,0.001328257,0.0001058163,0.0001255873,0.0001238875,0.000003097287,0.00008002219],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9531912,"threshold_uncertainty_score":0.2085884,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1149706351165215,"score_gpt":0.5230203048782074,"score_spread":0.4080496697616859,"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."}}