{"id":"W4414759776","doi":"10.2196/72005","title":"Cancer Diagnosis Categorization in Electronic Health Records Using Large Language Models and BioBERT: Model Performance Evaluation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Cancer","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Documentation; Categorization; Health records; Cancer; Electronic health record; MEDLINE; Data collection; Health informatics","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":[0.01491372,0.002176335,0.001354266,0.002784827,0.0005505802,0.002214465,0.001648452,0.001485309,0.0007210703],"category_scores_gemma":[0.02612879,0.000482916,0.001868473,0.001803761,0.0004955447,0.001581707,0.001269754,0.001477823,0.0004087406],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.003663479,"about_ca_system_score_gemma":0.001904237,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0293464,"about_ca_topic_score_gemma":0.01426623,"domain_scores_codex":[0.9958697,0.002551664,0.0003531314,0.0006207119,0.0004254241,0.0001792827],"domain_scores_gemma":[0.9635189,0.03111964,0.001377106,0.001145116,0.002480684,0.0003587278],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"simulation_or_modeling","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.003853165,0.001967376,0.1284881,0.0006218473,0.002330097,0.0003172332,0.0005043339,0.6779512,0.001530702,0.0007908767,0.005175663,0.1764695],"study_design_scores_gemma":[0.0000563961,0.000475289,0.007294833,0.00003242366,0.0001772257,0.00007534709,0.00008416227,0.9902508,0.0008847738,0.0003882429,0.0002550301,0.00002551779],"study_design_candidate":"simulation_or_modeling","study_design_consensus":"simulation_or_modeling","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9526701,0.002388963,0.03767786,0.0009385641,0.0001388368,0.0003810683,0.002632032,0.001874684,0.00129785],"genre_scores_gemma":[0.9505468,0.0007867352,0.04059594,0.0002179488,0.00009002152,0.0003720288,0.006545266,0.00008582863,0.0007594383],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.0293464,"threshold_uncertainty_score":0.07887226,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0396032572425928,"score_gpt":0.4156220820246641,"score_spread":0.3760188247820713,"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."}}