{"id":"W4411792792","doi":"10.2196/68955","title":"Efficient Detection of Stigmatizing Language in Electronic Health Records via In-Context Learning: Comparative Analysis and Validation Study","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Machine Learning in Healthcare","field":"Computer Science","cited_by":2,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto","funders":"","keywords":"Preprint; Context (archaeology); Health records; Electronic health record; Computer science; Data science; Natural language processing; Psychology; World Wide Web; Health care","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.001946482,0.0001048481,0.0003987757,0.0008809334,0.00006978749,0.00003271731,0.0002567632,0.00007216534,0.000006045814],"category_scores_gemma":[0.0002048609,0.00009666313,0.00003238998,0.002298269,0.00003329049,0.0001183222,0.0001484849,0.0006978067,0.00000181678],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002378193,"about_ca_system_score_gemma":0.0002485883,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001362636,"about_ca_topic_score_gemma":0.004641511,"domain_scores_codex":[0.997816,0.0004573207,0.0008784487,0.000138632,0.0004476234,0.0002619917],"domain_scores_gemma":[0.9990081,0.0003130822,0.0003307969,0.0002105019,0.00005466417,0.00008287256],"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":[0.00003213133,0.0003733469,0.2654474,0.0005951146,0.0001050126,0.000003750931,0.3095191,0.02183882,0.00001596247,0.00120329,0.000005707909,0.4008604],"study_design_scores_gemma":[0.0005255506,0.000279493,0.1077689,0.000103246,0.000007621706,0.000001435395,0.01317461,0.8779627,0.0000616208,0.00002811359,0.0000224767,0.00006424078],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8030879,0.0001116716,0.1959532,0.0001960383,0.00004334067,0.000432685,2.172084e-7,0.00003566537,0.0001393003],"genre_scores_gemma":[0.998906,0.00001160195,0.0008857418,0.000138767,0.000005211106,0.0000374065,0.00000400862,0.000001915506,0.000009385037],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.8561239,"threshold_uncertainty_score":0.3941808,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.01302340310798402,"score_gpt":0.3547310234171829,"score_spread":0.3417076203091989,"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."}}