{"id":"W7116777300","doi":"10.2196/83318","title":"Artificial Intelligence Models for Predicting Triage in Emergency Departments: Seven-Month Retrospective Comparative Study of Natural Language Processing, Large Language Model, and Joint Embedding Predictive Architectures","year":2025,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Data Visualization and Analytics","field":"Computer Science","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Triage; Retrospective cohort study; Emergency department; Emergency response; MEDLINE","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":[0.004932033,0.0005273002,0.0004964449,0.001268084,0.000263335,0.0008098473,0.0004678293,0.0005879743,0.0005836132],"category_scores_gemma":[0.02229253,0.000198761,0.0006459443,0.0008052978,0.000347981,0.0008767183,0.0006607783,0.0008736082,0.0003553914],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0005823378,"about_ca_system_score_gemma":0.0004596059,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004443971,"about_ca_topic_score_gemma":0.003518306,"domain_scores_codex":[0.9979364,0.001022243,0.0002318882,0.0003348281,0.000358443,0.0001163051],"domain_scores_gemma":[0.9870834,0.006910692,0.002280721,0.0009566231,0.002184639,0.0005839784],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0005041843,0.0002209522,0.9867263,0.00003395849,0.0001635484,0.0001998522,0.0003191786,0.002144012,0.0001874539,0.00003967216,0.000566977,0.0088941],"study_design_scores_gemma":[0.00004427594,0.002063512,0.9573544,0.00005322374,0.0001913911,0.0008457146,0.001446174,0.03638718,0.0004885787,0.0002137977,0.0008705514,0.00004126602],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9978724,0.0002516793,0.001024507,0.00009201723,0.00001340117,0.0000255318,0.0005430925,0.00001712797,0.0001602474],"genre_scores_gemma":[0.9979455,0.0001361547,0.0004326569,0.00002087797,0.00001227787,0.00002638025,0.001330053,0.000006763853,0.0000892984],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004932033,"threshold_uncertainty_score":0.02608341,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.0312730149266279,"score_gpt":0.3842273950826731,"score_spread":0.3529543801560452,"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."}}