{"id":"W7117235203","doi":"10.24911/sjemed.72-1740484871","title":"Ai Chatbots In Emergency Medicine: Analyzing Agreement with Expert Physician Triage Decisions","year":2025,"lang":"en","type":"article","venue":"Saudi Journal of Emergency Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Triage; Chatbot; Cohen's kappa; Early warning score; Emergency department; Resource (disambiguation); Reliability (semiconductor); Telemedicine","routes":{"ca_aff":false,"ca_fund":false,"ca_venue":false,"about_ca":true,"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":[0.04545684,0.0005683529,0.0007550921,0.003941777,0.001029926,0.001995523,0.001470171,0.00158275,0.002026358],"category_scores_gemma":[0.2243823,0.0003686698,0.0008581509,0.002030218,0.001568399,0.002339393,0.002686716,0.001339226,0.0007024354],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001313001,"about_ca_system_score_gemma":0.001181306,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003324407,"about_ca_topic_score_gemma":0.003298057,"domain_scores_codex":[0.956689,0.02758028,0.004219807,0.00523018,0.005285449,0.0009952306],"domain_scores_gemma":[0.6642019,0.2807853,0.02634794,0.006288722,0.01896229,0.00341392],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.003158523,0.0006806122,0.8773326,0.001882007,0.0007167975,0.0003547923,0.02328064,0.008446286,0.002104768,0.001210078,0.007492031,0.07334077],"study_design_scores_gemma":[0.0002795522,0.001480493,0.8248895,0.0009727616,0.0004523035,0.0007859811,0.01818202,0.134469,0.0047342,0.00485405,0.008603988,0.00029608],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9831229,0.0008178525,0.01126983,0.0003287217,0.0001399422,0.0003612528,0.0009129348,0.0001911855,0.002855375],"genre_scores_gemma":[0.9910313,0.0001036014,0.007074706,0.0001325521,0.00005433609,0.0003793558,0.0009051793,0.00003424644,0.000284688],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.04545684,"threshold_uncertainty_score":0.2404016,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1752623417108514,"score_gpt":0.4866427908921004,"score_spread":0.311380449181249,"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."}}