{"id":"W4411010319","doi":"10.2196/72706","title":"Assessing the Accuracy of ChatGPT in Appropriately Triaging Common Postoperative Concerns Regarding Mohs Micrographic Surgery","year":2025,"lang":"en","type":"article","venue":"JMIR Dermatology","topic":"Nonmelanoma Skin Cancer Studies","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Preprint; Triage; SCOOP; Medicine; Surgery; Medical emergency; Computer science; World Wide Web; Operating system","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":"codex-gemma-dda1882f352a","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.0003475798,0.0001918601,0.0009376263,0.0004321981,0.000139921,0.00003224134,0.0001538056,0.0001350584,0.00001063263],"category_scores_gemma":[0.0003453964,0.0001391877,0.0001601226,0.0007400779,0.0003678234,0.0001571497,0.0001458503,0.0004005815,0.000002649217],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00008916749,"about_ca_system_score_gemma":0.00025429,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0001462261,"about_ca_topic_score_gemma":0.00009623608,"domain_scores_codex":[0.9982685,0.000332484,0.000634836,0.0002871643,0.0001361461,0.0003408058],"domain_scores_gemma":[0.9971786,0.002045853,0.0002652085,0.0003708247,0.000108553,0.00003102774],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0001613103,0.0001193786,0.9847823,0.0002891538,0.0002167859,0.0002150517,0.001951181,0.000002244243,0.001858847,0.0008096049,0.002060418,0.007533748],"study_design_scores_gemma":[0.003332811,0.00004881853,0.9478625,0.001481402,0.0001370561,0.0007293093,0.00605013,0.0002310553,0.02930702,0.0004962023,0.01005708,0.0002666093],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9759502,0.004287304,0.00008677307,0.01532478,0.0003743084,0.0007459024,0.000002645217,0.00003547074,0.003192543],"genre_scores_gemma":[0.9980564,0.0002656714,0.0001416263,0.00125154,0.00004136431,0.0001759709,0.000008352204,0.00001625648,0.00004277824],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03691977,"threshold_uncertainty_score":0.5675908,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04084716065902948,"score_gpt":0.385464544703262,"score_spread":0.3446173840442325,"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."}}