{"id":"W4386710076","doi":"10.2196/48808","title":"ChatGPT-Generated Differential Diagnosis Lists for Complex Case–Derived Clinical Vignettes: Diagnostic Accuracy Evaluation","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":135,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Differential diagnosis; Medical diagnosis; Medicine; Radiology; Pathology","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":["metaresearch","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001845013,0.0002294851,0.0005037449,0.0002130472,0.0002909749,0.00006908324,0.0001762852,0.0004556694,0.00172487],"category_scores_gemma":[0.01997758,0.0001918545,0.0002022645,0.0005722288,0.0002110259,0.0002263848,0.00007619557,0.0004478114,0.0005153716],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0001374363,"about_ca_system_score_gemma":0.0007982797,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00009446785,"about_ca_topic_score_gemma":0.00008440697,"domain_scores_codex":[0.9960736,0.0001710312,0.001902293,0.0002181055,0.001094678,0.0005403082],"domain_scores_gemma":[0.9913231,0.006442694,0.0003626632,0.0003990867,0.000771917,0.0007005564],"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":[0.0001706231,0.0006851147,0.01835295,0.0009015232,0.0001390391,0.0001323156,0.011914,0.00003078638,0.00006398957,0.0001083434,0.1535621,0.8139393],"study_design_scores_gemma":[0.001525008,0.001363442,0.03238315,0.0008123448,0.000442822,0.0004341496,0.009602868,0.9228379,0.001546825,0.0006865863,0.02783247,0.00053244],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9857705,0.00005101939,0.002804343,0.006268243,0.001714763,0.002967911,0.00004421506,0.0002587168,0.0001202594],"genre_scores_gemma":[0.9884288,0.0006646374,0.000750311,0.003768423,0.001957378,0.002304959,0.002043298,0.00003147651,0.00005071714],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9228071,"threshold_uncertainty_score":0.9991877,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3661866926666326,"score_gpt":0.5378131142135208,"score_spread":0.1716264215468882,"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."}}