{"id":"W4386173035","doi":"10.2196/48933","title":"Applications of the Natural Language Processing Tool ChatGPT in Clinical Practice: Comparative Study and Augmented Systematic Review","year":2023,"lang":"en","type":"article","venue":"JMIR Medical Informatics","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":81,"is_retracted":false,"has_abstract":true,"ca_institutions":"","funders":"","keywords":"Systematic review; Concordance; Artificial intelligence; Computer science; Health care; Reliability (semiconductor); Natural language processing; Data science; MEDLINE; Medicine; Machine learning","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":[0.002209149,0.00008309412,0.0004719331,0.00007632112,0.00006420097,0.00001194888,0.0001285368,0.00007660057,0.00001318632],"category_scores_gemma":[0.002662283,0.00004886125,0.00004482068,0.0008130654,0.0001245121,0.0001536669,0.00006720916,0.0004419266,0.00003426778],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00003836407,"about_ca_system_score_gemma":0.0003445254,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00002842262,"about_ca_topic_score_gemma":0.00002801338,"domain_scores_codex":[0.9973817,0.0001715537,0.001621903,0.00007700106,0.0006091439,0.0001387044],"domain_scores_gemma":[0.9983281,0.0006531738,0.0004509735,0.0002676123,0.0002072425,0.00009284561],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"qualitative","study_design_scores_codex":[0.0001004961,0.002117894,0.01398162,0.5366358,0.0001618979,0.00001816943,0.2610417,0.000001109465,0.000004100335,0.0003122879,0.003751121,0.1818738],"study_design_scores_gemma":[0.0006710812,0.0006020321,0.01924537,0.1612708,0.0006939811,0.0001544881,0.7518933,0.06401457,0.00006685574,0.0000856457,0.001003051,0.0002988256],"study_design_candidate":"qualitative","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9733259,0.008279993,0.0001542113,0.006148966,0.0002293704,0.01096912,0.000003315955,0.0000798301,0.0008093005],"genre_scores_gemma":[0.9961603,0.0006621594,0.0001527734,0.002229968,0.00006089471,0.0006364843,0.00001517614,0.000004696214,0.00007753354],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.4908516,"threshold_uncertainty_score":0.3187194,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.1249508598316767,"score_gpt":0.5494528445603836,"score_spread":0.4245019847287069,"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."}}