{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.2363235,0.002007384,0.007449758,0.04490325,0.001947706,0.006392414,0.003318325,0.002261814,0.005410291],"category_scores_gemma":[0.5802275,0.001813147,0.01069767,0.03122186,0.002809568,0.008585932,0.007699289,0.001477241,0.0005759699],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.009244063,"about_ca_system_score_gemma":0.02684639,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004151316,"about_ca_topic_score_gemma":0.01583462,"domain_scores_codex":[0.6121169,0.2468574,0.1042191,0.009367544,0.02616169,0.001277304],"domain_scores_gemma":[0.2263396,0.6673958,0.05087077,0.02048971,0.03366965,0.001234342],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"systematic_review","study_design_gemma":"systematic_review","study_design_scores_codex":[0.001175776,0.0001271629,0.008067762,0.8145456,0.02009308,0.0004502607,0.008462287,0.0005353764,0.0008342924,0.001252769,0.002804714,0.1416509],"study_design_scores_gemma":[0.002247697,0.002314874,0.02654027,0.8103412,0.1023513,0.001711336,0.01129073,0.002359943,0.001836557,0.004398705,0.03409389,0.000513625],"study_design_candidate":"systematic_review","study_design_consensus":"systematic_review","genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.07880843,0.7980345,0.04524779,0.005833401,0.001430634,0.05362825,0.01047917,0.0007636689,0.005774203],"genre_scores_gemma":[0.4359228,0.2310986,0.1955296,0.003791757,0.0006108594,0.1267904,0.005174425,0.0002469854,0.0008345518],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.2363235,"threshold_uncertainty_score":0.9417491,"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."}}