{"id":"W4417524748","doi":"10.5435/jaaosglobal-d-25-00341","title":"Evaluation of ChatGPT-4o as a Patient Information Tool for Common Orthopaedic Surgeries: Accuracy, Completeness, and Clinical Utility","year":2025,"lang":"en","type":"article","venue":"JAAOS Global Research and Reviews","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"Object Research Systems (Canada)","funders":"","keywords":"Information system; Orthopedic surgery; MEDLINE; Patient care","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":false,"about_ca":false,"invisible_to_affiliation_only":false},"retraction":null,"screen":null,"direct_labels":[],"prediction":{"model_version":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.03989015,0.0006021369,0.0008350274,0.00367901,0.0005554932,0.001473626,0.0009842353,0.0007177944,0.001946656],"category_scores_gemma":[0.1623169,0.0003052439,0.0008055634,0.001836696,0.0009644139,0.001480886,0.003521648,0.0005592952,0.0006430596],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.001253658,"about_ca_system_score_gemma":0.001865847,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001700026,"about_ca_topic_score_gemma":0.002773148,"domain_scores_codex":[0.9703387,0.01687812,0.004926472,0.001177726,0.00600814,0.0006707076],"domain_scores_gemma":[0.7564499,0.1702187,0.02791821,0.008641111,0.03301292,0.0037592],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.004909989,0.0009028011,0.7325495,0.001837787,0.0001890237,0.0004123863,0.01653325,0.001595185,0.00359765,0.0003786578,0.004483999,0.2326099],"study_design_scores_gemma":[0.0005041704,0.005603256,0.9326438,0.0009953257,0.0003629748,0.001528137,0.01039855,0.02481189,0.009004638,0.0008494349,0.01302918,0.0002686983],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9841481,0.0003453531,0.006891305,0.0004650566,0.00008255173,0.002207126,0.001922149,0.0004746965,0.003463632],"genre_scores_gemma":[0.9720789,0.0003263242,0.02129885,0.000251219,0.00008327366,0.002919397,0.001786098,0.00009122572,0.001164717],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.03989015,"threshold_uncertainty_score":0.2109618,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.528192822504121,"score_gpt":0.6212552639638854,"score_spread":0.09306244145976439,"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."}}