{"id":"W4408976398","doi":"10.1093/jhps/hnaf011.084","title":"OP1.5 Evaluating ChatGPT’s Performance in Answering Patients’ Questions Relating to Femoroacetabular Impingement Syndrome and Arthroscopic Hip Surgery","year":2025,"lang":"en","type":"article","venue":"Journal of Hip Preservation Surgery","topic":"Hip disorders and treatments","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":true,"ca_institutions":"McMaster University","funders":"","keywords":"Femoroacetabular impingement; Medicine; Hip arthroscopy; Arthroscopy; Surgery; Physical therapy","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":[0.01462715,0.0007881476,0.0007976032,0.001429099,0.0004731773,0.001222192,0.0008448988,0.0008049395,0.005458144],"category_scores_gemma":[0.08447741,0.0003572139,0.001333371,0.0007062037,0.0007471862,0.001057439,0.002648157,0.0009254511,0.0017711],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007022042,"about_ca_system_score_gemma":0.0009965526,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001153887,"about_ca_topic_score_gemma":0.001634128,"domain_scores_codex":[0.989862,0.005894438,0.001551355,0.0008527272,0.001381984,0.0004574618],"domain_scores_gemma":[0.8963553,0.0724044,0.01284956,0.004010641,0.0111748,0.003205346],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"not_applicable","study_design_scores_codex":[0.01030316,0.002533399,0.6473762,0.003439887,0.000518506,0.0007312546,0.02997624,0.002940844,0.008513535,0.0005483664,0.01135264,0.2817659],"study_design_scores_gemma":[0.0006446702,0.01082217,0.9335495,0.0006131805,0.0005327934,0.001591518,0.01414343,0.01651154,0.00824784,0.0008450771,0.01215404,0.0003441725],"study_design_candidate":"not_applicable","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9893461,0.0001530109,0.003573461,0.0003314334,0.0001236094,0.001130103,0.001325028,0.0004430217,0.003574324],"genre_scores_gemma":[0.978811,0.0002409725,0.0143937,0.0003529263,0.0001011378,0.002409677,0.001282739,0.00009520259,0.002312646],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01462715,"threshold_uncertainty_score":0.07735664,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.04802582627426989,"score_gpt":0.3218167469445914,"score_spread":0.2737909206703215,"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."}}