{"id":"W4416891240","doi":"10.1007/s12178-025-10001-y","title":"Evaluating the Efficacy and Efficiency of GPT-5 for Automated Title and Abstract Screening in Orthopedic Surgery Systematic Reviews","year":2025,"lang":"en","type":"article","venue":"Current Reviews in Musculoskeletal Medicine","topic":"Artificial Intelligence in Healthcare and Education","field":"Medicine","cited_by":5,"is_retracted":false,"has_abstract":false,"ca_institutions":"Hamilton Health Sciences; McMaster University","funders":"","keywords":"Orthopedic surgery; Inclusion and exclusion criteria; Predictive value; Sports medicine; Inclusion (mineral); Systematic review","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.1283954,0.003005923,0.01312396,0.009970533,0.0009108411,0.005070158,0.003640272,0.003515816,0.009292678],"category_scores_gemma":[0.3826075,0.002351784,0.03495828,0.006656819,0.001753652,0.007879472,0.004992087,0.002677189,0.001039998],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002533979,"about_ca_system_score_gemma":0.007967451,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.004442163,"about_ca_topic_score_gemma":0.01718424,"domain_scores_codex":[0.8474287,0.08669605,0.04421179,0.01033467,0.01034601,0.0009827508],"domain_scores_gemma":[0.4762613,0.4548012,0.04175655,0.01327186,0.01217913,0.001729953],"domain_codex":null,"domain_gemma":"methods","domain_candidate":"methods","domain_consensus":null,"study_design_codex":"meta_analysis","study_design_gemma":"simulation_or_modeling","study_design_scores_codex":[0.0246594,0.000234969,0.03872535,0.3619024,0.3927266,0.0002487512,0.001132732,0.003305993,0.00126002,0.00143558,0.006978124,0.1673901],"study_design_scores_gemma":[0.02144594,0.004567398,0.05315441,0.05538112,0.8335977,0.0007782237,0.0005204906,0.01123705,0.001310492,0.003818446,0.01369722,0.0004915462],"study_design_candidate":"simulation_or_modeling","study_design_consensus":null,"genre_codex":"review","genre_gemma":"empirical","genre_scores_codex":[0.1556775,0.7361497,0.04279534,0.008444948,0.002982391,0.01700743,0.02833604,0.003051204,0.005555439],"genre_scores_gemma":[0.7508596,0.09428865,0.1271694,0.004475901,0.001289116,0.01075577,0.009056429,0.000574098,0.00153109],"genre_candidate":"empirical","genre_consensus":null,"teacher_disagreement_score":0.8716046,"threshold_uncertainty_score":0.679028,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.3686849703064726,"score_gpt":0.5599606477077739,"score_spread":0.1912756774013013,"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."}}