{"id":"W4317951735","doi":"10.1001/jamasurg.2022.5831","title":"The Benefits of Benchmarking—A New Perspective on Measuring Quality in Surgery","year":2023,"lang":"en","type":"article","venue":"JAMA Surgery","topic":"Clinical practice guidelines implementation","field":"Medicine","cited_by":20,"is_retracted":false,"has_abstract":true,"ca_institutions":"McGill University; Royal Victoria Hospital","funders":"","keywords":"Medicine; Benchmarking; Perspective (graphical); Quality (philosophy); MEDLINE; Quality management; General surgery; Medical physics; Intensive care medicine; Surgery; Operations management; Marketing; Artificial intelligence","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":"codex-gemma-dda1882f352a","candidate_categories":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.009492183,0.0001318226,0.0004933627,0.0003386441,0.00007618463,0.00002775652,0.00006968701,0.00009457426,0.00007424224],"category_scores_gemma":[0.05457001,0.0001024491,0.0002745989,0.0009843275,0.00005277983,0.0001653776,0.00005822178,0.0003348463,0.00005602965],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002217239,"about_ca_system_score_gemma":0.0005571865,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003428858,"about_ca_topic_score_gemma":0.0007068217,"domain_scores_codex":[0.9969631,0.00035634,0.001384794,0.0002999847,0.00066641,0.0003293054],"domain_scores_gemma":[0.9542154,0.04447516,0.0004745273,0.0004165179,0.0003007981,0.0001176091],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001374806,0.000165639,0.6426071,0.00007549048,0.00009761943,0.00004100571,0.0004952299,0.0001733871,0.0002671972,0.001669554,0.08661149,0.2664215],"study_design_scores_gemma":[0.000483202,0.00003341231,0.9925069,0.0003091788,0.00002722618,0.000005207582,0.001894584,0.0001043263,0.0006584744,0.0007303162,0.003135082,0.0001121583],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9686321,0.0002207393,0.00001212,0.02908548,0.0007449368,0.0002389263,0.000008845414,0.00006468177,0.0009921398],"genre_scores_gemma":[0.9974341,0.0006873993,0.00005117522,0.0008753508,0.0006340375,0.00001479239,0.00003897894,0.00002375487,0.0002403508],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3498997,"threshold_uncertainty_score":0.9533938,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.4151562763802944,"score_gpt":0.4792549402302482,"score_spread":0.06409866384995383,"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."}}