{"id":"W3174008823","doi":"10.1097/sla.0000000000005015","title":"Gender Bias in the Evaluation of Surgical Performance","year":2021,"lang":"en","type":"article","venue":"Annals of Surgery","topic":"Diversity and Career in Medicine","field":"Social Sciences","cited_by":21,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of British Columbia","funders":"National Cancer Institute; National Institutes of Health","keywords":"Medicine; Checklist; Inter-rater reliability; Cronbach's alpha; Gender bias; Respondent; Clinical psychology; Rating scale; Psychometrics; Psychology; Social psychology; Developmental psychology","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":[0.0632342,0.0005488051,0.0007036917,0.001751983,0.0006311869,0.0012226,0.0006945762,0.0004604305,0.002849185],"category_scores_gemma":[0.1832946,0.0002252586,0.0007683102,0.001086973,0.001709403,0.001336723,0.001970123,0.0004680779,0.0004683348],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0008911578,"about_ca_system_score_gemma":0.001235952,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001196983,"about_ca_topic_score_gemma":0.00154836,"domain_scores_codex":[0.9274396,0.0460798,0.006338207,0.004249583,0.01484612,0.001046591],"domain_scores_gemma":[0.7962517,0.1256605,0.05119245,0.009061319,0.01628271,0.001551327],"domain_codex":null,"domain_gemma":"evaluation","domain_candidate":"evaluation","domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001937225,0.0001429947,0.8826824,0.0008353621,0.000452467,0.0001944148,0.007654511,0.0004704928,0.001820072,0.001150322,0.001458176,0.1012015],"study_design_scores_gemma":[0.00009542957,0.002555824,0.977438,0.001058672,0.000185454,0.00107164,0.004743033,0.001381763,0.002960463,0.001811449,0.006599745,0.00009853276],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9663359,0.004298915,0.0165823,0.0009912722,0.0005615134,0.0006272611,0.0006843615,0.00005989601,0.009858721],"genre_scores_gemma":[0.9917401,0.0006383318,0.005565202,0.0003076244,0.0001600612,0.0004710825,0.0001656477,0.00002945668,0.0009224676],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.9367658,"threshold_uncertainty_score":0.3344184,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.7118481104725602,"score_gpt":0.4509731255451161,"score_spread":0.2608749849274442,"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."}}