{"id":"W4378190381","doi":"10.1016/j.jogc.2023.03.048","title":"Validity evidence for procedure-specific competence assessment tools in obstetrics and gynecology","year":2023,"lang":"en","type":"article","venue":"Journal of Obstetrics and Gynaecology Canada","topic":"Innovations in Medical Education","field":"Medicine","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of Alberta","funders":"","keywords":"Medicine; Competence (human resources); Obstetrics and gynaecology; Documentation; Medical education; Obstetrics; Gynecology; Pregnancy; Management","routes":{"ca_aff":true,"ca_fund":false,"ca_venue":true,"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":[0.3251934,0.002148842,0.004950886,0.009771784,0.004138564,0.01319461,0.007769279,0.009603713,0.00670206],"category_scores_gemma":[0.7682123,0.003101106,0.01226025,0.008183547,0.01624135,0.007279533,0.009069322,0.005929076,0.001223456],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.008011547,"about_ca_system_score_gemma":0.01441343,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.01521346,"about_ca_topic_score_gemma":0.01963316,"domain_scores_codex":[0.6264693,0.1989146,0.05606902,0.02772652,0.08597949,0.004841126],"domain_scores_gemma":[0.05807268,0.8459139,0.02901772,0.02237895,0.04312732,0.001489388],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"bench_or_experimental","study_design_scores_codex":[0.00951081,0.001019633,0.6715288,0.04561592,0.05123542,0.0003431588,0.01068481,0.003802637,0.001183688,0.02423704,0.009138644,0.1716995],"study_design_scores_gemma":[0.005555823,0.004451451,0.7207719,0.0811054,0.0450117,0.001413041,0.006562314,0.01623561,0.004627718,0.05619524,0.05704259,0.001027182],"study_design_candidate":"bench_or_experimental","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.4709314,0.3290393,0.05813539,0.03634378,0.01301876,0.00672464,0.009424585,0.0002553809,0.07612678],"genre_scores_gemma":[0.9411903,0.01775836,0.02616569,0.005460519,0.002397485,0.001836301,0.003532194,0.0001908737,0.001468379],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3251934,"threshold_uncertainty_score":0.8321567,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.09299586318212753,"score_gpt":0.356464638740618,"score_spread":0.2634687755584905,"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."}}