{"id":"W4386213493","doi":"10.1055/a-2161-1982","title":"The Bethesda ERCP Skills Assessment Tool (BESAT) can reliably differentiate endoscopists of different experience levels","year":2023,"lang":"en","type":"article","venue":"Endoscopy International Open","topic":"Gallbladder and Bile Duct Disorders","field":"Medicine","cited_by":3,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Calgary; The Wilson Centre; Hospital for Sick Children; University of Toronto","funders":"","keywords":"Medicine; Endoscopic retrograde cholangiopancreatography; Intraclass correlation; Discriminative model; Reliability (semiconductor); Validity; Medical physics; Psychometrics; Surgery; Clinical psychology; 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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.004671009,0.0003694755,0.00031133,0.00138444,0.0001471168,0.0005225993,0.0003371466,0.0004289645,0.002008817],"category_scores_gemma":[0.02535608,0.0001567637,0.0004500027,0.0004965028,0.0002715771,0.0007925066,0.0008892768,0.0003024542,0.0004842274],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0002781023,"about_ca_system_score_gemma":0.000447276,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001041586,"about_ca_topic_score_gemma":0.00216689,"domain_scores_codex":[0.9971946,0.0009245318,0.0005164351,0.0003274262,0.0008020317,0.0002349629],"domain_scores_gemma":[0.9827991,0.00857551,0.003632694,0.0005896951,0.003686405,0.0007165991],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0003453244,0.0001822561,0.9322876,0.000119237,0.00009943199,0.00007566659,0.0005904309,0.0006246472,0.001710718,0.00008259475,0.0008351114,0.06304698],"study_design_scores_gemma":[0.00003911285,0.000591732,0.9925983,0.00008067564,0.00002883156,0.0002959287,0.0004350388,0.003915688,0.001146882,0.0001491922,0.0006974275,0.00002114097],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.99418,0.0001612739,0.003292042,0.00006541502,0.00002030719,0.00007806026,0.0001925804,0.00004199838,0.001968446],"genre_scores_gemma":[0.9948751,0.00009363051,0.004228739,0.00004233274,0.00001345815,0.0001114422,0.0003212169,0.000008028879,0.0003061153],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.004671009,"threshold_uncertainty_score":0.02470297,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02661151850372033,"score_gpt":0.3664988632271707,"score_spread":0.3398873447234504,"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."}}