{"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":"codex-gemma-dda1882f352a","candidate_categories":["insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.0002634629,0.0002352799,0.0003691668,0.00010303,0.0001563905,0.0001694847,0.00103532,0.00006078308,0.001009999],"category_scores_gemma":[0.0002343816,0.0001527389,0.0001462023,0.000220261,0.0001459015,0.0001469638,0.0005475525,0.0002294669,0.00007864282],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.000154011,"about_ca_system_score_gemma":0.0001869955,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.0002850205,"about_ca_topic_score_gemma":0.0001303377,"domain_scores_codex":[0.9978264,0.00008101169,0.0005369702,0.0003827569,0.0008093883,0.000363493],"domain_scores_gemma":[0.9987402,0.0002760319,0.0001920724,0.000493598,0.0001891914,0.0001088812],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0008546048,0.002226464,0.7304317,0.0002081101,0.001968818,0.0001506171,0.009094522,0.00006035652,0.1824848,0.01857754,0.03247158,0.02147086],"study_design_scores_gemma":[0.00351882,0.0003354299,0.9209954,0.0003611693,0.00007176393,0.00001475459,0.001741797,0.0004671323,0.05828794,0.002015545,0.01191007,0.0002801359],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.980122,0.00003664564,0.00008783817,0.005107434,0.001461653,0.001017862,0.0001125908,0.000054124,0.01199984],"genre_scores_gemma":[0.985135,0.0003334202,0.0003949121,0.0003811441,0.0001082942,0.0002443973,0.0001484351,0.00003143261,0.01322303],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1905637,"threshold_uncertainty_score":0.9999032,"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."}}