{"id":"W1978972307","doi":"10.1007/s00464-010-0882-8","title":"Global Assessment of Gastrointestinal Endoscopic Skills (GAGES): a valid measurement tool for technical skills in flexible endoscopy","year":2010,"lang":"en","type":"article","venue":"Surgical Endoscopy","topic":"Surgical Simulation and Training","field":"Medicine","cited_by":172,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University; McGill University Health Centre","funders":"","keywords":"Medicine; Intraclass correlation; Cronbach's alpha; Concurrent validity; Colonoscopy; Construct validity; Reliability (semiconductor); Inter-rater reliability; Rating scale; Surgery; Physical therapy; Psychometrics; Patient satisfaction; Statistics; Internal consistency; Internal medicine; Mathematics","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":["metaepi_narrow","insufficient_payload"],"consensus_categories":[],"category_scores_codex":[0.001622707,0.0004168153,0.001005877,0.0002000152,0.00009527876,0.0000390092,0.0002525999,0.0002233203,0.001144707],"category_scores_gemma":[0.001316564,0.0003546836,0.0004268595,0.00066367,0.0002648638,0.0001157952,0.00009660641,0.0006905522,0.00001595776],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0003138105,"about_ca_system_score_gemma":0.0005630661,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00004804828,"about_ca_topic_score_gemma":0.00004730035,"domain_scores_codex":[0.9959492,0.0001393661,0.001151516,0.0007000867,0.001218254,0.0008415697],"domain_scores_gemma":[0.9974639,0.0009568724,0.0002488093,0.0005387218,0.0003747578,0.0004169453],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002138777,0.004935275,0.9350788,0.0003650937,0.0001155443,0.0004555965,0.0000448801,0.0003059964,0.02616851,0.02298171,0.0001625578,0.007247276],"study_design_scores_gemma":[0.167098,0.004017306,0.7607499,0.002210002,0.0003340094,0.000817809,0.00008348629,0.002975972,0.02716072,0.002936569,0.03059498,0.001021269],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9686366,0.00006185459,0.004572619,0.0004015915,0.0005189241,0.00188934,0.00006633266,0.0002476133,0.02360515],"genre_scores_gemma":[0.9448572,0.00001285373,0.0542795,0.000158822,0.0002263365,0.0002296203,0.00005284595,0.00003893744,0.0001439039],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.1743289,"threshold_uncertainty_score":0.9998905,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02867236155473876,"score_gpt":0.3658419886035423,"score_spread":0.3371696270488035,"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."}}