{"id":"W4285678491","doi":"10.1016/j.gie.2022.07.009","title":"Validation of a natural language processing algorithm to identify adenomas and measure adenoma detection rates across a health system: a population-level study","year":2022,"lang":"en","type":"article","venue":"Gastrointestinal Endoscopy","topic":"Colorectal Cancer Screening and Detection","field":"Medicine","cited_by":18,"is_retracted":false,"has_abstract":false,"ca_institutions":"McMaster University; Women's College Hospital; Cancer Care Ontario; Sunnybrook Health Science Centre; St. Joseph's Hospital; Institute for Clinical Evaluative Sciences; Brantford Energy (Canada); Health Sciences Centre; University of Toronto","funders":"","keywords":"Medicine; Interquartile range; Artificial intelligence; Adenoma; Population; Confidence interval; Natural language processing; Test set; Machine learning; Algorithm; Radiology; Pathology; Internal medicine; Computer science","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":[],"consensus_categories":[],"category_scores_codex":[0.0007433982,0.0001628534,0.0003283543,0.0001904052,0.0004289637,0.00004880655,0.00006597237,0.00001561943,0.000006930334],"category_scores_gemma":[0.0002171689,0.0001658079,0.00005134249,0.0005605512,0.00002132375,0.0001101236,0.00009835828,0.0002920713,0.000001289615],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0004495447,"about_ca_system_score_gemma":0.0001089191,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.003785163,"about_ca_topic_score_gemma":0.0001967844,"domain_scores_codex":[0.9982389,0.0002191007,0.0003936893,0.000364846,0.0005115146,0.000271987],"domain_scores_gemma":[0.9992269,0.00005525651,0.0002713019,0.0001510282,0.0001603861,0.0001351229],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"design_other","study_design_gemma":"observational","study_design_scores_codex":[0.01555107,0.0007810721,0.2102857,0.001591764,0.0001206949,0.000212009,0.02761322,0.0008986069,0.3248667,0.000002010254,0.0000446158,0.4180326],"study_design_scores_gemma":[0.004570341,0.02099671,0.8531718,0.001177144,0.0001350201,0.01490606,0.07263726,0.02271608,0.009381702,0.000003685895,0.000005707393,0.0002985075],"study_design_candidate":"observational","study_design_consensus":null,"genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9654942,0.0005732663,0.03249522,0.0001246988,0.0001520261,0.00097069,0.00004051604,0.0001412435,0.000008198042],"genre_scores_gemma":[0.9824237,2.14424e-7,0.01713722,0.00003032844,0.000133517,0.0001696946,0.0000304059,0.00002464904,0.00005024069],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.642886,"threshold_uncertainty_score":0.6761449,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02733405693761766,"score_gpt":0.3492438096581496,"score_spread":0.3219097527205319,"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."}}