{"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":"metacan-v3-hybrid-931329e0061c","candidate_categories":[],"consensus_categories":[],"category_scores_codex":[0.01319589,0.0006621996,0.0007665572,0.001343849,0.0008633026,0.001432556,0.001273052,0.001434134,0.001247563],"category_scores_gemma":[0.03632101,0.0004030224,0.001335353,0.0008945087,0.0009321261,0.001084494,0.000946107,0.001116095,0.0005973291],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007885954,"about_ca_system_score_gemma":0.001333871,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.008014374,"about_ca_topic_score_gemma":0.007608091,"domain_scores_codex":[0.9927326,0.004125692,0.0007381777,0.001365251,0.0008384472,0.0001997937],"domain_scores_gemma":[0.9641311,0.02422913,0.002178868,0.002930733,0.005941079,0.0005889887],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.006288904,0.007823459,0.898291,0.0003602564,0.001776782,0.0004712901,0.002287762,0.007353087,0.01040653,0.0003105387,0.001871498,0.06275886],"study_design_scores_gemma":[0.001285912,0.007139327,0.8985356,0.00006530542,0.0011048,0.001074013,0.00136435,0.08096965,0.005796982,0.0006566104,0.001913437,0.00009398842],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9880654,0.00006152062,0.009773655,0.0001032333,0.00001896711,0.0004891896,0.0009815725,0.000122152,0.0003843921],"genre_scores_gemma":[0.9763618,0.00005045415,0.0190273,0.0001643183,0.00003220552,0.000707417,0.003217105,0.00004955506,0.0003899622],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01319589,"threshold_uncertainty_score":0.06978738,"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."}}