{"id":"W4308187602","doi":"10.1136/jcp-2022-208506","title":"Use of the ISUP e-learning module improves interrater reliability in prostate cancer grading","year":2022,"lang":"en","type":"article","venue":"Journal of Clinical Pathology","topic":"Prostate Cancer Diagnosis and Treatment","field":"Medicine","cited_by":6,"is_retracted":false,"has_abstract":true,"ca_institutions":"University Health Network","funders":"","keywords":"Medicine; Grading (engineering); Prostate cancer; Inter-rater reliability; Kappa; Wilcoxon signed-rank test; Cohen's kappa; Medical physics; Radiology; Nuclear medicine; Pathology; Cancer; Machine learning; Internal medicine; Computer science; Statistics; Mann–Whitney U test; Rating scale; 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":[],"consensus_categories":[],"category_scores_codex":[0.001702425,0.00008763286,0.0005956987,0.00007049444,0.00005121196,0.000005550105,0.0001110733,0.00004784996,0.00009453702],"category_scores_gemma":[0.001201414,0.00005163666,0.0002916323,0.0001439001,0.0001526,0.00008081322,0.0001607669,0.001033546,5.012906e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00015668,"about_ca_system_score_gemma":0.0002378237,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.00005236921,"about_ca_topic_score_gemma":0.00001891926,"domain_scores_codex":[0.9975827,0.000643838,0.00123451,0.0001689453,0.0002059021,0.0001640905],"domain_scores_gemma":[0.9983115,0.0004602704,0.0008248063,0.0001903572,0.0001421416,0.00007091864],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.0009787452,0.0005929421,0.9806924,0.0000325499,0.00005351942,0.0002703542,0.0005043142,0.001016729,0.00125288,0.00001312198,0.0004962506,0.01409617],"study_design_scores_gemma":[0.002343082,0.002719977,0.9862505,0.0001343818,0.0001007625,0.0001934233,0.0002038142,0.0002908274,0.0009738085,0.0004156475,0.006319123,0.00005468484],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9924108,0.0004698214,0.000004451511,0.005748211,0.001085297,0.0002369295,0.0000107134,0.000003203496,0.00003056269],"genre_scores_gemma":[0.997837,0.0007507883,0.000252521,0.0007929382,0.00009706648,0.00003118975,0.000001146166,0.00001127982,0.0002260508],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.01404149,"threshold_uncertainty_score":0.44903,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.08222556493770797,"score_gpt":0.3984873411640993,"score_spread":0.3162617762263913,"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."}}