{"id":"W2465767413","doi":"10.1093/ecco-jcc/jjw120","title":"Effect of Standardised Scoring Conventions on Inter-rater Reliability in the Endoscopic Evaluation of Crohn’s Disease","year":2016,"lang":"en","type":"article","venue":"Journal of Crohn s and Colitis","topic":"Inflammatory Bowel Disease","field":"Biochemistry, Genetics and Molecular Biology","cited_by":36,"is_retracted":false,"has_abstract":true,"ca_institutions":"University of Toronto; St. Michael's Hospital; Robarts Clinical Trials; Western University","funders":"","keywords":"Inter-rater reliability; Crohn's disease; Reliability (semiconductor); Disease; Medicine; Physical therapy; Internal medicine; Statistics; Mathematics; Physics; Rating scale","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.3493598,0.001112107,0.001461644,0.001895972,0.001501817,0.002369026,0.001851836,0.001562662,0.0009185917],"category_scores_gemma":[0.4926372,0.001471096,0.002115102,0.001780151,0.004193863,0.002303599,0.003524855,0.00211479,0.000565663],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.002205722,"about_ca_system_score_gemma":0.001861306,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001622981,"about_ca_topic_score_gemma":0.003438854,"domain_scores_codex":[0.4980744,0.3978952,0.04022159,0.01814177,0.04352517,0.002141916],"domain_scores_gemma":[0.2419725,0.5899486,0.05096659,0.05117319,0.06416592,0.001773102],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.01342392,0.001294377,0.5975352,0.00468689,0.00563619,0.0004749888,0.04338519,0.01443461,0.01967846,0.003791209,0.00760213,0.2880569],"study_design_scores_gemma":[0.001059739,0.0102881,0.9137182,0.002289935,0.002041333,0.001002167,0.003171584,0.035964,0.01643078,0.002794556,0.01064703,0.0005925567],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.8847737,0.005230664,0.09024894,0.0008156238,0.00158409,0.004868095,0.0004672589,0.0005981972,0.01141342],"genre_scores_gemma":[0.9516279,0.0003501202,0.04342642,0.0002699488,0.000191378,0.003012852,0.0003457443,0.0001771433,0.0005985036],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.3493598,"threshold_uncertainty_score":0.8023552,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.009860362704388687,"score_gpt":0.289036159677372,"score_spread":0.2791757969729833,"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."}}