{"id":"W3203942361","doi":"10.1016/j.ejrad.2021.109992","title":"Diagnostic performance of US for suspected appendicitis: Does multi-categorical reporting provide better estimates of disease in adults, and what factors are associated with false or indeterminate results?","year":2021,"lang":"en","type":"article","venue":"European Journal of Radiology","topic":"Appendicitis Diagnosis and Management","field":"Medicine","cited_by":2,"is_retracted":false,"has_abstract":false,"ca_institutions":"Queen Elizabeth II Health Sciences Centre; Dalhousie University","funders":"","keywords":"Medicine; Indeterminate; Appendicitis; False positive paradox; False Negative Reactions; Logistic regression; Appendix; Radiology; Surgery; Internal medicine; Statistics","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":["metaresearch"],"consensus_categories":[],"category_scores_codex":[0.0008553509,0.0001764191,0.0007876612,0.0001913412,0.00003625106,0.00002027569,0.00009570665,0.0000377454,0.00001146747],"category_scores_gemma":[0.009717543,0.000104287,0.00008727321,0.0001563877,0.0001683144,0.0002148513,0.00006224806,0.0002253689,2.330661e-7],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.00005431238,"about_ca_system_score_gemma":0.00008906757,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.000009636618,"about_ca_topic_score_gemma":0.00004633682,"domain_scores_codex":[0.9975538,0.0002623221,0.001513134,0.0002489305,0.0001728621,0.0002489011],"domain_scores_gemma":[0.9956626,0.0009641532,0.002757188,0.000160869,0.0003030169,0.0001522024],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.002763414,0.0003870735,0.9810637,0.0005433313,0.0002042484,0.008654515,0.0006296159,0.00009873068,0.000250542,0.000004796715,0.00009128367,0.005308749],"study_design_scores_gemma":[0.006471576,0.002304716,0.9824421,0.004604859,0.0002583308,0.0004104002,0.001402674,0.0005063911,0.001419475,0.000009189288,0.00004816835,0.0001220811],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9977512,0.001019663,0.0002320068,0.0004645593,0.0001174338,0.0003757127,0.00002312982,0.000008601447,0.000007649699],"genre_scores_gemma":[0.9970071,0.001629771,0.00110545,0.0001012338,0.00005520962,0.000005822579,0.00004999038,0.00002976098,0.00001571488],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.008862192,"threshold_uncertainty_score":0.998624,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.02369486844994675,"score_gpt":0.2694970753285191,"score_spread":0.2458022068785724,"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."}}