{"id":"W2077240755","doi":"10.1016/j.annemergmed.2011.06.423","title":"391 Between-Rater Reliability of Historical Variables Used in a Routine Emergency Department Patient Evaluation","year":2011,"lang":"en","type":"article","venue":"Annals of Emergency Medicine","topic":"Reliability and Agreement in Measurement","field":"Decision Sciences","cited_by":0,"is_retracted":false,"has_abstract":false,"ca_institutions":"University of British Columbia","funders":"","keywords":"Medicine; Emergency department; Inter-rater reliability; Reliability (semiconductor); Medical emergency; Emergency medicine; Statistics; Nursing","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":[0.05398653,0.0004337313,0.0009685659,0.001456768,0.0008363705,0.001281948,0.0006725236,0.0007382304,0.0007061189],"category_scores_gemma":[0.1379827,0.0004597711,0.001257947,0.001037519,0.000835041,0.00125201,0.001104046,0.0007713436,0.0006078566],"about_ca_system_candidate":false,"about_ca_system_consensus":false,"about_ca_system_score_codex":0.0007049069,"about_ca_system_score_gemma":0.000630235,"about_ca_topic_candidate":false,"about_ca_topic_consensus":false,"about_ca_topic_score_codex":0.001245115,"about_ca_topic_score_gemma":0.001806905,"domain_scores_codex":[0.939168,0.03827823,0.008201624,0.004051432,0.00947338,0.0008272456],"domain_scores_gemma":[0.7881362,0.1545888,0.01175422,0.01857619,0.02621948,0.0007251287],"domain_codex":null,"domain_gemma":null,"domain_candidate":null,"domain_consensus":null,"study_design_codex":"observational","study_design_gemma":"observational","study_design_scores_codex":[0.001974791,0.000196711,0.9506557,0.0002318125,0.001547773,0.0001337628,0.004465983,0.001594133,0.003982752,0.0005410799,0.000604514,0.03407098],"study_design_scores_gemma":[0.00009846632,0.001615957,0.972124,0.0001331421,0.0008014388,0.0005864922,0.001199007,0.01176164,0.008524269,0.0006114828,0.002460906,0.00008314286],"study_design_candidate":"observational","study_design_consensus":"observational","genre_codex":"empirical","genre_gemma":"empirical","genre_scores_codex":[0.9859244,0.0006713998,0.009982307,0.00005980469,0.0001603356,0.0001606693,0.0003424282,0.00004911168,0.002649497],"genre_scores_gemma":[0.996467,0.00008291601,0.002824179,0.0000225745,0.00003549036,0.00008426295,0.0002590079,0.0000220626,0.0002025018],"genre_candidate":"empirical","genre_consensus":"empirical","teacher_disagreement_score":0.05398653,"threshold_uncertainty_score":0.2855115,"prediction_status":"machine_predicted_unvalidated"},"machine_scores":{"provisional":true,"baseline":true,"maturity_gate_passed":false,"score_opus":0.5818926185803324,"score_gpt":0.4632303106880315,"score_spread":0.1186623078923009,"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."}}