Original ArticlesPredictors of Disagreement Between Diagnoses From Consult Requesters and Consultation-Liaison PsychiatryOtani, Victor MD, MSc*,†; Otani, Thaís MD*; Freirias, Andrea MD*; Calfat, Elie MD, MSc*; Aoki, Patricia MD*; Cross, Sean MD‡; Sumskis, Susan BN (Hons), PhD, RN, MHN, FACMHN§; Kanaan, Richard PhD, CCT, MB BS, MA, BA∥; Cordeiro, Quirino MD, PhD*; Uchida, Ricardo MD, PhD*Author Information *Department of Mental Health, Santa Casa Medical School †Department of Psychiatry, Instituto Superior de Medicina, Sao Paulo, Brazil ‡Maudsley Simulation, South London & Maudsley Foundation NHS Trust, Lambeth Hospital, London, United Kingdom §School of Nursing, Faculty of Science, Medicine, and Health University of Wollongong, Wollongong ∥Austin Health, University of Melbourne, Melbourne, Australia. Send reprint requests to Victor Otani, MD, MSc, Department of Mental Health, Santa Casa Medical School, Rua Major Maragliano, 241, Sao Paulo, SP, Brazil 01239–011. E-mail: email@example.com. The Journal of Nervous and Mental Disease: December 2019 - Volume 207 - Issue 12 - p 1019-1024 doi: 10.1097/NMD.0000000000001018 Buy Metrics Abstract We evaluated disagreement between reported symptoms and a final diagnosis of depression, anxiety, withdrawal, psychosis, or delirium through regression models assessing individual and combined diagnoses. Highest disagreement rates were reported for services classified as others (88.2%), general surgery (78.5%), and bone marrow transplant (77.7%). Disagreement rates varied widely across different diagnoses, with anxiety having the highest disagreement rate (63.3%), whereas psychosis had the lowest disagreement rate (10.6%). When evaluating kappa coefficients, the highest agreement occurred with diagnoses of withdrawal and psychosis (0.66% and 0.51%, respectively), whereas anxiety and depression presented the lowest values (0.31% and 0.11%, respectively). The best-performing predictive model for most outcomes was random forest, with the most important predictors being specialties other than the ones focused on single systems, older age, lack of social support, and the requester being a resident. Monitoring disagreement rates and their predictors provides information that could lead to quality improvement and safety programs. Copyright © 2019 Wolters Kluwer Health, Inc. All rights reserved.