Predicting U.S. veteran dropout from trauma-focused eye movement desensitization and reprocessing therapy using machine learning.

= 114; 40.9%) reported worse posttraumatic stress disorder and depression symptoms than veterans who completed EMDR. Optimal performance for predicting dropout varied by algorithm and metric, including the final ensemble (precision-recall area under the receiver operating characteristic curve [AUC] = .83, Brier score = .13), random forest (precision-recall AUC = .83, AUC = .89, accuracy = .85), and logistic regression models (precision-recall AUC = .83, specificity = .94). Variables yielding the highest feature importance values that were linked with higher dropout included younger age, more trauma types, worsening posttraumatic stress disorder symptoms, improving depression symptoms, and missing symptom measures. Together, results underscore the importance of better understanding EMDR dropout among veterans. Findings suggest that dropout risk factors may operate differently across veteran subgroups and that machine learning may be useful for characterizing this heterogeneity. Therapists, researchers, and clinics may consider using research-based dropout prevention strategies for veterans at risk of dropping out. (PsycInfo Database Record (c) 2026 APA, all rights reserved).