LAUSR

OBJECTIVE Rather than relying on a single psychotherapeutic orientation, most clinicians draw from a range of therapeutic approaches to treat their clients. To date, no data-driven approach exists for personalized predictions of which skill domain would be most therapeutically beneficial for a given patient. The present study combined ecological momentary assessment (EMA) and machine learning to test a data-driven approach for predicting patient-specific skill-outcome associations. METHOD Fifty (Mage = 37 years old, 54% female, 84% White) adults received training in behavioral therapy (BT) and dialectical behavior therapy (DBT) skills within a behavioral health partial hospital program (PHP). Following discharge, patients received four EMA surveys per day for 2 weeks (total observations = 2,036) assessing the use of therapeutic skills and positive/negative affect (PA/NA). Clinical and demographic characteristics were submitted to elastic net regularization to predict, via cross-validation, patient-specific associations between the use of BT versus DBT skills and level of PA/NA. RESULTS Cross-validated accuracy was 81% (sensitivity = 93% and specificity = 63%) in predicting whether a patient would exhibit a stronger association between the use of BT versus DBT skills and PA level. Predictors of positive DBT skills-PA associations included higher levels of nonsuicidal self-injury (NSSI) and sleep disturbance, whereas predictors of positive BT skills-PA relations included higher emotional lability and anxiety disorder comorbidity, and lower psychomotor retardation/agitation and worthlessness/guilt. Corresponding models with NA yielded no predictors. CONCLUSIONS Findings from this initial proof-of-concept study highlight the potential of data-driven approaches to inform personalized prescriptions of which skill domains may be most therapeutically beneficial for a given patient. (PsycInfo Database Record (c) 2021 APA, all rights reserved).