Stroop task data does not improve prediction of CBT partial hospital treatment outcomes beyond routine clinical data: An elastic net analysis

BACKGROUND: Partial hospitalization is a psychiatric treatment for patients who require structured, intensive outpatient intervention, many of whom present with features of borderline personality disorder (BPD). Findings to date suggest that baseline clinical severity and demographic characteristics are predictive of partial hospital outcomes, but it is possible that prediction could be further improved by including cognitive data into predictive models. METHODS: Naturalistic data from 113 patients admitted to a brief CBT partial hospital were analyzed. Elastic net regression was employed to compare prediction of four outcomes using i) baseline severity and demographics only and ii) models that included cognitive data derived from Classic and Emotional Stroop Tasks. Interactions between BPD trait severity and Stroop performance were also included as predictors in the Stroop Task models. Predicted outcomes included depression, interpersonal functioning, emotional lability, and self-harm ideation. RESULTS: Including Stroop Task data did not improve the prediction outcomes. RMSE values for models with and without data from the Stroop Tasks were virtually equivalent or differed in magnitude by only 0.1 on the five-point outcome scale. Baseline symptom severity was the most robust predictor of all outcomes. CONCLUSIONS: The addition of Stroop Task data did not improve prediction of partial hospital outcomes in a naturalistic sample. Findings underscore the importance of examining clinical data routinely collected at admission when considering who may benefit from a partial hospital treatment.