Personalized predictions of treatment outcome in patients with post-stroke depressive symptoms

Abstract

Objective: Post-stroke depressive symptoms have a vast individual and societal impact. However, research into interventions for such symptoms show contradictory results; it is unclear what works for which patients. In addition, clinical prediction tools are lacking. This study aimed to develop a prognostic index model for treatment outcome in patients with post-stroke depressive symptoms. Methods: Data from a randomized controlled trial (n?=?61) evaluating 2 interventions for post-stroke depressive symptoms were used to predict post-treatment post-stroke depressive symptoms and participation. From 18 pre-treatment variables of patients and caregivers, predictors were selected using elastic net regression. Based on this selection, prognostic index scores (i.e. predictions) for both out-comes were computed for each individual patient. Results: The depression model included all pre-treatment variables, explaining 44% of the variance. The strongest predictors were: lesion location, employment, participation, comorbidities, mobility, sex, and pre-treatment depression. Six predictors of post-treatment participation were identified, explaining 51% of the variance: mobility, pre-treatment participation, age, satisfaction with participation, caregiver strain, and psychological distress of the spouse. The cross-validated prognostic index scores correlated highly with the actual outcome scores (depression: correlation?=?0.672; participation: correlation?=?0.718). Conclusion: Post-stroke depressive symptoms form a complex and multifactorial problem. Treatment outcome is influenced by the characteristics of the stroke, the patients, and their spouses. The results show that psychological distress is probably no obstacle to attempting to improve participation. The personalized predictions (prognostic index scores) of treatment outcome show promising results, which, after further replication and validation, could aid clinicians with treatment selection.