Optimizing
Wellness
(OWL) Lab
Precision Mental Health
We have developed a keen interest in researching the precise matching of patients to the most suitable MH interventions. For instance, we recently wrote a meta-analysis that showed that psychotherapy-only and combined antidepressant medication (ADM) and psychotherapy were more effective than ADM alone in reducing serious psychiatric adverse events for depressed patients, possibly because psychotherapies (alone or with ADM) offer strategies to strengthen the will to live (Zainal, in press). My collaborators and I have also identified predictive factors for psychopharmacological treatments for common mental disorders (CMDs). During this work, our team has gained expertise in machine learning (ML) algorithms, which utilize parametric and flexible ML methods. This aligns with the core of precision MH, where treatment effects can vary depending on the specific population and subgroups being studied.
Currently, we are working on identifying which depressed and anxious college students would benefit the most from guided internet-delivered cognitive-behavioral therapy (i-CBT) (as opposed to self-guided i-CBT or treatment-as-usual) to optimize resource allocation (Benjet et al., 2023). We discovered that actionable and clinically valuable treatment allocation rules could be constructed to determine which patient subgroups would benefit the most from guided i-CBT (Benjet et al., 2023). Also, using machine learning, we found that individuals with body dysmorphic disorder who have EF issues (such as difficulty developing new rules, learning from errors, and inhibiting unhelpful impulses) were less likely to respond to a guided CBT smartphone app (Zainal et al., in preparation-b). This highlights the need to consider alternative treatments for such patients. In separate studies, military men with lower recent depression severity and frequency, severe financial distress, and self-reported strong emotion regulation skills were more likely to respond to antidepressant medication (ADM) (Puac-Polanco et al., 2023) and combined ADM and psychotherapy (Bossarte et al., 2023). Together, these findings imply the potential value of combining emotion regulation training with ADM for patients with treatment-resistant depression.
Additionally, we have been learning latent profile analysis (LPA), another machine-learning technique that identifies distinct patient subgroups based on their pre-treatment symptoms, demographics, and related attributes. LPA helps predict different treatment outcomes for these subgroups. Our research found that subgroups derived from real-time ecological momentary assessments (EMA) were more effective in predicting panic disorder patients' treatment responses than traditional retrospectively-recalled self-report symptom assessments (Zainal et al., in preparation). These results suggest clinicians should prioritize real-time assessments of symptoms over retrospective ones to tailor interventions. Identifying treatment predictors and moderators (effect modifiers) can provide valuable insights for MH providers.
