Conclusion🔗

This dissertation argued that heterogeneity is a foundational problem in psychological treatment research. I presented six articles that, from a largely meta-analytic lens, aimed to improve the impact and precision of psychological intervention research.

One major innovation was the development of a Meta-Analytic Research Domain: a comprehensive set of harmonized living databases and custom software designed to enable rapid and fine-grained evidence generation. This infrastructure supported a comprehensive synthesis of psychotherapy effects across twelve mental health conditions, and enabled analysis of the worldwide expansion of psychotherapy research.

Another key advance was the creation of meta-analytic predictive models, which aim to forecast individualized treatment benefits while addressing cross-contextual variation. The value of this approach was demonstrated for an indirect digital intervention for depression as well as an Internet-based treatment for depression in chronic back pain patients. A further application explored the potential of ecological momentary assessment and digital phenotyping to forecast depressive symptoms within a digital psychological intervention. Here, we found that deep learning architectures can be improved by combining personalized learning with information from data across patients.

Finally, I presented an individual participant data meta-analysis summarizing the efficacy of psychological interventions in people with subthreshold depression. This analysis showed moderate effects on depressive symptom severity lasting up to 12 months—effects that remained clinically meaningful even for patients with very mild symptoms. By applying targeted superlearning, the study revealed considerable variability in estimated individual treatment benefits both within and across studies, in sharp contrast to the limited number of effect modifiers identified through conventional univariate analyses.