Transcranial direct current stimulation is a safe and efficient intervention for treating major depressive disorder. However, research has suggested heterogeneity of response between patients. The emerging field of precision psychiatry aims to use statistical modeling of multi-modal data to tailor treatment to the single patient. To this end, more in-depth analysis of randomized controlled trials will be relevant due to limited availability of other large datasets with high phenotypic detail and to develop tools for personalization within counterfactually controlled environments to (...) distinguish specific vs. non-specific patterns in treatment data. Previous research has aimed at identifying patient-related factors associated with better response. However, most analyses have operated on the group-level, ignoring natural clusters within the patients' constituting factors, their individual trajectories of symptom improvement, and their presented symptoms. Furthermore, group-based modeling strategies were limited to explanatory approaches using in-sample hypothesis-testing, that are ill-suited to prognosticate outcomes of single patients. This dissertation provides a methodological framework for reevaluation of existing clinical trial data to provide future investigations with more differentiated units of analysis and to complement explanatory approaches with predictive modeling strategies enabling prediction of single-patient outcomes. Using data from a landmark 3-arm clinical trial paradigmatic for a rigorously controlled experimental design the dissertation provides three blueprint studies for modeling heterogeneity of tDCS response: Study 1 characterized response to tDCS by considering patient-individual dynamics of symptom change over the course of treatment. Distinct trajectories of tDCS response could be identified, representing patient subgroups with varying strength and speed of improvement. These results suggest development of individualized treatment protocols and exploration of prolonged treatment courses. Study 2 reevaluated the efficacy of tDCS, in distinct, naturally occurring clusters of depressive symptoms. Using unsupervised machine learning, a global depression measure was parsed into 4 distinct symptom clusters. Analysis of cluster-scores showed superiority of tDCS and escitalopram over placebo in core depressive symptoms, but only tDCS was superior in improving sleep and only escitalopram was superior in improving guilt/anxiety symptoms, suggesting treatment selection based on patients' symptom profiles. In Study 3 supervised ML algorithms were employed to predict response to tDCS. In this proof-of-concept approach, response could be predicted above chance on the single-patient level, but overall accuracy was modest. Features employed for model training were explored using interpretable ML methods. Trained algorithms were provided to the field for expansion as well as tests of generalizability and incremental utility. The presented studies illustrate how in-depth secondary analyses of clinical trial data can aid personalization of treatment. The provided methodological framework can be expanded and generalized to other contexts and interventions that show heterogeneity of treatment effects. Yet, the empirical studies also epitomize challenges precision psychiatry is faced with, including low data availability, low outcome granularity, and limited external validation opportunities. The dissertation concludes with a discussion of challenges and future directions resulting from infrastructural demands in data acquisition, data management, data sharing, and interdisciplinary collaboration. (shrink)

Certain characteristics make machine learning a powerful tool for processing large amounts of data, and also particularly unsuitable for explanatory purposes. There are worries that its increasing use in science may sideline the explanatory goals of research. We analyze the key characteristics of ML that might have implications for the future directions in scientific research: epistemic opacity and the ‘theory-agnostic’ modeling. These characteristics are further analyzed in a comparison of ML with the traditional statistical methods, in order to demonstrate what (...) it is specifically that makes ML methodology substantially unsuitable for reaching explanations. The analysis is given broader philosophical context by connecting it with the views on the role of prediction and explanation in science, their relationship, and the value of explanation. We proceed to show, first, that ML disrupts the relationship between prediction and explanation commonly understood as a functional relationship. Then we show that the value of explanation is not exhausted in purely predictive functions, but rather has a ubiquitously recognized value for both science and everyday life. We then invoke two hypothetical scenarios with different degrees of automatization of science, which help test our intuitions on the role of explanation in science. The main question we address is whether ML will reorient or otherwise impact our standard explanatory practice. We conclude with a prognosis that ML would diversify science into purely predictively oriented research based on ML-like techniques and, on the other hand, remaining faithful to anthropocentric research focused on the search for explanation. (shrink)