Big data guided interventions: Predicting treatment response

Abstract

Big data analytics and advanced statistical learning methods held an auspicious entry into neuropsychiatric research over the last decade. Especially for common multifactorial diseases as major depressive disorder (MDD), decisive advantages for diagnostics and prediction of treatment outcome phenotypes were both promised and expected. While a substantial amount of research was brought forward over the last years that already acknowledged the high potential of big data analytics for precision medicine in psychiatry, these expectations have so far been curbed by data management and methodological issues as well as difficulties inherent to the heterogeneous nature of neuropsychiatric disorders. Based on the example of MDD and treatment resistance in depression, this chapter will first give an overview of unsupervised machine learning algorithms targeting heterogeneity by surfacing subtypes of depression in a data driven manor. Supervised learning algorithms discussed next in this chapter are focused on predicting treatment outcome for antidepressant trials, based on clinical, genetic and imaging predictors. Finally, state-of-the-art machine learning design with prerequisites for successful and clinically meaningful application are discussed and prospects of their future in neuropsychiatric research are presented.