Prognostic enhancement in depressed patients on different non-implantable neuromodulation interventions

Abstract

A complementary approach to tackle the issue of modest tDCS clinical efficacy and severe cognitive effects of ECT is by identifying those more prone to experience such (adverse) effects. However, no reliable predictors of tDCS response have been identified. Although treatment-resistant depression is a robust clinical predictor of poor response for tDCS, this might be related more to the depressive episode per se rather than the intervention. Other predictors have shown mixed results. For instance, for tDCS, higher "dose" was associated with better outcomes in one meta-analysis, but the RCT using a higher dose than previous ones yielded non-significant findings. For ECT, even first line recommendations such as age, psychosis and melancholic features have not been consistently replicated. Likewise, no individualized risk approach has been developed for ECT, although some predictors of cognitive impairment that present large effects, such as bilateral ECT, higher charges, older age, and use of certain psychotropic and anesthetic agents have been identified. Here, we propose to enhance prediction by: (1) taking the biological pathways of the disease into account; represented in this project by the Neuroimaging module (M2) that will allow us to evaluate depression at multiple neurobiological levels; (2) multimodal big data collection, exemplified here by the neuroimaging (M2), cognitive (M1) and clinical phenotyping that will include cross-cutting transdiagnostic domain scales, as described below and; (3) methodological approaches for analyzing multi-dimensional and complex patterns of a manifold of data collected at multiple biological levels and finding generalizable predictive patterns to forecast future behaviors regardless of their mechanistic basis, which include supervised (e.g., XGBoost tree boosting, and elastic net) and unsupervised (e.g., hierarchical clustering and growth mixture modeling) methods to respectively enhance response prediction and identify latent subpopulations that manifest heterogeneous outcome trajectories based on clinical, cognitive and neuroimaging data. Finally, we will also use biomarker change as an outcome variable to explore which depression subtypes are particularly associated with changes in cognitive and neuroimaging assessment, which will further aid in disentangling the heterogeneous construct of depression by linking cluster of symptoms to changes in putative depression-related endophenotypes, providing further validation to them. (AU)