Abstract:
|
When treating mental health conditions (such as depression), it is often the case that the distribution of outcome measures for different treatments show a high degree of overlap. The high level of heterogeneity makes it difficult to differentiate the impact of specific treatment effects on outcome measures. This problem complicates a common goal in precision medicine to identify baseline variables that can differentiate the effects of treatments on outcome. This poster illustrates a novel approach to precision psychiatry by identifying a linear biosignature (i.e., a linear combination of baseline features) that optimally separates outcome measures in terms of the biosignature. We illustrate this approach in the context of longitudinal outcomes in depression studies. A "convexity-based" clustering algorithm, a generalization of k-means clustering, is implemented that generates clusters that are maximally homogeneous with respect to the different treatments. Implementing this approach with depression trial data results in distinct outcome trajectories that distinguish specific effects of different treatments.
|