|
Abstract:
|
For surviving stroke victims, an unfortunate sequela is post-stroke depression (PSD), a disorder that increases mortality, cognitive impairment, and disability for years later. Not only do underlying biological processes increase a victim’s vulnerability to depression, but there is also a ‘psychosocial storm’ that originates due to the victim’s sudden disability. We analyzed a cohort of patients admitted within 3 months post-stroke to rehabilitation clinics at New York Presbyterian and Burke Rehabilitation Hospital who met DSM-IV criteria for major depression. Patients were randomized to receive either Ecosystem Focused Therapy or Education on Stroke and Depression. Upon initial modelling, no treatment differences existed in depression scores over time, as measured by the Hamilton Depression Rating Scale-24 items (HAM-D 24). To shed more light on patient PSD course for future development of personalized treatments, we propose to find ‘responder’ and ‘non-responder’ subgroups of HAM-D 24 scores through Latent Class Mixture Modelling (LCMM). We then propose to compare machine-learning (ML) methods (xgboost, random forests, neural networks, etc.) to predict the classified subgroups.
|
