Modelling Cardiovascular Condition Evolution In Hypertensive Population Using Graph Signal Processing

Graph signal processing (SP) is a new discipline that interprets data as a collection of signals defined on top of a graph. The nodes of the graph correspond to variables (features), with the links between nodes describing pair-wise relationships between the different variables. Graph signals are useful in several interesting fields, including medicine and health care. Our aim in this paper is to use graph SP to model and analyze clinical records of a chronic disease such as essential hypertension in a population. The ultimate goal is to identify prognostic factors and to assess the predictive value of features among the participants. Electronic clinical records of 1664 hypertensive patients were collected. The initial cohort was split into two groups: one group with patients with an incident cardiovascular (CV) event, and another group with patients without CV event. Clinical and analytic features were assessed, such as body mass index, blood pressure, cholesterol, albuminuria, and kidney function. By performing graph SP techniques, we provided a better understanding of pairwise interactions, correlation between features and conditional independence among them, which may help caregivers in designing an appropriate medical management in patients with chronic diseases such as essential hypertension, obesity and diabetes.