PREDICTING CHRONIC DISEASES FROM HEALTHCARE DATA-A FRAMEWORK BASED ON GRAPH THEORY AND SOCIAL NETWORK MEASURES
The study illustrates a framework to predict the progression of chronic diseases from a new perspective using graph theory and social network analysis methods. The framework utilizes large and untapped longitudinal administrative data sets that contain ICD-10-AM disease codes that describe the principal and secondary diagnosis recorded during hospital admissions. The primary focus of the framework is to understand the health trajectory of chronic disease patients by mapping their admission history into a baseline network where nodes represent a particular disease and edges represent co-morbid relations between diseases. The framework iteratively updates and builds the baseline network. Next we apply social network and graph theory based methods to find patterns of co-morbidities that lead to chronic disease patterns that have a bearing on factors such as admission cost and length of stay. Based on the trajectory of disease patterns found in the baseline network, we have proposed a method that predicts the potential health trajectory for new patients who could be at risk. The research framework is being applied on type-2 diabetes as the exemplar chronic disease. The attached poster shows the baseline network along with major clusters of ICD codes found in longitudinal history of Diabetic patients.