Chronic diseases are the leading cause of death in the United States, and diabetes in particular is one of the major underlying causes of stroke, coronary heart disease, kidney failure, and blindness. Despite extraordinary advances in the development of medications for treating chronic diseases like diabetes there are significant barriers to achieving their full benefits in practice. Existing treatment guidelines are often "one-size-fits-all" and do not account for the heterogeneous nature of patients. They are frequently focused on a single risk factor, while many chronic diseases have multiple risk factors that interact in a complex way. The goal of this project is to develop methods for measuring the benefits and the costs of treatment guidelines for chronic diseases, such as diabetes. Specifically, mathematical models will be created that can simulate the progression of chronic diseases over a patient's lifetime. Moreover, new algorithms will be designed that can be used to discover treatment guidelines that are optimized to account for individual patient risk factors, uncertainty health outcomes, and other factors influencing treatment decisions.
Formulation and validation of mathematical models for chronic disease progression requires the use of very large data sets with longitudinal population level data including laboratory, claims, and other data that makeup the complete electronic health record for a patient. Model parameter estimates are subject to unavoidable uncertainty as a result of patient heterogeneity, missing data, and conflicting estimates from different data sources. Additionally, complex interactions between disease risk factors, treatment efficacy, and medication adherence give rise to new large-scale optimization models that are difficult to solve. To address these challenges, this project will (a) create and validate new stochastic models for robust optimization of treatment guidelines that explicitly account for the heterogeneous nature of patients and their variation in adherence behavior; (b) analyze the mathematical structure of these models and develop methods for computing optimal treatment policies; (c) develop ways to integrate patient adherence behavior into treatment models; and (d) use insights from the optimal policies obtained to develop fast, easy-to-implement, near optimal guidelines, that account for individual patient preferences, and which can be implemented as an aid to clinicians and policy makers to improve population health.
