Data-driven healthcare management an analytics approach to blood pressure control and cardiovascular disease prevention
Bonifonte, Anthony J.
MetadataShow full item record
This thesis develops analytics based tools using operations research and statistical methodologies to address problems in the domain of cardiovascular disease and blood pressure management. Cardiovascular disease is the leading cause of death in the United States and worlwide, and a major component of healthcare costs. Elevated blood pressure (hypertension) has been shown as a significant risk factor for cardiovascular disease. We develop data-driven mathematical models to optimize treatment decisions, monitor blood pressure from mobile health technologies, and maximize population health. In chapter I, we show that history of a patient's blood pressure is an important predictor of future cardiovascular risk. Standard risk prediction models consider only a patient's current blood pressure and ignore history. Using Cox model survival analysis and the Framingham Heart Study data set, we demonstrate improved predictive ability by incorporating antecedent blood pressure measurements. The results of this chapter motivate the following chapters by emphasizing the importance of a time integrated measure of blood pressure as a risk for cardiovascular disease. In chapter II, we develop a population level optimal antihypertensive treatment policy. We model blood pressure as a continuous time, continuous state stochastic process, specifically a geometric Brownian motion mixture model, which we demonstrate is a good statistical fit. Using published parameters of cardiovascular disease risk as a function of blood pressure, we create a closed form analytical expression for the expectation and variance of hazard a patient experiences over the following T years. Using meta-analyses of randomized control trials, we estimate the effects of different dosages of antihypertensive treatment and optimize over treatment decisions. We create a threshold based population level optimal treatment policy for initiation and intensification of antihypertensive treatment, and show significant improvement over current guidelines in a large scale simulation model. In chapter III, we develop two changepoint detection-based algorithms for monitoring blood pressure from wearables and other mobile health technologies, which can gather many more measurements than traditional clinical measurements. The first algorithm uses knowledge of the disease progression to maintain a Bayesian belief of the true state. This method is highly accurate, but may be difficult to implement in practice due to the parameter estimation and necessity of simulation to calibrate the algorithm. We subsequently develop a Naive changepoint detection algorithm that is simple and generalizable to other continuous health characteristics such as cholesterol, glucose level, and pulse.