Date of Award

Fall 5-2019

Degree Name

Doctor of Philosophy (PhD)

Advisor(s)

Barry Davis, MD, PHD

Second Advisor

Wenyaw Chan, PHD

Third Advisor

Clifton Fuller, MD, PHD

Abstract

Cardiovascular disease (CVD), defined by the World Health Organization (WHO) as diseases that involve the heart and/or blood vessels is the number one cause of morbidity and mortality worldwide. In the United States more health care dollars are spent managing and treating CVD and/or its complications than any other disease process. Coronary heart disease (CHD) is the leading cause of deaths (43.8%) attributable to CVD, followed by stroke (16.8%), hypertension (9.4%) and heart failure (HF) (9%). CVD-related deaths and attendant morbidities, which include lifelong disability are in many cases preventable. This research proposes a dynamic risk model that handles multi-type recurrent events with a dependent terminating event in a competing risk framework, specifically nonfatal MI, stroke and HF, with all-cause mortality (death) as the dependent terminating event. A unique feature of this model is that it directly quantifies the baseline hazard for each recurrent CVD event and death, and the additional hazard that each recurrent event confers to its own recurrence and all other events. Positive and negative associations and relationships between all event types, recurrent and terminating, are established. The baseline hazard is dynamically updated with each event occurrence and affected by the types and number of events up to that point. The model is validated with a simulation study and applied to the Antihypertensive and Lipid-Lowering Treatment to Prevent Heart Attack Trial (ALLHAT) study. A procedure to assess the goodness of fit of the model is detailed. The model is further extended to incorporate risk factors for MI, stroke, HF and death such that each event type has unique risk factors [intrinsic hazards covariate model 1 (IHCM 1)]. Risk factors for the 4 event types imparted by antecedent nonfatal events is also established [intrinsic and recurrent hazards covariate model (IRHCM)]. Heterogeneity of ALLHAT treatment arm effects (amlodipine vs chlorthalidone; lisinopril vs chlorthalidone) on hazards by subgroup [sex, diabetes, race (black/nonblack), age, kidney disease, atrial fibrillation, hypertension treated at baseline, and stage 1/stage 2 hypertension] is studied (IHCM 2). Stabilization, fine tuning and validation of the model is performed by supervised learning, utilizing bagged training sets (70% and 60%) and test sets (30% and 40%) of IHCM 1 (250 bagged sets) and IRHCM (200 bagged sets, 70/30 training/test sets). Parameters are tuned and 95% confidence intervals (CI’s) constructed by the mean and standard deviation of the estimated parameters of the bagged training sets, respectively. Training set parameters applied to corresponding test sets yield similar and consistent goodness of fit measures for IHCM 1 and IRHCM, which suggests good generalization of the model without overfitting. Given the enormous global burden of CVD, this model is of great clinical import with significant potential to prevent and reduce future CVD events and develop into a risk assessment tool/decision rule, particularly in high-risk patients and delineate optimal treatment strategies tailored to the individual’s clinical profile.

Share

COinS