Date of Award


Document Type

Open Access Dissertation

First Advisor

James R. Hébert


T Introduction: There are several unique characteristics in the epidemiology of prostate cancer (PrCA) that make it an interesting and important cancer to study. The first is that while prostate cancer is the most common cancer that men develop, it is one of the least common cancers that men die from. This indolent nature of PrCA has led to the idiom among health scientists that “men are more likely to die with PrCA than due to PrCA”. Just like other cancers, several individual-level risk factors (e.g., family history of the disease, age, and race) are well established for both PrCA incidence and mortality. It is becoming more common for health scientists to utilize innovative modeling techniques to describe the current epidemiologic characteristics of PrCA given its unique etiological nature and established individual-level risk factors. One such avenue of research is to understand how geography impacts the epidemiology of PrCA. The vast majority of studies incorporating a geographical component will model an area-level characteristic(s) and nest PrCA subjects within a geographical area where that area is treated as a random component (either as a random intercept or random effect). Yet another, more recent, approach is to account for the spatial autocorrelation of the geographical area of interest. Modeling individual-level geographical components (e.g., distance each subject travels to a healthcare facility), area-level geographical components (area-level risk factors and geographical area as a random effect) and the spatial autocorrelation of the vi geographical area of interest within a succinct modeling framework. It is this gap in knowledge in simultaneously accounting for multiple geographical levels and components in PrCA epidemiology that this research was undertaken. Methods: To evaluate how geographic context impacts the epidemiology of PrCA, this research was divided into three parts. The first investigated the incidence of PrCA, the second the mortality, and the third the mortality-to-incidence ratio (MIR). Electronic medical records from the United States (US) Veteran’s Health Administration (VHA) served as the data sources for all individual-level information and aggregate demographic information. The US 2010 Decennial Census and 2007-2011 5- Year American Community Survey (ACS) were used for all area-level information. The Social Vulnerability Indexes (SoVI®) was obtained from the University of South Carolina Hazards and Vulnerability Institute. The 2004 Rural-Urban Commuting Area Codes (RUCA) were obtained from the University of Washington. The geographical area of interest were all ZIP code tabulated area linked (ZCTA) ZIP codes in the state of South Carolina (SC) during the timeframe January 1, 1999 to December 31, 2015. Results: It was found that PrCA incidence among SC veterans who receive care at VHA facilities from 1999 – 2015 were at 35% and 39% increased risk if they reside in ZCTA-linked ZIP codes with SoVI® scores that were medium and high, respectively, as compared to those veterans residing in ZIP codes with low SoVI® scores. Also, the bestfitting model for PrCA incidence accounted for the random effect of SC ZCTA-linked ZIP codes and accounted for the spatial autocorrelation between SC ZCTA-linked ZIP codes. vii It was found that PrCA mortality among SC veterans who received are at VHA facilities from 1999 – 2015 were at 25% increased risk of death if they traveled more than 55 miles to receive care. Also, the best-fitting model for PrCA mortality accounted for the random effect of SC ZCTA-linked ZIP codes. It was found that the PrCA mortality-to-incidence ratio (MIR) was 0.17 for SC veterans who receive are at VHA facilities from 1999 – 2015. Stratified MIRs for standard risk factors for PrCA were found to vary by racial category, age at first VA visit, ZCTA-linked ZIP code level social vulnerability and ZIP code rurality. Collectively, it was found that MIRs by ZCTA-linked ZIP codes did not vary significantly in SC; however, two distinct clusters of ZCTA-linked ZIP code MIRs was found. Discussion: This research found that the epidemiology of PrCA among veterans who received care at VHA facilities from 1999 – 2015 varies by geographic location and context. This research has successfully used three distinct measures to characterize where a veteran resides: 1) linear distance to his most frequented VHA facility, 2) neighborhood characteristics of the ZIP code he resides in, and 3) the spatial autocorrelation of the ZIP code he resides in. The successful demonstration of the approach undertaken in this research has the potential to be replicated and expanded to use different geographical boundaries (e.g. counties), incorporate temporal factors, and evaluate other cancers and chronic diseases. This research approach has the potential to 1) allow health scientists to target areas of higher than- and lower than expected risk for PrCA, 2) allow clinicians to know if where a patient resides increases the risk for PrCA, and 3) provides further and detailed viii evidence to policy makers to understand that where a veteran resides can influence his/her health

Included in

Epidemiology Commons