Date of Graduation

2017

Document Type

Dissertation

Degree Type

PhD

College

School of Public Health

Department

Epidemiology

Committee Chair

Tom Hulsey

Committee Co-Chair

Kim Innes

Committee Member

Jamison Conley

Committee Member

Miguella Mark-Carew

Abstract

Background and Objectives: Mapping and exploratory spatial data analyses are ideal tools for characterizing spread and occurrence of human Lyme disease infection. Unfortunately, many mapped displays utilizing Lyme disease surveillance data are prone to bias due to a lack of consideration for geographical confounders. The objectives of our study were to 1) characterize the geographic effects that boundary and travel related biases have on visualization of human Lyme disease occurrence and 2) apply these findings to develop a more precise methodology for evaluating efficacy of animal sentinel surveillance programs in predicting incidence of human Lyme disease infection.;Methods: County-level human Lyme disease and companion animal tick surveillance data were obtained from relevant state health departments. Data were organized within Microsoft Excel spreadsheets, and sorted by relevant reporting year and county. In Study 1, boundary effects were evaluated for the region containing Kentucky, Maryland, Ohio, Pennsylvania, Virginia, and West Virginia 2010-2014, utilizing a combination of rate smoothing and local indicators of spatial autocorrelation. Trends in disease clustering over time within our multistate region were evaluated utilizing logistic generalized estimating equations. In Study 2, travel associated biases were evaluated only for West Virginia confirmed Lyme disease cases 2011-2015, utilizing a combination of paired t-test, Wilcoxon Rank Signed test, and local indicators of spatial autocorrelation. In Study 3, the efficacy of the companion animal (dog and cat) sentinel surveillance program in West Virginia 2014-2016, was evaluated utilizing a combination of ordinary least squares and spatial regression techniques as well as local indicators of spatial autocorrelation on regression residuals.;Results: Study 1. Analyses indicated statistically significant ( P = 0.05) clustering of human Lyme disease incidence over time. High-high clusters aggregated near counties bordering high incidence states, while low-low clusters aggregated near shared county borders in non-high incidence states. Study 2. Analyses indicated statistical non-equivalency using paired t-test (t = 3.99, df = 54, P = 0.0002) and the non-parametric Wilcoxon Signed Rank test (S=264, P < 0.001) between total overall cases and those obtained within patient's home county, suggesting significant travel-associated bias. Additionally, local indicators of spatial autocorrelation detected statistically significant ( P = 0.05) patterns of clustering in the county level proportion of cases attributable to travel. Study 3. Regression analyses identified significant associations between confirmed cases of human Lyme disease and average number of Ixodes scapularis removed from dogs (ordinary least squares (beta=0.20 P < 0.001) and spatial lag (beta = 0.12, P = 0.002) models) but not cats for the period 2014-2016. Local indicators of spatial autocorrelation produced for spatial lag regression residuals indicated a decrease in model over and underestimation, but identified a higher number of statistically significant outliers than ordinary least squares regression.;Conclusions: Results of spatial and regression analyses 1) indicate significant differential clustering of incident human Lyme disease within WV and surrounding states over time; 2) suggest substantial travel-associated bias in Lyme disease case visualization within WV; and 3) strongly support the use of companion animal, and specially dog sentinel surveillance programs for estimation of human Lyme disease risk within WV. These findings suggest that geographic biases significantly affect visualization of human Lyme disease incidence and support the effectiveness of utilizing dogs as sentinel populations to estimate human risk. Findings of these three studies highlight the importance of using statistical methodologies that can accommodate the spatial structure imbedded within public health surveillance data.

Download

Share

COinS