West Nile Virus (WNV) first appeared in Texas equids during June 2002. It has since spread rapidly across the state and apparently become endemic. Data from outbreaks occurring between 2002 and 2004 were analyzed to determine hotspots of equine WNV disease, identify environmental factors associated with outbreaks, and to create risk maps of locations with horses at a higher risk of the disease. Kriging was used to model the smoothed WNV attack rates, and interpolated rates were mapped to describe the spatial distribution of WNV disease risk in Texas. A retrospective time-space analysis using a Poisson model was conducted on each year’s data to identify clusters with high attack rates. The resulting overlapping yearly clusters were considered areas of hyperendemicity (hotspots). The counties identified as hotspots included Hockley, Lubbock, and Lynn (primary cluster) and Leon and Roberstson (secondary cluster). Environmental and geographic features were added to the disease maps and analyzed to determine possible environmental factors associated with outbreaks. Locations in close proximity to lakes, bird breeding routes, migratory flyway zones, crop farm and agricultural land, and all dense vegetation were found to be important environmental predictors. Finally, risk maps were created that combined surveillance data on WNV positive mosquito collections and wild bird WNV cases with previously identified environmental risk factors to predict areas of high occurrence of WNV. These risk maps could be used to implement various preventative measures to reduce the transmission of WNV in the Texas equine industry.

West Nile Virus (WNV) first appeared in Texas equids during June 2002. It has since spread rapidly across the state and apparently become endemic. Data from outbreaks occurring between 2002 and 2004 were analyzed to determine hotspots of equine WNV disease, identify environmental factors associated with outbreaks, and to create risk maps of locations with horses at a higher risk of the disease. Kriging was used to model the smoothed WNV attack rates, and interpolated rates were mapped to describe the spatial distribution of WNV disease risk in Texas. A retrospective time-space analysis using a Poisson model was conducted on each year's data to identify clusters with high attack rates. The resulting overlapping yearly clusters were considered areas of hyperendemicity (hotspots). The counties identified as hotspots included Hockley, Lubbock, and Lynn (primary cluster) and Leon and Roberstson (secondary cluster). Environmental and geographic features were added to the disease maps and analyzed to determine possible environmental factors associated with outbreaks. Locations in close proximity to lakes, bird breeding routes, migratory flyway zones, crop farm and agricultural land, and all dense vegetation were found to be important environmental predictors. Finally, risk maps were created that combined surveillance data on WNV positive mosquito collections and wild bird WNV cases with previously identified environmental risk factors to predict areas of high occurrence of WNV. These risk maps could be used to implement various preventative measures to reduce the transmission of WNV in the Texas equine industry.

The tsetse transmitted fatal disease of humans, sleeping sickness, is caused by two morphologically identical subspecies of the parasite T. brucei; T. b. rhodesiense and T. b. gambiense. Current distributions of the two forms of disease are not known to overlap in any area, and Uganda is the only country with transmission of both. The distribution of Rhodesian sleeping sickness in Uganda has expanded in recent years, with five districts newly affected since 1998. This movement has narrowed the gap between Rhodesian and Gambian sleeping sickness endemic areas, heightening concerns over a potential future overlap which would greatly complicate the diagnosis and treatment of the two diseases. An improved understanding of the social, environmental and climatic determinants of the distribution of Rhodesian sleeping sickness is required to allow more effective targeting of control measures and to prevent further spread and possible concurrence with Gambian sleeping sickness. The work presented in this thesis investigates the drivers of the distribution and spread of Rhodesian sleeping sickness in districts of central and eastern Uganda which form part of the recent disease focus extension. The spatial distribution of Rhodesian sleeping sickness was examined in Kaberamaido and Dokolo districts where the disease was first reported in 2004, using three different methodologies. A traditional one-step logistic regression analysis of disease prevalence was compared with a two-step hierarchical logistic regression analysis. The two-step method included the analysis of disease occurrence followed by the analysis of disease prevalence in areas with a high predicted probability of occurrence. These two methods were compared in terms of their predictive accuracy. The incorporation of a stochastic spatial effect to model the residual spatial autocorrelation was carried out using a Bayesian geostatistical approach. The geostatistical analysis was compared with the non-spatial models to assess the importance of spatial autocorrelation, to establish which method had the highest predictive accuracy and to establish which factors were the most significant in terms of the disease’s distribution. Links between Rhodesian sleeping sickness and landcover in Soroti district were also assessed using a matched case-control study design. Temporal trends in these relationships were observed using an annually stratified analysis to allow an exploration of the disease’s dispersion following its introduction to a previously unaffected area. This work expands on previous research that demonstrated the source of infection in this area to be the movement of untreated livestock from endemic areas through a local livestock market. With regards to the comparison of regression frameworks, the two-step regression compared favourably with the traditional one-step regression, but the Bayesian geostatistical analysis outperformed both in terms of predictive accuracy. Each of these regression methods highlighted the importance of distance to the closest livestock market on the distribution of Rhodesian sleeping sickness, indicating that the disease may have been introduced to this area via the movement of untreated cattle from endemic areas, despite the introduction of regulations requiring the treatment of livestock prior to sale. In addition, several other environmental and climatic variables were significantly associated with sleeping sickness occurrence and prevalence within the study area. The temporal stratification of the matched case-control analysis highlights the dispersion of sleeping sickness away from the point of introduction (livestock market) into more suitable areas; areas with higher proportions of seasonally flooding grassland, lower proportions of woodland and dense savannah and lower elevations. These findings relate to the habitat preferences of the predominant vector species in the study area; Glossina fuscipes fuscipes, which prefers riverine vegetation. The findings presented highlight the importance of the livestock reservoir as well as the climatic and environmental preferences of the tsetse fly vector for the introduction of Rhodesian sleeping sickness into previously unaffected areas, the subsequent spread of infection following an introduction and the equilibrium spatial distribution of the disease. By enhancing the knowledge base regarding the spatial determinants of the distribution of Rhodesian sleeping sickness within newly affected areas, future control efforts within Uganda may be better targeted to decrease prevalence and to prevent further spread of the disease.

The past twenty years of research that has associated air pollution with health outcomes has brought remarkable advance in statistical techniques that effectively tease out the intricacies of the relationship. However, while statistical techniques progressed, an assumption based on seminal work in the field persisted: that concentrations of particulate matter (PM) air pollution are spatially homogeneous within urban areas, and consequently, that personal exposures could be based on central monitoring site data alone. Although this assumption went unaddressed for years, it has now come to researchers' attention that it may be flawed and that the assumption may induce exposure misclassification error under certain conditions. This thesis explores intraurban spatial variability in PM through a systematic review of the literature, experimental field testing, modelling, and new methodological approaches. The key outcomes of the thesis are as follows: (i) the publication of the first systematic review of the intraurban particulate literature, challenging the widely-held assumption that PM concentrations are spatially uniform; (ii) an experimental test was conducted in Christchurch, New Zealand, revealing that the homogenous assumption was false for a city with high wintertime particulate matter concentrations; (iii) an integrated meteorological-emission model was evaluated for the first time at the intraurban level for PM and a new study design was suggested; and (iv) the spatial modification effect of social and ecological confounders was analysed with respect to respiratory hospital admissions and PM. Collectively, these outcomes provide a new body of knowledge informing researchers focused on assessing the relationship between air pollution and health in applications ranging from small-area exposure assessment to the wider field of environmental epidemiology.

Birth defects are the leading cause of infant death in the United States. Research has linked poor maternal micronutrient intake to birth defects including neural tube defects, heart defects, and oral clefts. After investigating spatial patterns of these birth defects in the City of Dallas and the neighborhood characteristics within clusters, geospatial access to supermarkets and convenience stores measured by proximity and concentrations are examined as environmental risk factors for nutrition-related birth defects. Spatial clusters of all three nutrition-related birth defects exist in the City of Dallas. Cluster for NTD occurs in vulnerable places with lower income and high minority population specifically Hispanics with no supermarkets. Cluster for heart defects mostly occurs in high income and predominantly white neighborhoods with many supermarkets. Clusters of oral clefts mostly occurs in middle-class income with relatively high minority populations with many convenience stores. For the entire study area, geographical access to supermarkets that include healthy foods are shown to be spatially reachable from most of mothers of infants with nutrition-related birth defects as well as convenience stores that typically include the majority of unhealthy processed foods with very few nutrients. Thus, not only easy geographical access to healthy food vendors but to convenience stores with low quality produces is observed at the same time.

In many spatial epidemiologic studies, investigators use residential location at diagnosis as a surrogate for unknown environmental exposures or as a geographic basis for assigning measured exposures. Inherently, they make assumptions about the timing and location of pertinent exposures which may prove problematic when studying long latency diseases such as cancer. In this work we explored how the association between environmental exposures and disease risk for long-latency health outcomes like cancer is affected by residential mobility. We used simulation studies conditioned on real data to evaluate the extent to which the commonly held assumption of no residential mobility 1) affected the ability of generalized additive models to detect areas of significantly elevated historic environmental exposure and 2) increased bias in the estimates of the relationship between environmental exposures and disease in a case-control study. While the literature suggests that some researchers have begun to develop methods to incorporate historic locations in studies of health outcomes, a number of questions remain. One reason for the knowledge gap is that residential histories have not been collected in most U.S. epidemiologic studies. In our work we evaluated the impact of using public-record database generated histories to estimate the effects of exposure in lieu of using subject-reported addresses collected during a study. Finally, we evaluated the effect of environmental exposure on cancer risk in a case-control study using an approach that combined a multiple membership conditional autoregressive (CAR) model with an environmental exposure index for temporally correlated time-varying exposure assigned based on residential histories. We used this model in a data application to explain bladder cancer risk in the New England Bladder Cancer Study. We included a temporal arsenic exposure index in the model to assess a large number of correlated arsenic exposures.

This dissertation explores the spatial patterns and place-based characteristics of colorectal cancer (CRC) late stage incidence and CRC-specific mortality in Miami-Dade County. Because CRC is the second leading cause of death among all cancers and is almost 90 percent preventable through medical screenings, investigations of CRC disparities across groups and communities are extremely relevant in the fight against cancer. This paper analyzes the geographic distribution of CRC cases in Miami-Dade County between two periods, 1988-1992 and 1998-2002 to: a) identify significant "hot spots" or clusters of disease; b) investigate associations of CRC patterns with neighborhood level characteristics such as socio-economic status, race/ethnicity, and poverty; and c) explore the policy implications of the spatial trends identified for the disease, with particular reference to the Welfare Reform Act of 1996. This dissertation analyzes data from the Florida Cancer Data Registry and tract level U.S. Census data, to identify the spatial distribution of CRC and study its relation to place-based variables using Geographic Information Systems (GIS) and spatial statistical modeling. Identifying spatial clusters of disease can assist in targeting public health interventions and improving social service delivery, particularly for uninsured populations. Identifying communities facing greater obstacles to screenings and quality medical care through the use of spatial analysis is an effort to mitigate these barriers while simultaneously providing empirically based evidence linking neighborhood-level social and economic conditions to health disparities.

An important prerequisite for running a successful insurance business is to predict risk. By forecasting the future in as much detail as possible, competitive advantages are created in terms of price differentiation. This work aims at using spatial point processes to provide a proposal for how the geographical position of the customer can be used in developing risk differentiation tools. For spatial variation in claim frequency an approach is presented which is common in spatial epidemiology by considering a group of policyholders, with and without claims, as a realisation of a multivariate Poisson point process in two dimensions. Claim costs are then included by considering the claims as a realisation of a point process with continuous marks. To describe the spatial variation in relative risk, demographic and socio-economic information from Swedish agencies have been used. The insurance data that have been used come from the insurance company If Skadeförsäkring AB, where also the work has been carried out. The result demonstrates problems with parametric modelling of the intensity of policyholders, which makes it difficult to validate the spatial varying intensity of claim frequency. Therefore different proposals of non-parametric estimation are discussed. Further, there are no tendencies that the selected information is able to explain the variation in claim costs. / En viktig förutsättning för att kunna bedriva en framgångsrik försäkringsverksamhet är att prediktera risk. Genom att på en så detaljerad nivå som möjligt kunna förutse framtiden skapas konkurrensfördelar i form av prisdifferentiering. Målet med detta arbete är att med hjälp av spatiala punktprocesser ge ett förslag på hur kunders geografiska position kan utvecklas som riskdifferentieringsverktyg. För spatial variation i skadefrekvens presenteras ett tillvägagångssätt som är vanligt inom spatial epidemiologi genom att betrakta en grupp försäkringstagare, med och utan skador, som en realisering av en multivariat Poissonprocess i två dimensioner. Skadekostnaderna inkluderas sedan genom att betrakta skadorna som en punktprocess med kontinuerliga märken. För att beskriva spatial variation i relativ risk används demografisk och socioekonomisk information från svenska myndigheter. De försäkringsdata som använts kommer från If Skadeförsäkring AB, där också arbetet har utförts. Resultatet påvisar problem med att parametriskt modellera intensiteten för försäkringstagare, vilket medför svårigheter att validera den skattade spatiala variationen i skadefrekvens, varför olika ickeparametriska förslag diskuteras. Vidare upptäcktes inga tendenser till att variationen i skadekostnad kan förklaras med den utvalda informationen.

Aim: The aim of this study was to: 1) investigate the geographic distribution of six pregnancy complications associated with future maternal cardiovascular disease risk in the province of Ontario and 2) to identify regions where women are likely to benefit from post-partum cardiovascular disease screening, based on the development of complications during pregnancy. Rationale: Cardiovascular disease is the leading cause of death in Canadian women. Pregnancy has been likened to a cardiovascular stress test and provides an early opportunity to assess a female’s lifetime risk of cardiovascular disease. Methods: This study was a retrospective analysis of data collected for the Niday Perinatal Database, provided by the Better Outcomes Registry & Network. Crude and age-standardized cumulative incidences of six pregnancy complications, and one or more pregnancy complications, were calculated for each Public Health Unit area in Ontario. The cumulative incidence of one or more pregnancy complications for women with no previous history of cardiovascular disease or traditional cardiovascular risk factors was calculated at the Public Health Unit and census subdivision area levels. Spatial statistics were applied to locate statistically significant clusters of high cumulative incidence. Results: Crude and age-standardized cumulative incidences of each pregnancy complication and one or more pregnancy complications varied across Public Health Unit areas in Ontario. The crude cumulative incidence of one or more complications ranged from 74 to 224 cases per 1000 pregnancies. The spatial analysis identified one statistically significant cluster of high cumulative incidence at the Public Health Unit area level, spanning the Lambton, Chatham-Kent, and Windsor-Essex Health Unit areas. Seven statistically significant clusters of high cumulative incidence census subdivisions were located within the following Public Health Unit areas: Chatham-Kent, Lambton, Middlesex-London, Ottawa, Leeds, Grenville and Lanark, Renfrew County, Simcoe Muskoka, Grey Bruce, and Eastern Ontario. Conclusion: Regional variation in the cumulative incidence of six pregnancy complications associated with cardiovascular disease risk was observed in Ontario. Statistically significant clusters of high cumulative incidence of one or more of these pregnancy complications were identified. These regions in particular may benefit from post-partum screening clinics and increased awareness regarding the association between pregnancy complications and cardiovascular disease. / Thesis (Master, Community Health & Epidemiology) -- Queen's University, 2012-07-27 13:56:51.945

The past twenty years of research that has associated air pollution with health outcomes has brought remarkable advance in statistical techniques that effectively tease out the intricacies of the relationship. However, while statistical techniques progressed, an assumption based on seminal work in the field persisted: that concentrations of particulate matter (PM) air pollution are spatially homogeneous within urban areas, and consequently, that personal exposures could be based on central monitoring site data alone. Although this assumption went unaddressed for years, it has now come to researchers' attention that it may be flawed and that the assumption may induce exposure misclassification error under certain conditions. This thesis explores intraurban spatial variability in PM through a systematic review of the literature, experimental field testing, modelling, and new methodological approaches. The key outcomes of the thesis are as follows: (i) the publication of the first systematic review of the intraurban particulate literature, challenging the widely-held assumption that PM concentrations are spatially uniform; (ii) an experimental test was conducted in Christchurch, New Zealand, revealing that the homogenous assumption was false for a city with high wintertime particulate matter concentrations; (iii) an integrated meteorological-emission model was evaluated for the first time at the intraurban level for PM and a new study design was suggested; and (iv) the spatial modification effect of social and ecological confounders was analysed with respect to respiratory hospital admissions and PM. Collectively, these outcomes provide a new body of knowledge informing researchers focused on assessing the relationship between air pollution and health in applications ranging from small-area exposure assessment to the wider field of environmental epidemiology.