Oral clefts are the most common form of birth defects in the United States (US) and the State of Utah has among the highest prevalence of oral clefts in the nation. The overall objective of this dissertation was to examine the spatial distribution of oral clefts and their linkage with a broad range of demographic, behavioral, social, economic, and environmental risk factors through the application of Geographic Information Systems (GIS) and spatial statistics. Using innovative linked micromaps plots, we investigated the geographic patterns of oral clefts occurrence from 1998 to 2002 and their relationships with maternal smoking rates and proportion of American Indians and Alaskan Natives (AIAN) at large scales across the US. The findings indicated higher oral clefts occurrence in the southwest and the midwest and lower occurrence in the east. Furthermore, these spatial patterns were significantly related to the smoking rates and AIAN. Then at the small area level, hierarchical Bayesian models were built to examine the spatial variation in oral clefts risk in the State of Utah from 1995 to 2004 and to assess association with mothers using tobacco, mothers consuming alcohol during pregnancy, and the proportion of mothers with no high school diploma. Next, multi-scalar spatial clustering and cluster techniques were used to test the hypothesis whether there was spatial clustering of oral clefts anywhere in the State of Utah and whether there were statistically significant local clusters with elevated oral cleft cases. Results generally revealed modest spatial variation in oral clefts risk in the State of Utah, with no pronounced spatial clustering, indicating environmental exposures are unlikely plausible cause of oral clefts. However, a few notable areas within Tri-County Local Health District, Provo/Brigham Young University, and North Orem had a tendency toward elevated oral clefts cases. Investigation of the maternal characteristics of these potential clusters supports the hypotheses that maternal smoking, lower education level, and family history are possible causes of oral clefts. Throughout this dissertation, we demonstrated how birth defects data collected by state and local surveillance systems coupled with GIS and spatial statistics methods can be useful in exploratory etiologic research of birth defects.

Several vector-borne and zoonotic diseases have emerged or re-emerged in Europe over these last decades. Besides climate change that influences disease risk at a regional scale, landscape changes could be responsible for local heterogeneities in disease risk. Spatial epidemiology tries to understand and predict spatial variations in disease risk by using spatial tools and spatially-explicit modelling methods. This study investigated the impact of fine-grained landscape patterns on the transmission of vector-borne and zoonotic infections in terms of habitat suitability for vectors and/or hosts and of exposure of people to infectious agents. This was studied through three human diseases emerging or at risk of re-emergence in Europe: the rodent-borne Puumala hantavirus, the tick-borne Lyme borreliosis and the mosquito-borne malaria infections. Statistical models were first used to study the relationships between environmental variables and host abundance, host prevalence, and human cases of Puumala hantavirus. Environmental factors were also combined with socio-economic factors to explain Puumala hantavirus and Lyme borreliosis incidence rates. The combination of factors explaining disease transmission and the complexity of such systems led to the development of an innovative, spatially-explicit modelling method: multi-agent simulation (MAS). The MALCAM simulation model was developed to assess the risk of malaria re-emergence in southern France and simulates spatial and temporal variations in contact rate between people and potential malaria vectors. The effect of changes in potential drivers of malaria re-emergence was also simulated. The different case studies showed that fine-grained landscape patterns influence the presence and abundance of vectors and hosts. Moreover, environmental conditions may also influence disease transmission through pathogen dispersal and the exposure of people to infectious agents. Finally, this study showed that people-vector contacts not only depend on the spatial distribution of people and potential vectors, but also on their behaviours and interactions.

Health studies often rely on aggregated instead of individual-level data to protect patient privacy. However, aggregated data are subject to the modifiable areal unit problem (MAUP), meaning results of statistical analyses may differ depending on the data's scale and areal unit. Past studies have suggested MAUP is context-specific and analyzing multiple spatial scales may provide richer understandings of examined phenomena. More research is needed to understand the role of scale and areal unit in health-related analyses. This study examines associations between surface mining and birth outcomes from 1989 to 2015 in Central Appalachia at the individual; postal; county; and county-sized, non-administrative scales. Evidence from previous studies suggests associations exist between health outcomes and county-level measures of mining activity. This is the first study to examine associations between mining and birth outcomes at more spatially refined exposure estimates. We identified surface mines using Landsat imagery and geocoded birth records. Airsheds, used to quantify the influence area of potential airborne pollutants from surface mining activity, were built using HYSPLIT4. The frequency values of each airshed that intersected each geocoded birth record were summed. These cumulative frequency airshed values were then aggregated. Finally, we implemented multiple regression models, each at a different scale, to examine associations between airsheds and birth outcomes. Results suggest MAUP has minimal impacts on the statistical results of examining associations between surface mining and birth outcomes in Central Appalachia. Results also indicate surface mining is significantly associated with preterm birth and reduced birthweight at each scale. / Master of Science / Health studies often rely on data that has been grouped together within political boundaries (e.g. counties) instead of individual-level data to protect patient privacy. However, results from analyses using grouped data may differ depending on the data's scale and areal unit, which describes the modifiable areal unit problem (MAUP). Past studies have suggested MAUP is specific to the situation being analyzed and examining multiple scales may provide richer understandings of the situation. More research is needed to understand the role of scale and areal unit choice in health-related analyses. This study examines associations between surface mining and birth outcomes from 1989 to 2015 in Central Appalachia at the individual; postal; county; and county-sized, non-administrative scales. Evidence from previous studies suggests associations exist between health outcomes and county-level measures of mining activity. This is the first study to examine associations between mining and birth outcomes at finer scales. Surface mines were identified using satellite images, and we identified the locations of birth records using the mother's home address. Airsheds, used to determine the influence area of airborne pollutants from surface mining activity, were created. We then used statistical models, to examine associations between airsheds and birth outcomes at four spatial scales. Results suggest MAUP has minimal impacts on the statistical results of examining associations between surface mining and birth outcomes in Central Appalachia. Results also indicate surface mining is significantly associated with preterm birth and decreased birthweight in grams at each scale.

Mandatory notification of disease forms the backbone of disease surveillance in New Zealand and overseas. Notification data is used by public health professionals and academics to identify cases requiring public health control, monitor disease incidence and distribution, and in epidemiological research. However, there is emerging evidence that notification rates do not accurately reflect the true extent of notifiable diseases within the community, resulting in the underascertainment of many notifiable cases. While adequate surveillance does not necessarily require that all cases of notifiable disease be captured, the systematic underascertainment of disease can have significant implications for perceived spatial and demographic trends in disease prevalence; potentially threatening the credibility of spatial epidemiological research by under or overestimating the burden of disease in different populations. There is evidence that systematic underascertainment occurs as a result of the differential actions of laboratories and general practitioners. It has also been recognised that that underascertainment can be influenced by a patient's willingness to seek medical attention and participate in laboratory tests. However, few studies have investigated whether these factors systematically influence notification either in New Zealand or overseas. Furthermore, the discipline of health geography has been slow to engage with this topic of public health importance, despite the inherently spatial nature of the processes involved, and the close ties to the geographic literature on health service utilization and healthcare provision. This thesis explores the spatial and temporal variation in notification rates in New Zealand for the period 1997-2005 and the potential relationships between notification rates and different variables. Unlike many underascertainment studies, which have used individual data and capture-recapture methods, data constraints inspired a unique ecological approach to investigating the factors which may be associated with notification in New Zealand. Variables were divided into categories based on Anderson's behavioural model for healthcare utilization and the influence of these variables on notification was determined through multiple regression analyses. The main findings of this research indicate that in New Zealand notification rates have increased during the period 1997-2005 and that there is a north-south gradient in notifications, with substantially lower rates in the North Island than in the South Island. Furthermore, it is also evident that the variables associated with notification vary according to disease, spatial aggregation and spatial scale. Notification rates are significantly associated with a range of predisposing and enabling factors which might influence patient choice to consult for many frequently underascertained diseases. More variation in enteric diseases is explained by the independent variables analysed than the variation in non-enteric diseases. However, further research into these relationships, and underascertainment in general, is required before firm conclusions can be drawn.

What explains the geographic distribution of pathogens? Better understanding and characterising disease patterns will help scientists to identify areas likely to host future epidemics and epizootics and to prioritise surveillance and intervention. However, the use of disease surveillance data to assess the risk of transmission and generate risk maps raises conceptual and methodological issues. Indeed, pathogens and more particularly viruses aren't ”species” like others that live in the open environment and must be studied with methods and concepts of their own. Avian influenza (AI), a disease caused by a virus infecting bird populations, has been selected to study these issues. AI has a major economic impact on the poultry industry in many countries, raises concerns of livelihood in low and middle-income countries, and represents a major concern for human health. The aim of this PhD thesis was to improve the knowledge on the spatial epidemiology of AI in different settings and conditions (i). For this, recent epizootics caused by the subtypes A (H5N1) and A (H7N9) were selected as case studies. First, highly pathogenic subtypes of the A (H5N1) virus have been studied in poultry farms (ducks and chickens) at different spatial scales: at the continental scale and the regional scale in the Mekong (Cambodia, Laos, Vietnam, Thailand) and the Nile Delta in Egypt. All these cases occurred between 2003, the date on which the virus starts to spread outside China, and 2015; the HPAI A (H5N1) subtypes are still reported today in many countries. Human infections caused by the A (H7N9) virus in China from March 2013 to 2017 were also studied. Studied different AI subtypes at different spatial scales within different host species also allowed to develop a conceptual model of AI transmission and to discuss the issue of the transferability of results in epidemiology (ii). Lastly, this PhD thesis leads to a discussion about the transfer of methods and concepts from ecology to spatial epidemiology, with a particular emphasis on their possible limitations (iii). / Doctorat en Sciences agronomiques et ingénierie biologique / info:eu-repo/semantics/nonPublished

Mandatory notification of disease forms the backbone of disease surveillance in New Zealand and overseas. Notification data is used by public health professionals and academics to identify cases requiring public health control, monitor disease incidence and distribution, and in epidemiological research. However, there is emerging evidence that notification rates do not accurately reflect the true extent of notifiable diseases within the community, resulting in the underascertainment of many notifiable cases. While adequate surveillance does not necessarily require that all cases of notifiable disease be captured, the systematic underascertainment of disease can have significant implications for perceived spatial and demographic trends in disease prevalence; potentially threatening the credibility of spatial epidemiological research by under or overestimating the burden of disease in different populations. There is evidence that systematic underascertainment occurs as a result of the differential actions of laboratories and general practitioners. It has also been recognised that that underascertainment can be influenced by a patient's willingness to seek medical attention and participate in laboratory tests. However, few studies have investigated whether these factors systematically influence notification either in New Zealand or overseas. Furthermore, the discipline of health geography has been slow to engage with this topic of public health importance, despite the inherently spatial nature of the processes involved, and the close ties to the geographic literature on health service utilization and healthcare provision. This thesis explores the spatial and temporal variation in notification rates in New Zealand for the period 1997-2005 and the potential relationships between notification rates and different variables. Unlike many underascertainment studies, which have used individual data and capture-recapture methods, data constraints inspired a unique ecological approach to investigating the factors which may be associated with notification in New Zealand. Variables were divided into categories based on Anderson's behavioural model for healthcare utilization and the influence of these variables on notification was determined through multiple regression analyses. The main findings of this research indicate that in New Zealand notification rates have increased during the period 1997-2005 and that there is a north-south gradient in notifications, with substantially lower rates in the North Island than in the South Island. Furthermore, it is also evident that the variables associated with notification vary according to disease, spatial aggregation and spatial scale. Notification rates are significantly associated with a range of predisposing and enabling factors which might influence patient choice to consult for many frequently underascertained diseases. More variation in enteric diseases is explained by the independent variables analysed than the variation in non-enteric diseases. However, further research into these relationships, and underascertainment in general, is required before firm conclusions can be drawn.

At the intersection of geography and public health, the field of spatial epidemiology seeks to use the tools of geospatial analysis to answer questions about disease. In this work we explore two areas: the use of geostatistical modeling as an extension of niche modeling, and the use of mobility metrics to augment modeling for epidemic responses. Niche modeling refers to the practice of using statistical methods to relate the underlying spatially distributed environmental variables to an outcome, typically presence or absence of a species. Such work is common in disease ecology, and often focuses on exploring the range of a disease vector or pathogen. The technique also allows one to explore the importance of each underlying regressor, and the effect it has on the outcome. We demonstrate that this concept can be extended, through geostatistical modeling, to explore non-logistic phenomena such as incidence. When combined with weather forecasts, such efforts can even predict incidence of an upcoming season, allowing us to estimate the total number of expected cases, and where we would expect to find them. We demonstrate this in Chapter 2, by forecasting the incidence of melioidosis in Australia given weather forecasts a year prior. We also evaluate the efficacy of this technique and explore the impact of environmental variables such as elevation on melioidosis. But these techniques are not limited to free-living and vector-borne pathogens. We theorize that they can also be applied to diseases that spread exclusively by person-to-person contact. Exploring this allows us to find areas of underreporting, as well as areas with unusual local forcing which might merit further investigation by the health department. We also explore this in Chapter 4, by relating the incidence of hepatitis C in rural Virginia to demographic data. The West African Ebola Outbreak of 2014 demonstrated the need to include mobility in predictive disease modeling. One can no longer assume that neglected tropical diseases will remain contained and immobile, and the assumption of random mixing across large areas is unwise. Our efforts with modeling mobility are twofold. In Chapter 3, we demonstrate the creation of mobility metrics from open source road and river network data. We then demonstrate the usefulness of such data in a meta-population patch model meant to forecast the spread of Ebola in the Democratic Republic of Congo. In Chapter 4, we also demonstrate that mobility data can be used to strengthen outbreak detection via hotspot analysis, and to augment incidence models by factoring in the incidence rates of neighboring areas. These efforts will allow health departments to more accurately forecast incidence, and more readily identify disease hotspots of atypical size and shape. / Doctor of Philosophy / The focus of this work is called “spatial epidemiology”, which combines geography with public health, to answer the where, and why, of disease. This is a growing field, and you’ve likely seen it in the news and media. Have you ever seen a map of the United States turning red in some virus disaster movie? The real thing looks a lot like that. After the Ebola outbreak of 2014, public health agencies wanted to know where the next one might hit. Now that there is another outbreak, we need to ask where and how will it spread? What areas are hardest hit, and how bad is it going to get? We can answer all these questions with spatial epidemiology. Our work adds to two aspects of spatial epidemiology: niche modeling, and mobility. We use niche modeling to determine where we could find certain diseases, usually those that are spread by insects or animals. Consider Lyme disease, you get it from the bite of a tick, and the tick gets it from a white-footed mouse. But both the mice and ticks only live in certain parts of the country. With niche modeling we can determine where those are, and we can also guess at what makes those areas attractive to the mice and ticks. Is it winter harshness, summer temperatures, rainfall, and/or elevation? Is it something else? In Chapter 2, we show that you can extend this idea. Instead of just looking at where the disease is, what if we could guess how many people will get infected? What if we could do so, a year in advance? We show that this can be done, but we need a good idea of what the weather will be like next year. In Chapter 4, we show that you can do the same thing with hepatitis C. Instead of Lyme’s ticks and mice, hepatitis C depends on drug-use, unregulated tattooing, and unsafe sex. And like with Lyme, these things are only found in certain places. Instead of temperature or rainfall, we now need to find areas with drug-problems and poverty. But we can get an idea of this from the Census Bureau, and we can make a map of hepatitis C as easily as we did for Lyme. But hepatitis C spreads person-to-person. So, we need some idea of how people move around the area. This is where mobility comes in. Mobility is important for most public health work, from detecting outbreaks to estimating where the disease will spread next. In Chapter 3, we show how one could create a mobility model for a rural area where few maps exist. We also show how to use that model to guess where the next cases of Ebola will show up. In Chapter 4, we show how you could use mobility to improve outbreak and hotspot detection. We also show how it’s used to help estimate the number of cases in an area. Because that number depends on how many cases are imported from the surrounding areas. And the only way to estimate that is with mobility.

La campylobactériose représente la principale cause de gastro-entérite bactérienne dans les pays industrialisés. L’épidémiologie de la maladie est complexe, impliquant plusieurs sources et voies de transmission. L’objectif principal de ce projet était d’étudier les facteurs environnementaux impliqués dans le risque de campylobactériose et les aspects méthodologiques pertinents à cette problématique à partir des cas humains déclarés au Québec (Canada) entre 1996 et 2006. Un schéma conceptuel des sources et voies de transmission de Campylobacter a d’abord été proposé suivant une synthèse des connaissances épidémiologiques tirées d’une revue de littérature extensive. Le risque d’une récurrence de campylobactériose a ensuite été décrit selon les caractéristiques des patients à partir de tables de survie et de modèles de régression logistique. Comparativement au risque de campylobactériose dans la population générale, le risque d’un épisode récurrent était plus élevé pour les quatre années suivant un épisode. Ce risque était similaire entre les genres, mais plus élevé pour les personnes de régions rurales et plus faible pour les enfants de moins de quatre ans. Ces résultats suggèrent une absence d’immunité durable ou de résilience clinique suivant un épisode déclaré et/ou une ré-exposition périodique. L’objectif suivant portait sur le choix de l’unité géographique dans les études écologiques. Neuf critères mesurables ont été proposés, couvrant la pertinence biologique, la communicabilité, l’accès aux données, la distribution des variables d’exposition, des cas et de la population, ainsi que la forme de l’unité. Ces critères ont été appliqués à des unités géographiques dérivées de cadre administratif, sanitaire ou naturel. La municipalité affichait la meilleure performance, étant donné les objectifs spécifiques considérés. Les associations entre l’incidence de campylobactériose et diverses variables (densité de volailles, densité de ruminants, abattoirs, température, précipitations, densité de population, pourcentage de diplomation) ont ensuite été comparées pour sept unités géographiques différentes en utilisant des modèles conditionnels autorégressifs. Le nombre de variables statistiquement significatives variait selon le degré d’agrégation, mais la direction des associations était constante. Les unités plus agrégées tendaient à démontrer des forces d’association plus élevées, mais plus variables, à l’exception de l’abattoir. Cette étude a souligné l’importance du choix de l’unité géographique d’analyse lors d’une utilisation d’un devis d’étude écologique. Finalement, les associations entre l’incidence de campylobactériose et des caractéristiques environnementales ont été décrites selon quatre groupes d’âge et deux périodes saisonnières d’après une étude écologique. Un modèle de Poisson multi-niveau a été utilisé pour la modélisation, avec la municipalité comme unité. Une densité de ruminant élevée était positivement associée avec l’incidence de campylobactériose, avec une force d’association diminuant selon l’âge. Une densité de volailles élevée et la présence d’un abattoir de volailles à fort volume d’abattage étaient également associées à une incidence plus élevée, mais seulement pour les personnes de 16 à 34 ans. Des associations ont également été détectées avec la densité de population et les précipitations. À l’exception de la densité de population, les associations étaient constantes entre les périodes saisonnières. Un contact étroit avec les animaux de ferme explique le plus vraisemblablement les associations trouvées. La spécificité d’âge et de saison devrait être considérée dans les études futures sur la campylobactériose et dans l’élaboration de mesures préventives. / Campylobacteriosis is a leading cause of acute bacterial gastro-enteritis in industrialized countries. The epidemiology of the disease is complex, involving many sources and transmission pathways. The principal objective of this project was to study environmental factors and methodological aspects pertinent to the spatial epidemiology of human campylobacteriosis using cases reported in Quebec (Canada) between 1996 and 2006. A conceptual diagram of sources and transmission pathways of Campylobacter was first proposed following a synthesis of current epidemiological knowledge based on a comprehensive literature review. The risk of recurrent episodes in relation to patient characteristics was described. Life table estimates and logistic regression were used for modeling. Compared to campylobacteriosis risk in the general population, the risk for a recurrent disease event was higher for a period of four years with a decreasing trend. This increased risk was similar across gender but higher for people from rural areas and lower for children under four years old. These results may suggest the absence of durable immunity or clinical resilience following a first episode and/or periodic re-exposure, at least among reported cases. Next, criteria were proposed and applied to ascertain the best geographical unit to use. Nine measurable criteria were proposed, including biological relevance, communicability of results, ease of data access, distribution of exposure variables, cases and population, and unit shape. These criteria were applied to various geographical units derived from administrative, health services and natural frameworks. Ultimately, the municipal geographical unit performed the best, given the specific objectives of the study. Future research areas for optimizing the choice of geographical unit were discussed. Another objective was to estimate and compare the associations between incidence and various environmental characteristics (poultry density, ruminant density, slaughterhouse, temperature, and precipitation) and demographic characteristics (population density, diploma) using seven different geographical units. Conditional autoregressive models were used for statistical modeling. In general, the number of significant predictors decreased as the aggregation level increased but directions of associations were consistent. More aggregated scales tended to show larger but more variable estimates for all variables, with the exception of the presence of slaughterhouses. This study highlighted the need for careful selection and analysis of geographical units when using ecological designs in epidemiological studies. Finally, the association between environmental characteristics and incidence in relation to four age groups and deux seasonal periods was studied. A multi-level Poisson regression model was used for modeling at the municipal level. High ruminant density was positively associated with incidence but decreased with age. High poultry density and presence of a large poultry slaughterhouse were also associated with higher incidence for people aged 16-34. Associations were also detected with population density and average daily precipitation. Except for population density, associations were constant across seasonal periods. Close contact with farm animals is most likely involved in the associations observed. Clearly, age and season must be considered in future studies on campylobacteriosis and in the design of preventive measures.

This dissertation examined risk factors for hypertensive disorders of pregnancy (HDP) — specifically whether neighborhood socioeconomic deprivation exacerbates individual socioeconomic disadvantage (deprivation amplification) to increase the likelihood of developing HDP. To select the optimal areal unit at which to investigate HDP, geographic proxies for neighborhoods were explored. A thematic review qualitatively examined nontraditional neighborhood boundaries identified through internet sources. Data from 2008–2012 Miami-Dade County, Florida birth records (n=121,421) and the U.S. Census Bureau were used for the remaining analyses. Ordinary least squares (OLS) and geographically weighted regression (GWR) analysis empirically compared the proportion of HDP prevalence explained by six areal units: census block groups, census tracts, ZIP code tabulation areas (ZCTAs), and three types of natural neighborhood — census units clustered based on an eight-item Neighborhood Deprivation Index. Multilevel logistic regression examined relationships between HDP, neighborhood deprivation, and individual-level factors. Odds ratios (OR) and adjusted odds ratios (aOR) were calculated. The thematic review found 22 potential alternatives to census boundaries developed through techniques such as crowd-sourcing and qualitative research. In the sensitivity analysis, census tracts aggregated at the scale of ZCTAs performed twice as well as any other model (GWR2 = 0.27) and were used as the Aim 3 unit of analysis. In the multilevel logistic regression, HDP was associated with moderate (aOR=1.13; CI: 1.05, 1.21) and high neighborhood deprivation (aOR=1.16; CI: 1.07, 1.26). Compared with mothers with private insurance, uninsured women (aOR=1.69; CI: 1.56, 1.84) and Medicaid recipients (aOR=1.12; CI: 1.05, 1.18) had higher HDP odds. Non-Hispanic Black women’s HDP odds were 1.58 times those of non-Hispanic White women. Cross-level interactions — between neighborhood deprivation and educational attainment and neighborhood deprivation and insurance status — did not reach statistical significance. Private sector neighborhood boundaries hold promise for developing new public health tools. Because they are relatively easy to generate from census data, natural neighborhoods may balance tradition and innovation. While no evidence of deprivation amplification was found, results suggested that individual-level and neighborhood deprivation are HDP risk factors. Interventions that target expectant mothers in deprived neighborhoods — particularly non-Hispanic Black and Hispanic women who lack health insurance — may help reduce HDP prevalence and disparities.

Background South Africa has experienced the most severe consequences of the HIV/AIDS pandemic. Every community has been affected in some way, many experiencing huge increases in mortality,particularly before antiretroviral therapies (ART) were readily available. However, the micro-level understanding of the HIV epidemic in South Africa is weak, because of a lack of detailed data for most of the population. This thesis is based on detailed individual follow-up in the Agincourt Health and Demographic Surveillance Site (HDSS) located in the Agincourt subdistrict of Mpumalanga Province and investigates micro-level determinants of HIV epidemiology and the impact of treatment provided. Methods The Agincourt HDSS has followed a geographically defined population since 1992,approximately the time when the HIV/AIDS epidemic first became apparent. This population based surveillance has included capturing details of all deaths, with cause of death determined by verbal autopsy, as well as the geographical location of individual households within the overall Agincourt area. Background information on the roll-out of ART over time was also recorded. Results A comparison immediately before and after the major roll-out of ART showed a substantial decrease in HIV-related mortality, greater in some local communities within the area than others. Individual determinants associated with a decreased risk of HIV/AIDS mortality included proximity to ART services, as well as being female, younger, and in higher socioeconomic and educational strata. There was a decrease in the use of traditional healthcare sources and an increase in the use of biomedical healthcare amongst those dying of HIV/AIDS between periods before and after the roll-out of ART. Conclusions Understanding micro-level determinants of HIV/AIDS infection and mortality was very important in terms of characterising the overall epidemic in this community. This approach will enable public health interventions to be more effectively targeted towards those who need them most in the continuing evolution of the HIV/AIDS epidemic.