Statistical Applications in Geographical Health Studies

Résumé

This thesis consists of two related parts based on the study of health in a geographical region divided in a set of zones (small areas). The first part considers studies based on health information aggregated for each area into which the region under study has been divided. Specifically, it is a disease mapping application, based on generation of an Atlas of mortality in small areas of Catalonia over the period 1984-1998, using empirical Bayes methods. The second part considers an innovative approach, based on an integration of aggregated and individual health data in each of the zones of the region under study, using an estimating equation approach. Specifically, we consider this new approach as an extension of geographical regression. The elaboration of the first part of this thesis is justified for different reasons. First, health atlases and the mapping of health indicators in general, has demonstrated its great utility in identifying geographical localizations of health problems, in formulation of hypotheses about disease causes, and in monitoring public health interventions. Second, most atlases of mortality at the small area level present patterns of relative mortality risk for the most important causes of death using maps with a high level of geographical resolution. The first goal of this thesis was to construct a mortality Atlas involving a decomposition of the Autonomous Community of Catalonia into 289 small areas (municipalities or aggregates thereof) and 66 primary health areas of Barcelona city (being a small area but with a large population) for the period 1984-1998. For Catalonia as a whole, these maps presented, using a double-page format, the age adjusted relative risk, significantly high and low relative risk areas, relative risk in Barcelona City with respect to Catalonia and internally with respect to Barcelona, relative risk by age group (0-64 and ?65) and additionally the relative risk evolution over time in each area summarized in an single map, using spatial and temporal information modeled through Bayesian methods. Specifically, the atlas uses a strategy to include both: 1) relative risk evolution throughout the study period of each area compared to the average trend for all Catalonia and 2) the absolute relative risk evolution of each area. To our knowledge, this is the first time that both types of information have been combined in a single map. In addition, this is the first Atlas that presents information about geographical patterns in zones within small areas having a large population such as the cities of a country and includes life expectancy obtained with an empirical Bayes approach. The second part of this thesis can be useful in epidemiological studies where we include exposure and confounding variables that may have different sources of within and between-population variability. Specifically, analyses of individual disease-exposure data within a population are useful when exposure of interest varies sufficiently within the population. When the within-population variance of exposure is limited power of the individual-data analysis within a population is reduced. In such situations, aggregated-data analyses of disease data across populations, with a sample of individual exposure data from populations, can be powerful in estimating the exposure effect if between-population variation of exposure is large. However, although we may have knowledge of which variations dominate in each variable, exposure and/or confounding variables with different types of variation can be considered jointly. The second goal of this thesis was to consider a new analytical framework that is a combination of the individual- and aggregated-data analyses, based on an estimating equation approach (population-based estimating equation (PBEE) approach). The proposed analysis utilizes strengths from individual and aggregated health data approaches in the estimation of the exposure effect of interest, depending on which of the exposure variations (within- vs. between-population) dominates. Simulation studies under different scenarios were performed to show the strengths of the proposed approach in the estimation of the exposure effects of interest. Finally, we hope that some of the methods and topics employed may be of use to researchers who want to improve the study of health in space and time.