Estimability of association and effect measures in spatial case-control studies

Abstract

Case-control studies make a large contribution to epidemiological methodology, providing an efficient way of estimating population relative risk and attributable fraction. During the second half of the 20th Century methodological evolution was substantial in this area, incorporating diverse and complex sample designs, developing ways of estimating risk and statistical significance in stratified and multivariate designs, better control of confounding variables, identifying interactions in a more precise manner, and also, in particular situations, dispensing control samples to obtain valid results. Even with outstanding developments in geographic information systems and spatial analysis of data in the last 20 years, there still seems to be little use of the spatial dimension in case-control studies, although important studies have appeared in this area, particularly in the field of environmental epidemiology and the so called geographic epidemiology. In view of the above, this projects theme is to investigate a method: the spatial case-control study, understood as a case-control study which explicitly incorporates spatial location as a covariate of interest. This, given the non linearity of spatial distribution for most occurrences of epidemiological interest, implies a sweeping change to the ordinary analytical instrument used in case-control studies. The objectives of this project are in Case-control studies: a) to verify the behavior and epidemiological significance of the spatial relative risk function in function of case-base sampling, risk-set sampling sample designs and sampling from survivors of the studied disease; b) to develop a computer model which gives a set of predefined demographic, geographic, and epidemiological parameters, simulating the execution of spatial case-control studies in different epidemic configurations, more as an approach for a study of behavior and precision of spatial relative risk function estimator in function of the above cited sample designs; c) to apply a multinomial logistic regression model to estimate spatial distribution of occurrence risk when occurrences are classified according to their severity, this is in contraposition to the binary model which classifies studied individuals as either cases or controls, developing a multinomial treatment, which enables estimation of the spatial distribution of occurrence risk for cases in function of their severity; d) to develop significance tests for parameters of the non-linear component from surface estimates of risk in spatial case-control studies, which can be used in parallel or in place of the currently used Monte-Carlo procedure; e) to test the methodological results obtained, particularly the one in item c) above, in the reanalysis of two case-control studies: spatial distribution of work accident risk in the informal Piracicaba work market (FAPESP 05/03491-9) and spatial distribution of dengue fever risk in the south region of Campinas Municipal Area (FAPESP 06/01224-6)... (AU)