Spatially varying effects of measured confounding variables on disease risk

Abstract

Background: The presence of considerable spatial variability in incidence intensity suggests that risk factors are unevenly distributed in space and influence the geographical disease incidence distribution and pattern. As most human common diseases that challenge investigators are complex traits and as more factors associated with increased risk are discovered, statistical spatial models are needed that investigate geographical variability in the association between disease incidence and confounding variables and evaluate spatially varying effects on disease risk related to known or suspected risk factors. Information on geography that we focus on is geographical disease clusters of peak incidence and paucity of incidence.

Methods: We proposed and illustrated a statistical spatial model that incorporates information on known or hypothesized risk factors, previously detected geographical disease clusters of peak incidence and paucity of incidence, and their interactions as covariates into the framework of interaction regression models. The spatial scan statistic and the generalized map-based pattern recognition procedure that we recently developed were both considered for geographical disease cluster detection. The Freeman-Tukey transformation was applied to improve normality of distribution and approximately stabilize the variance in the model. We exemplified the proposed method by analyzing data on the spatial occurrence of sudden infant death syndrome (SIDS) with confounding variables of race and gender in North Carolina.

Results: The analysis revealed the presence of spatial variability in the association between SIDS incidence and race. We differentiated spatial effects of race on SIDS incidence among previously detected geographical disease clusters of peak incidence and incidence paucity and areas outside the geographical disease clusters, determined by the spatial scan statistic and the generalized map-based pattern recognition procedure. Our analysis showed the absence of spatial association between SIDS incidence and gender.

Conclusion: The application to the SIDS incidence data demonstrates the ability of our proposed model to estimate spatially varying associations between disease incidence and confounding variables and distinguish spatially related risk factors from spatially constant ones, providing valuable inference for targeted environmental and epidemiological surveillance and management, risk stratification, and thorough etiologic studies of disease.

Keywords: Disease cluster; Hierarchical disease cluster; Spatial association; Spatial scan statistic; Spatially varying; Sudden infant death syndrome.

Fig. 1

County-Specific SIDS Incidence Intensity Map in North Carolina with County Names

Fig. 2

A Normal Probability Plot of SIDS Incidence Rates. B Normal Probability Plot of Freeman-Tukey Transformed SIDS Incidence Rates

Fig. 3

A SIDS Incidence Intensity-Level Map in North Carolina by Generalized Map-based Pattern Recognition Procedure. Fifty-two medium-risk counties are indicated in white that are not considered to be used in constructing hierarchical intensity clusters of peak and low SIDS incidence. B Most Likely and Secondary SIDS Clusters Map in North Carolina by Spatial Scan Statistic. Seventy-six counties are indicated in white that do not lie in the most likely and secondary disease clusters of peak SIDS incidence or the most likely disease cluster of low SIDS incidence

Fig. 4

A Plot of Freeman-Tukey Transformed Non-White Live-Birth Proportion X^FT₁ versus Freeman-Tukey Transformed SIDS Incidence Y^FT for 99 North Carolina Counties and Fitted Regression Lines based on Generalized Map-based Pattern Recognition Procedure. Red Symbol and Blue Symbol Indicate Counties in Hierarchical Intensity Clusters of Peak Incidence and Incidence Paucity, Respectively. B Plot of Freeman-Tukey Transformed Non-White Live-Birth Proportion X^FT₁ versus Freeman-Tukey Transformed SIDS Incidence Y^FT for 99 North Carolina Counties and Fitted Regression Lines based on Spatial Scan Statistic. Red Symbol and Blue Symbol Indicate Counties in Likely SIDS Clusters of Peak Incidence and Incidence Paucity, Respectively

Fig. 5

Plot of SIDS Incidence × 1000 versus Non-White (Blue Symbol ) and Male (Red Symbol ) Live-Birth Proportion