Diarrhea Morbidities in Small Areas: Accounting for Non-Stationarity in Sociodemographic Impacts using Bayesian Spatially Varying Coefficient Modelling

F B Osei,A Stein

doi:10.1038/s41598-017-10017-6

Abstract

Model-based estimation of diarrhea risk and understanding the dependency on sociodemographic factors is important for prioritizing interventions. It is unsuitable to calibrate regression model with a single set of coefficients, especially for large spatial domains. For this purpose, we developed a Bayesian hierarchical varying coefficient model to account for non-stationarity in the covariates. We used the integrated nested Laplace approximation for parameter estimation. Diarrhea morbidities in Ghana motivated our empirical study. Results indicated improvement regarding model fit and epidemiological benefits. The findings highlighted substantial spatial, temporal, and spatio-temporal heterogeneities in both diarrhea risk and the coefficients of the sociodemographic factors. Diarrhea risk in peri-urban and urban districts were 13.2% and 10.8% higher than rural districts, respectively. The varying coefficient model indicated further details, as the coefficients varied across districts. A unit increase in the proportion of inhabitants with unsafe liquid waste disposal was found to increase diarrhea risk by 11.5%, with higher percentages within the south-central parts through to the south-western parts. Districts with safe and unsafe drinking water sources unexpectedly had a similar risk, as were districts with safe and unsafe toilets. The findings show that site-specific interventions need to consider the varying effects of sociodemographic factors.

Highlights

Model-based estimation of diarrhea risk and understanding the dependency on sociodemographic factors is important for prioritizing interventions
The proportion of the population without safe toilets ranged from ≈19% to ≈98%; the proportion without safe drinking sources ranged from ≈8% to ≈98%; the proportion without access to safe liquid waste disposal ranged from ≈42% to ≈99%
Our study represents a contribution to spatial epidemiology literature by demonstrating both the methodological and epidemiological benefits of spatially varying coefficient modeling in estimating diarrhea risk

Summary

Introduction

Model-based estimation of diarrhea risk and understanding the dependency on sociodemographic factors is important for prioritizing interventions. The findings show that site-specific interventions need to consider the varying effects of sociodemographic factors Disease indices, such as the relative risk, of common morbidities are important criteria for comparison of neighborhood health status, neighborhood health planning, and health budgetary allocations. When geographic information on neighborhoods is available, the inclusion of independent Gaussian and conditional autoregressive (CAR) processes as spatially varying intercepts could be adequate to account for residual spatial effects These inclusions have the advantage to account for variance instabilities due to heterogeneous populations, unobserved influential factors, spatial interactions induced by similar sociodemographic conditions, and improve prediction accuracy. The assumption of stationarity in neighborhood sociodemographic effects is difficult to meet because of differences in neighborhood specific characteristic and unobserved factors that can locally influence disease outcomes. The manuscript intends to demonstrate the methodological significance and the substantive epidemiological implications

Methods

Results

Discussion

Conclusion