Comparing definitions of spatial relations for the analysis of geographic disparities in mortality within a Bayesian mixed-effects framework

• Published in: Revista brasileira de epidemiologia 2017-09, Vol.20 (3), p.487-500
• Main Author: Rojas-Gualdrón, Diego Fernando
• Format: Article
• Language: Portuguese
• Subjects: Health Policy & Services
• ISSN: 1415-790X
• DOI: 10.1590/1980-5497201700030011

Objective: To analyze the conceptual and technical differences between three definitions of spatial relations within a Bayesian mixed-effects framework: classical multilevel definition, spatial multiple membership definition and conditional autoregressive definition with an illustration of the estimate of geographic disparities in early neonatal mortality in Colombia, 2011-2014. Methods: A registry based cross-sectional study was conducted. Births and early neonatal deaths were obtained from the Colombian vital statistics registry for 2011-2014. Crude and adjusted Bayesian mixed effects regressions were performed for each definition of spatial relation. Model fit statistics, spatial autocorrelation of residuals and estimated mortality rates, geographic disparity measures, relative ratios and relative differences were compared. Results: The definition of spatial relations between municipalities based on the conditional autoregressive prior showed the best performance according to both fit statistics and residual spatial pattern analyses. Spatial multiple membership definition had a poor performance. Conclusion: Bayesian mixed effects regression with conditional autoregressive prior as an analytical framework may be an important contribution to epidemiological design as an improved alternative to ecological methods in the analyses of geographic disparities of mortality, considering potential ecological bias and spatial model misspecification.