Longitudinal Beta GEE Modelling for Analysing Global and Regional Prevalence of Anaemia in Women

Abstract

- In this study, we use a beta regression approach to model the worldwide longitudinal prevalence of anaemia in pregnant and non-pregnant women. The estimates of anaemia prevalence from 1990 to 2016 are extracted for each country from the WHO Data Repository. Since the data for the subjects (i.e., countries) are clustered within sampling units, and the measurements within the same country are correlated, a beta-distributed Generalized Estimating Equation (GEE) model allowing for a population-averaged interpretation of the regression coefficients is fitted. The analysis is implemented in the SAS GLIMMIX procedure. Regardless, parameter coefficients in the GEE are estimated invariably; even if the covariance structure is miss-specified, a careful selection of the working correlation structure is performed to improve the efficiency of the estimates. Pregnancy and WHO regions had significant effects on the prevalence of anaemia. The significant interaction between pregnancy and time suggested that the decline in prevalence over time was larger in non-pregnant women than in pregnant women.