Estimating maternal mortality using data from national civil registration vital statistics systems: A Bayesian hierarchical bivariate random walk model to estimate sensitivity and specificity of reporting

TL;DR

This paper introduces a Bayesian hierarchical model to evaluate and adjust for reporting errors in civil registration data, improving the accuracy of global maternal mortality estimates.

Contribution

A novel Bayesian bivariate random walk model that estimates sensitivity and specificity of maternal death reporting in CRVS systems, enhancing data reliability.

Findings

Model accurately predicts maternal death proportions in validation tests.

Effective adjustment factors improve maternal mortality estimates.

Model performs well without specialized study data.

Abstract

Civil registration vital statistics (CRVS) data are used to produce national estimates of maternal mortality, but are often subject to substantial reporting errors due to misclassification of maternal deaths. The accuracy of CRVS systems can be assessed by comparing CRVS-based counts of maternal and non-maternal deaths to those obtained from specialized studies, which are rigorous assessments of maternal mortality for a given country-period. We developed a Bayesian bivariate random walk model to estimate sensitivity and specificity of the reporting on maternal mortality in CRVS data, and associated CRVS adjustment factors. The model was fitted to a global data set of CRVS and specialized study data. Validation exercises suggest that the model performs well in terms of predicting CRVS-based proportions of maternal deaths for country-periods without specialized studies. The new model is used by the UN Maternal Mortality Inter-Agency Group to account for misclassification errors when estimating maternal mortality using CRVS data.