How informative are vital registration data for estimating maternal mortality? A Bayesian analysis of WHO adjustment data and parameters

Monitoring maternal mortality is extremely challenging due to issues with data availability and quality. The maternal mortality estimates published by the WHO in 2012 include data adjustment parameters to account for data quality issues, but there is a discrepancy between the WHO assumption about, and observed variability in, misclassification errors in VR observations. Bayesian modeling approaches can be used to provide more data-driven insights into maternal mortality estimates and data adjustment parameters. We propose a Bayesian time series model for the VR adjustment parameters to assess the extent of VR misclassification errors and to provide a plausible assessment of the uncertainty associated with VR observations for which no external quantification of misreporting is available. We find that the proposed model gives a distribution for VR adjustments that is more comparable to the observed biases than the WHO expert distribution. However, given the lack of, and issues with maternal mortality data, validation of modeling assumptions and findings is challenging; more research on measuring maternal mortality and assessing data quality is urgently needed.