Counting Maternal Deaths? You Better Bayes It! a Systematic Assessment of Underreporting and Misclassification in Registration of Maternal Deaths in High and Middle Income Countrie

• Activity Number: 256 - Contributed Poster Presentations: Section on Bayesian Statistical Science
• Type: Contributed
• Date/Time: Monday, July 30, 2018 : 2:00 PM to 3:50 PM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Emily Peterson* and Leontine Alkema
• Companies: University of Massachusetts Amherst and University of Massachusetts Amherst
• Keywords: Bayesian methods; Maternal mortality estimation; Reporting errors; Misclassification; VR adjustment factors; Bias

Abstract

A maternal death is a death from any cause related to/aggravated by a pregnancy or its management. Vital registration(VR) systems record maternal deaths but this information is subject to substantial errors in cause classification, and in some instances, incompleteness of deaths. For selected country-periods, specialized studies have quantified reporting errors. A past assessment of such studies found that the median level of underreporting of maternal deaths was of an order of 50%. We propose a Bayesian hierarchical time series model to estimate VR adjustment factors, which are the relative adjustments necessary to remove bias from the number of maternal deaths reported in the VR. The model is parametrized around the sensitivity and specificity of maternal death classification and completeness of reporting,and uses data quality covariates and hierarchical time series sub models to construct estimates. This parametrization for VR adjustment factors allows for sharing of information across countries and within countries over time, and allows data of various forms to be included. We also assess whether specific subgroups of study design result in estimates of lower under-reporting.