The Extent of Underestimation of Maternal Mortality in Developing Countries from Demographic and Health Surveys (DHS) Data
Saifuddin Ahmed, Johns Hopkins University
Qingfeng Li, Johns Hopkins University
Carolyn Scrafford, Johns Hopkins University
Thomas W. Pullum, ICF International
Background: The Demographic Health Surveys (DHS) remain as the main source of empirical data on maternal mortality in developing countries. This paper shows the problems of underestimation of MMR from DHS data due to high non-response rate on the timing of female deaths. Method: We use data from 111 Demographic and Health Surveys with siblings history. We examine MMR underestimation by four statistical methods used for missing data: complete case analysis, weighting for non-response rate, hotdeck imputation and Rubin’s multiple imputation. Results: Many countries had very high non-response rate on the timing of female deaths. The MMR estimates of the standard DHS method and complete case analysis were almost similar ( average MMR: 526 deaths per 100,000 live births) and likely to be biased in the presence of high non-response rate. All other methods show considerably higher MMR estimates, suggesting substantial underestimation problems. Conclusion: The current reporting of DHS may significantly underestimate maternal mortality in developing countries.