Pregnancy reporting and biases in under-five mortality in three African HDSSs.

In the absence of complete civil registration and vital statistics, Health and Demographic Surveillance Systems (HDSSs) are important sources of population-based data throughout sub-Saharan Africa. However, HDSS data on the vital status of newborns are often unreliable due to omission of those who were born and died between two rounds of data collection and are therefore never enumerated. This study investigates whether pregnancy registration improves estimation of under-five mortality (U5M) in three HDSSs in The Gambia, Kenya, and South Africa. We find that mortality is higher for children whose mother's pregnancy was observed than for children who were first registered after birth. Cox proportional hazards models with inverse probability weights further suggest that this difference is probably due to improved ascertainment of deaths in pregnancy cohorts and unlikely to be driven by a selection effect. These results highlight the importance of pregnancy registration in HDSSs for the estimation of U5M.