Age Patterns of Mortality within Childhood in Sub-Saharan Africa

The age pattern of mortality in sub-Saharan Africa and how it varies across the continent remain poorly understood. The region lacks accurate registration statistics and assumptions about mortality patterns are needed to produce and smooth mortality estimates. These have had to be taken from model life table systems based on non-African data. Birth histories collected in national Demographic and Health Surveys are used to investigate age patterns of mortality in childhood in the sub-national regions of 26 countries of continental Sub-Saharan Africa. The majority of populations display a pattern of higher child relative to infant mortality than in any existing model system, including the Princeton "North" models. This reflects the existence of a "hump" of excess mortality in the late post-neonatal period and second year of life in more than three-quarters of sub-national populations. Age patterns of mortality vary markedly within and between countries, though adjacent parts of neighbouring countries sometimes have similar patterns. Particularly extreme relationships between infant and child mortality are most common in the Sahel, while a coastal belt exists adjoining the Indian Ocean with age patterns of mortality within the range of those in the Princeton models. A three-parameter model, which incorporates this "hump", is fitted to the data using Poisson regression and fitted national life tables are produced. Except for the southwest of Africa, no extensive areas exist with homogenous parameter values for the underlying downward slope of mortality with age in childhood and the size of the "hump" respectively. Thus, the scope for construction of "regional" childhood mortality models is limited. Nevertheless, age patterns of mortality in African populations tend to share features that differ from those of historical Western populations. Thus, using the national and regional average life tables in the indirect estimation of under-5 mortality yields more consistent series of estimates than are obtained using existing models.