• maternal mortality;
  • process indicators;
  • evaluation;
  • developing countries

Summary Objectives Process evaluation has become the mainstay of safe motherhood evaluation in developing countries, yet the extent to which indicators measuring access to obstetric services at the population level reflect levels of maternal mortality is uncertain. In this study we examine the association between population indicators of access to obstetric care and levels of maternal mortality in urban and rural West Africa.

Methods In this ecological study we used data on maternal mortality and access to obstetric services from two population-based studies conducted in 16 sites in eight West African countries: the Maternal Mortality and Obstetric Care in West Africa (MAMOCWA) study in rural Sénégal, Guinea-Bissau and The Gambia and the Morbidité Maternelle en Afrique de l'Ouest (MOMA) study in urban Burkina Faso, Côte d'Ivoire, Mali, Mauritanie, Niger and Sénégal.

Results In rural areas, maternal mortality, excluding early pregnancy deaths, was 601 per 100 000 live births, compared with 241 per 100 000 for urban areas [RR = 2.49 (CI 1.77–3.59)]. In urban areas, the vast majority of births took place in a health facility (83%) or with a skilled provider (69%), while 80% of the rural women gave birth at home without any skilled care. There was a relatively close link between levels of maternal mortality and the percentage of births with a skilled attendant (r = −0.65), in hospital (r = −0.54) or with a Caesarean section (r = −0.59), with marked clustering in urban and rural areas. Within urban or rural areas, none of the process indicators were associated with maternal mortality.

Conclusion Despite the limitations of this ecological study, there can be little doubt that the huge rural–urban differences in maternal mortality are due, at least in part, to differential access to high quality maternity care. Whether any of the indicators examined here will by themselves be good enough as a proxy for maternal mortality is doubtful however, as more than half of the variation in mortality remained unexplained by any one of them.