Background. To assess the completeness of various information sources and the subsequent estimates on maternal mortality.
Methods. Maternal deaths in the study area, rural northern Tanzania, in 1995 were identified from hospital records, health centers and dispensaries, registration by village leaders, follow up of an antenatal cohort, and a household survey. Data from some of these sources were also obtained in 1996.
Results. In 1995, 22 of a total of 26 maternal deaths were identified at the Haydom hospital. Three of the 15 deaths (20%) reported by the village leaders were not identified at any health facility. Four deaths were found in the antenatal cohort and one death in the household survey. Only two deaths were reported by the official statistics. Of the identified maternal deaths, 85% were found from health facility data. Including data from 1996, a total of 45 maternal deaths were identified; 13 of which were direct and 32 indirect obstetric deaths.
The 1995 estimated maternal mortality ratio, based on reports from the multiple source registrations, was 382 (95% confidence interval 250–560) per 100 000 live births. The antenatal cohort yielded an estimate of 322 (95% confidence interval 160–580). The ratio based on official figures for 1995 and 1996 combined was 123 (95% confidence interval 70–200).
Conclusions. Even a high quality routine registration of maternal deaths will miss a small proportion of cases. Investing in better registration of direct and indirect obstetric deaths will give better insight into this important health problem. Estimates based on official reports showed substantial underreporting.