Sumve Survey on Adult and Childhood Mortality 1995

During May to October 1995 in the Mwanza Region of northwestern Tanzania, an in-depth survey on adult and childhood mortality estimation was conducted. Entitled the "Sumve Survey on Adult and Childhood Mortality" (SACM), the study was implemented by the Tanzanian Bureau of Statistics (BOS) and the Demographic and Health Surveys (DHS) program with assistance provided by several local institutions.

The primary objective of the SACM was to establish whether data useful for the estimation of childhood mortality rates (birth histories) could be collected by proxy from the mothers' sisters. The proxy data on deceased sisters--that is, women not interviewed in a routine demographic survey could be used to adjust estimates of childhood mortality where adult mortality is on the rise due to the AIDS pandemic. This type of data collection had not been attempted in Africa. Aside from the methodological aims of the SACM, the study was also intended to provide descriptive information on the demographic situation and use of basic maternity service utilization in the study area for purposes of local program evaluation.

The SACM was a two-phase data collection exercise conducted in the Kwimba District of Mwanza Region which lies on the southern boundary of Lake Victoria. This is an area where approximately 100,000 persons, of predominantly Sukuma ethnic origins, reside. Very little modern sector development has occurred in the study area and the large majority of the population relies on subsistence agriculture and some cash cropping to make a living. Educational levels are very low: the SACM results show that about 40 percent of women age 15-49 had never been to school, and only 1 percent had reached secondary school. Most of the study population falls in the catchment area of the Sumve primary health care (PHC) program, which aims to provide health education and basic maternal and child health services through outreach and referral programs. The PHC program is (and the SACM study was) based in Sumve where a relatively large hospital serves much of the district's tertiary care needs as well.

In Phase I of the SACM, a representative sample of 1,488 households and 2,130 women age 15-50 were interviewed. In these interviews, full birth histories of the respondents ("own" reports) and full sibling histories were collected. Based on information in the latter, all sisters born 15-50 years ago were identified and full birth histories were collected on all of these sisters ("proxy" reports). In Phase II of the SACM, conducted a month after Phase I, all living sisters age 15-50 living in an expanded study area were "tracked" with 2,123 of 2,223 eligible sisters (96 percent) eventually interviewed. From Phase II respondents was elicited essentially the same information as was obtained from Phase 1 respondents. These data allow comparisons of own-reported and proxy-reported birth histories. One drawback of the design is that the SACM sister-pairs are not representative of all sister-pairs since they live closer to each other than the average sister-pair (i.e., by design, the Phase II sisters live in roughly the same area as Phase I respondents). The SACM found that nearly all women (99 percent) who gave birth in the five years before the survey had received some kind of antenatal care during their last pregnancy, with the majority of services provided by nurses, midwives, and maternal and child health (MCH) aides. Only 2 percent of the women received care by a doctor. Unfortunately, the data indicate that over 90 percent of these women did not initially receive services before the second trimester, and 15 percent did not before the third trimester, which indicates that the full benefits of antenatal care are not being realized for most women around Sumve. The SACM also found that 62 percent of deliveries still occur outside of health facilities. Nearly all of these home deliveries are assisted by relatives and friends. Thirty-nine percent of deliveries were assisted by a trained health professional; in 4 percent of deliveries, a doctor assisted. Previous use of antenatal services and advice by a health professional to deliver in a health facility is positively correlated with subsequent delivery in a hospital or clinic. Of women not delivering in a health facility, the most commonly reported reason for nonuse of a facility was transport- or distance-related; 61 percent said that it was "too far," and 44 percent said that no transport was available.

The Phase I SACM data provided an opportunity to establish representative estimates of fertility and mortality.Women living in the Sumve area bear, on average,7.4 children during their lifetime,and nearly 60 percent have begun their reproductive lives before reaching age 20.The under-five mortality rate was estimated to be 134 deaths per 1,000 live births, meaning that about 1 in 7 children in this area do not survive to their fifth birthday.Infant mortality stands at 83 deaths (under age 1) per 1,000 live births.The risk of dying in early childhood is closely linked to the length of the birth interval.Infant mortality is about twice as high among children with short intervals (less than 24 months) than among children born after long intervals (48 or more months).

Adult mortality is high in the study area.The mortality rate for adult females (age 15-49) is estimated to be 4 per1,000 person-years and male mortality (age 15-49) is 5 per 1,000 person-years. While high, these mortality levels indicate that AIDS has not yet impacted significantly on adult mortality during the 0-13 year period before the survey (circa 1982-1995). A measure of female mortality attributable to maternity-related causes, the maternal mortality ratio, was calculated using the SACM. The maternal mortality ratio for the Sumve area was found to be around 500 maternal deaths per 100,000 live births.

Phase II of the SACM provided for linkage of 2,711 own-reported birth histories with 3,719 proxy reported birth histories (1.37 proxy reports per own report). The analyses of proxy reports vis-a-vis own reports demonstrate that women are familiar with their sisters' experience regarding childbearing and child deaths. The quality of the proxy information is, in some respects, surprisingly good. Yet the study identified some important problems related to proxy reporting. The precision of dating of births was significantly worse in the proxy reports, and substantial birth date displacement was evident. Most importantly, a considerable 14 percent fewer non-surviving births were reported in the proxy birth histories than in the own reports.

These data quality problems had some impact on demographic estimates.The directly-estimated total fertility rate for ages 15-39 (TFR) in the five-year period before the survey was estimated to be 6.7 children per woman from the own data, but 5.9 children per woman from the proxy data.While the own and proxy data produce similar childhood mortality rates for the five years before the survey (due to offsetting underreports of surviving and nonsurviving births),the proxy effect resulted in a 23 percent underestimate of under-five mortality 5-9 years before the survey, and a 31 percent underestimate 10-14 years before the survey. Trend estimates from the proxy data thus produce a picture of rising mortality, whereas own data indicate falling or stable mortality. These results suggest that routine implementation of a methodology to correct for mother's survival bias involving use of proxy data is not realistic at this time. However, in settings where moderate to severe bias is expected (five-fold or greater increases in adult mortality), careful adjustment to mortality estimates based on proxy data, while difficult to support empirically, may be an improvement over no adjustment at all.The adjustment would need to involve estimation of a "proxy effect" as well as estimation of the substantive correction parameter that reflects the survival bias.

Evaluation and quantification of the biases influencing childhood mortality estimation in sub-Saharan Africa should be undertaken. In this study, the children of recently deceased women had significantly elevated mortality relative to children of survivors: under-five risk was more than doubled (340 versus 143 per 1,000 live births). Additional information on the fertility-inhibiting impact of HIV/AIDS and current levels and trend in adult HIV/AIDS-related mortality needs to be garnered. These data should be population based and refer to a recent time period in order to be useful for program and policy purposes.