This study has shown that even in remote, deprived populations a key informant system can produce reliable and plausible maternal mortality ratios at low cost. The key informant method has several important advantages. The system is designed to prospectively measure, rather than estimate, levels of maternal mortality in a given population. Most maternal mortality data for developing countries are estimates, derived retrospectively, from censuses, facility records and indirect methods such as adding questions to household or Demographic and Health Surveys (DHS). The 'sisterhood method' generates retrospective data and attempts to overcome sample size problems by asking all adult women in a household questions about the survival of their sisters, but often excludes information on the cause or circumstances surrounding the death [13] 'Networking' tries to further address sample size problems by asking women if they are aware of any women who died from maternal causes in the preceding year [14]. To avoid missing maternal deaths the key informant system identifies all deaths to women of reproductive age and subsequently exposes maternal deaths through a process of elimination. In this respect this system could be considered to be a prospective variant of 'Reproductive Age Mortality Surveys'. This has the further benefit that it can also provide useful information about non-maternal deaths to women of reproductive age if desired. Furthermore, since the key identifiers identify all births and neonatal outcomes this process also provides an accurate denominator.
Few Safer Motherhood programmes or research trials attempt to measure maternal mortality directly due to the large sample size required and the perceived high cost involved. The key informant system is acceptable, feasible and affordable for epidemiological surveillance of research or development projects. Mothers were only visited once, six to eight weeks after delivery. Compared with methods requiring frequent visits [4], which could potentially result in higher refusal rates, our system reduces costs and is less intrusive to families. The system may be seen as similar to conventional demographic surveillance systems, but differences include the use of incentivised key informants (who are not health workers and do not routinely visit every household in a defined area); the application across a large number of dispersed clusters; and the relatively low cost. The use of an incentive driven system avoids ethical dilemmas associated with the use of unpaid volunteers, while remaining low cost. The essential cost, in terms of deriving an estimate of the maternal mortality ratio is simply the cost of identification of deaths and verification, a monthly cost of US$386 in our area, or US $0.02 per capita per year.
Concerns that key informers may over report births or deaths to earn more money are invalid, as payment is only made once an identification has been verified by interviewers. In regards to missing births or deaths, the informants were allocated manageable geographical areas, based on whole hamlets or distinct parts of a village, which are familiar and accessible, minimising the risk of any births or deaths being missed. The death of any woman of reproductive age is a significant event that an informant living in the community would naturally be aware of, enabling deaths during pregnancy and late maternal deaths to be captured. However, migration does pose a problem.
The results presented in this paper are for women of the standard reproductive age group 15–49 years. In the study all deaths to women aged 10–50 are recorded to ensure there is no under-reporting of maternal deaths due to age restrictions, as early marriage is common in this setting and many people do not know their exact age. However, during the 110 week study period no maternal deaths have been found for this age-group and hence results presented in this paper are just for those aged 15–49.
This method for measuring maternal mortality appears to be robust, but as with all methods for measuring maternal mortality in the community there are clearly some limitations. Maternal deaths could be intentionally misclassified, especially with regards to unsafe abortion. There are a lot of suicides reported in this population, and the relatives of two women who died due to menstrual problems refused to give a detailed interview. It is possible that some of these women may have been pregnant or had unsafe abortions. Furthermore, using verbal autopsies to assign the primary cause of death is difficult when no formal provider is present. There was some uncertainty over assigning malaria, sepsis or anaemia as a cause of death without the availability of test results. Delays in collecting data could also lead to recall problems. The sample size of 13660 births, though large, may not be considered enough for precise maternal mortality measurement. However, an analysis of the MMR at 60 weeks from 7500 births produced a very similar MMR. Over time, or with a larger population, the increase in sample size would add to the robustness of the mortality measurements.
The findings from this surveillance system raise important issues for policymakers and health professionals. The system has produced high, but plausible birth and death rates in this rural, predominantly indigenous, population in India. Our observed crude birth rate of 28 births per 1000 population are consistent with state level reports of 29 in Jharkhand and 24 in Orissa [15]. The MMR of 722 per 100 000 live births was alarmingly high for a population unaffected by HIV. The most recent estimate of the MMR for India is 450 per 100 000 live births [10]. The MMR for Orissa has been reported as 367 per 100,000 live births [16]. No official figures are currently available for MMR for Jharkhand (independent of Bihar), but the state government has indicated that its goal for the MMR for 2006–07 was 407 per 100,000 live births and 325 for 2009–10 [17]. The high levels of mortality in our study area probably relate to socioeconomic deprivation in indigenous populations in remote areas in India and highlight a pressing need for interventions targeted at improving maternal and newborn care in these vulnerable populations.
The tenth version of the International Classification of Disease recommended two new additions with regard to measuring maternal mortality; pregnancy related deaths and late maternal deaths [18]. It is debatable whether the suicide and homicide deaths should be included as maternal deaths. There is mounting evidence that being pregnant may place women at greater risk of dying from suicide and homicide, and that unwanted pregnancies might be an important factor for the increased risk [19]. The inclusion of these two deaths from homicide and suicide would raise the MMR from 722 to 737 and including all pregnancy related deaths would raise it to 752 (i.e. remaining within the current confidence interval). This small difference suggests that pregnancy related deaths could be a useful proxy indicator for maternal deaths in similar contexts where cause of death data is unavailable. This supports the argument that it would be worthwhile including information on death certificates stipulating pregnancy status [20] to help measure maternal mortality. This system also enables the identification of late maternal deaths. One justification for this category is that modern life-sustaining procedures can prolong dying and delay death. This data show that although late maternal deaths even occur in a context with very limited access to quality health care, most maternal deaths did take place within the time boundaries of the current definition.
It has been estimated that two-thirds of maternal deaths occur in late pregnancy through to 48 hours after delivery [6]. This data showed a higher proportion (52%) of deaths occurring outside the intra-partum period, suggesting that there may be considerable variability in the contribution of the intra-partum period to maternal mortality in different contexts. Therefore policies solely focusing on interventions targeting this period may be less appropriate for vulnerable and excluded populations. Admittedly, some of those problems occurring in the postpartum period may be prevented by interventions in the intra-partum period, especially sepsis, and the higher proportion of deaths occurring outside the intra-partum period may be partly attributable to the high prevalence of malaria within this context.
The data show that in this remote rural population, most maternal deaths took place at home (60%) rather than in a facility (28%). Other studies suggest that most maternal deaths occur in facilities, but this evidence relies heavily on facility records or studies from countries with a higher proportion of facility deliveries rather than population data from poor rural communities [20]. It is also feasible that deaths at home in similar contexts are more likely to be missed by conventional methods for monitoring maternal mortality.