An option for measuring maternal mortality in developing countries: a survey using community informants

The study was conducted in all 708 villages in Serang and Pandeglang Districts. In Serang, 38 sub-district health centres (HCs) provide services for almost 1.9 million people, similarly 34 HCs in Pandeglang, with about 1.1 million populations [6]. Banten Province ANC coverage is about 86% but skilled attendance at birth is only about 52% [4].

In this study, as for instance in DHS studies, the MMR is estimated using information on the number of Pregnancy Related Deaths (PRD) [7]. A PRD is defined as the death of a woman while pregnant or within 42 days of termination of pregnancy, irrespective of the cause of death. This category of case is introduced to facilitate identification of maternal death in circumstances in which attribution of cause might be inadequate. From verbal autopsies, we found that the proportion of women classified as PRD who died from maternal causes in the study area was very high (i.e. 97.5%) meaning that choice of PRD or maternal death definition is not of major importance.

The pilot and data collection process ran from January to June 2006.

The pilot phases were aimed at examining the advantages and disadvantages of various possible informant networks, and of different modes of accessing the informants' knowledge. A further feasibility study was conducted to assess the logistics and costs of a full-scale survey, using the most efficient administrative and data capture approaches identified from the pilot.

Kaders were chosen as forming the best network for identifying maternal deaths: they are established in the community and registered both in HCs and sub-district offices, making them easier to contact; their attendance rate was high (on average 90% compared to 52% for RTs), they were able to understand and fill-out the forms with a low error rate (5% compared to 9% for RTs); and are typically one third as many kader as RTs in a village, which reduced costs of data capture. However, because population density is much greater in urban areas and knowledge of neighbours less, it was felt many PRDs might be missed by kaders, and so it was decided to also include the RT network in urban areas. To enable capture-recapture estimates, RTs were also used in one randomly-chosen village per sub-district in rural areas.

The key activity was the 'listing-meeting' of village informants, to list details of deaths of women of reproductive age (WRA) in the village. Note that in villages where both kader and RT networks were used, separate listing meetings were held for each network, allowing for independency between the networks. Prior to the meetings, information, instructions, and forms were distributed to the village informants, for them to list information about WRA deaths during the previous two years. The most important items requested on the form were timing of death in relation to any pregnancy, age of woman, date of death, residence of woman, and name and address of a relative. Informants were asked to bring the completed forms to the listing meeting, where the study and forms were explained verbally, questions answered, and an opportunity given for informants to correct their forms in the light of improved understanding. The meeting then discussed all deaths listed by any of the informants, and collectively agreed a 'consolidated' list of WRA deaths, and of likely PRDs. To strive for total coverage of deaths, the MADE-IN process also included a visit to village informants who did not attend the meeting.

This consisted of visits to the named relative of likely PRDs on a consolidated MADE-IN listing meeting list. This was primarily to confirm or correct the details on the form. For 'eligible' deaths (i.e. PRDs within the defined period, and district), additional information about the circumstances and cause of death were collected through structured verbal autopsy of specific variables to allow estimation of cause of death using a computerised algorithm InterVA-M [8]. Socio-economic status and information on use of health services during pregnancy and near the time of death were also collected. Together these two steps - MADE-IN and MADE-FOR - provided village level estimates of the number of deaths of WRA, and in particular of PRDs in a defined period. Further details about the method are available at http://​www.​immpact-international.​org/​toolkit/​module4/​mimf/​index.​html. The method is summarised schematically in the following diagram (Figure 1).

Many sources can be used to provide estimates of live births by district. In Indonesia they include Demographic and Health Surveys, National Socioeconomic Surveys or censuses, and District Basic Health Research surveys. However, to enable useful comparisons between PRDs and all women with recent pregnancies, a stratified cluster random sample survey was conducted by Immpact, not only to estimate live births but also to describe characteristics and health status, among other variables. With some help with the sampling, such a survey is well within the capability of a team undertaking MADE-IN MADE-FOR. Villages were stratified as urban, rural or rural-remote. A total of 125 villages were sampled with probability proportional to size and with replacement. Within each sampled village, all household members were listed to identify all women with a live birth or stillbirth between April 2004 and March 2006. Eight births per village were then randomly selected, and information was collected from the families on the same background variables as for maternal deaths.

Ethical consideration was one of several issues ensured in this study. It includes permission, confidentiality, and research benefit for the community. Before the study conducted, ethical approval was sought and then issued by the Ethical Committee of Faculty of Public Health University of Indonesia. Permission was also obtained from related district health offices. At community level, permissions were obtained from all sub-districts and villages in the two districts. Before an interview was conducted, written consent was asked from each respondent after data collectors explained the objective, benefit, and process of the interview. All individual data was treated in a strictest confidentiality and no information on name was disclosed, even to the village or sub-district authority. Following the study, researcher provided feed back to various parties from various levels to make sure that public will achieve benefit from the study conducted. Feed back has been continuously given through written dissemination (publication) or seminar, at local, national, and international level.

The main limitation of the method deals with recall of the event. This method relies on informants' memories and knowledge to record deaths, thus the accuracy of the method depends on the length of recall period and size of the area covers by informants. This recall issue also related with the memory of the family in describing the circumstances of the death. Further detail about limitations of the method is available at http://​www.​immpact-international.​org/​toolkit/​module4/​mimf/​index.​html.

In urban and in 10% of rural villages, where we used both kader and RT networks to capture eligible (i.e. within the set period, and location) PRDs, 116 eligible PRDs were identified: 13 were identified only by the Kader network, 30 identified only by the RT network, and 73 identified by both. This gives a point estimate of 0.71 for the probability of an eligible PRD being identified by MADE-IN/MADE-FOR via Kader information, and an estimate of total eligible PRDs of 121. In other words, there were an estimated 5 extra eligible PRDs in 2004/5 in the areas with both RT and Kader informants, which were not identified by MADE-IN/MADE-FOR (see table 1).

Overall, we estimated that the RT network was more likely to capture PRDs than the kader network. This does not, however, mean that individual RTs are necessarily 'better' than individual kaders in identifying PRDs, since within the same area the number of RTs involved is much higher than the number of kaders.

From the pilot phases dataset, when all identified WRA deaths were visited, it was found that the village informants mis-classified 2 out of 25 eligible PRDs as not being eligible PRDs. On the other hand, 8 out of 31 deaths village identified by village informants as being eligible PRDs were false positives. Table 2 below shows the accuracy of the village informants in reporting eligible PRD status, by year and by area, using the MADE-FOR visit result as the "gold standard". The table shows that the kader network is generally more accurate in reporting eligible PRD status. There is a trend towards the eligible PRD status of more recent WRA deaths, and WRA deaths from rural areas, being more accurately reported.