Comparing two survey methods for estimating maternal and perinatal mortality in rural Cambodia

Article Outline

• Summary
• Introduction
• Context and methods
• Results
• Discussion
• Conclusion
• Acknowledgements
• References
• Glossary
• Copyright

Summary 

Purpose

We need solid estimates of maternal mortality rates (MMR) to monitor the impact of maternal care programs. Cambodian health authorities and WHO report the MMR in Cambodia at 450 per 100,000 live births. The figure is drawn from surveys where information is obtained by interviewing respondents about the survival of all their adult sisters (sisterhood method). The estimate is statistically imprecise, 95% confidence intervals ranging from 260 to 620/100,000. The MMR estimate is also uncertain due to under-reporting; where 80–90% of women deliver at home maternal fatalities may go undetected especially where mortality is highest, in remote rural areas. The aim of this study was to attain more reliable MMR estimates by using survey methods other than the sisterhood method prior to an intervention targeting obstetric rural emergencies.

Procedures

The study was carried out in rural Northwestern Cambodia where access to health services is poor and poverty, endemic diseases, and land mines are endemic. Two survey methods were applied in two separate sectors: a community-based survey gathering data from public sources and a household survey gathering data direct from primary sources.

Findings

There was no statistically significant difference between the two survey results for maternal deaths, both types of survey reported mortality rates around the public figure. The household survey reported a significantly higher perinatal mortality rate as compared to the community-based survey, 8.6% versus 5.0%. Also the household survey gave qualitative data important for a better understanding of the many problems faced by mothers giving birth in the remote villages. There are detection failures in both surveys; the failure rate may be as high as 30–40%.

Principle conclusion

Both survey methods are inaccurate, therefore inappropriate for evaluation of short-term changes of mortality rates. Surveys based on primary informants yield qualitative information about mothers’ hardships important for the design of future maternal care interventions.

Keywords: Maternal mortality, Perinatal mortality, Survey, Rural, Cambodia

Back to Article Outline

Introduction 

Maternal mortality rate (MMR) is an indicator of the risk of death related to pregnancy. In order to evaluate if risk reduction actually results from a given maternal care intervention, the traditional approach has been to compare the MMR before and after the intervention. For this we need reliable estimates of maternal and perinatal mortality rates which prove to be a complex task. One problem relates to statistics as the proportions for comparison, pre- and post-MMR, are very small and therefore uncertain even in countries reporting high MMR figures. The MMR in Cambodia is reported at 450/100,000, 95% confidence intervals ranging from 260 to 620/100,000.¹ Thus, a given reduction of MMR from a baseline level at 0.5–0.25% may seem impressive at first glance, but the difference is statistically insignificant due to the wide confidence intervals of the proportions under study. To obtain more precise estimates by expanding survey sample sizes, the sisterhood method was developed during the late 1980s. Data gathered by this method reflects a wide window of time (10–12 years) which is obviously a drawback in settings where endemic diseases of rapidly changing prevalences (HIV, malaria) contribute to the maternal fatality risk. Also sisterhood surveys are inappropriate in areas of high internal migration, e.g. due to recent war and post-conflict unrest.² Cambodia, a country where war ended 10 years ago and post-war migration is still extensive, the public MMR figure is questionable as it draws on sisterhood survey data. Under-reporting further adds to the uncertainty of MMR estimates. Often the populations most in need for such health programs are living in remote rural communities where the majority of women deliver at home without trained midwife assistance. In this context fatalities often go unreported and public mortality estimates based on urban or township surveys may need to be inflated by a factor of at least 1.5.² To get at more precise estimates, other survey methods for maternal mortality have been developed for low-income countries such as cemetery studies and verbal autopsy of deaths and near-miss cases. These methods also report inaccurate estimates.², ³ Recently the verbal autopsy method has been refined using probabilistic analysis to validate villagers’ information against physicians’ second-hand assessment.⁴, ⁵ The method implies that urban based doctors with no experience with the realities of village life define the “golden standard” for inter-rater analysis of villagers’ information, a dubious approach especially where the context is characterized by extensive social stratification and conflict. Survey methods should be adapted to the actual social context; methods that may work well in a semi-urban context, may not fit a rural scenario where many families are landless, migrant, and oppressed. The particular impact of the poverty factor on maternal death risk is demonstrated by Graham et al. who found that poverty may increase MMR by a factor of 3–4.⁶ As societies are stratified regarding income and welfare, MMR surveys should consequently be designed to fit the specific scenario of the actual maternal care intervention.

The aim of this study was to gain local baseline MMR estimates before launching a pilot program to address obstetric and perinatal emergencies in a rural sector in Cambodia (Delivery Life Support program). We wanted to compare the accuracy of two different survey methods to gain as reliable as possible pre-intervention baseline data: The first, a community-based approach gathering data from secondary sources such as village leaders, health centers, midwives, and village health volunteers; secondly, a household survey gathering data direct from all households.

Back to Article Outline

Context and methods 

The study area for the planned Delivery Life Support program comprises three rural sectors (Sector 1, 2, and 3) in the provinces of Pailin and Battambang in Northwestern Cambodia. The study population mainly consists of illiterate poor farmers and landless migrant workers. Diseases such as malaria Falciparum, HIV, and hepatitis B are endemic in the study area, and the health infrastructure is broken by decades of armed conflict. Parts of the target area, especially remote villages, are located inside the massive belt of land mines along the Thai-Cambodian border where access to health centers is virtually blocked during the rainy season.⁷, ⁸

The community-based survey was conducted retrospectively for the year 2004 in all 139 villages located in Sectors 1 and 2, villages located close to health centers as well as remote jungle communities. Semi-structured questionnaires were established and tested. The data gathering teams headed by the authors (HC, YVH, HS) consisted of local health workers well experienced in local customs and traditions. The informants comprised of all village heads (n=137), health center heads (n=11), midwives (n=19), and village health volunteers (n=171) in the two sectors.

The household survey was conducted in study Sector 3. We assumed that unregistered fatalities would be higher where access to health services was poor and poverty high. To explore the problem of dark numbers we therefore deliberately selected for survey the villages in Sector 3 having difficult access to medical service due to remote location, low income, and where roads were mine infested. In those villages all households (n=2762) were visited by teams of local health workers trained and headed by the authors (HC, YVH, HS). Data was collected retrospectively from a 3-year period (2002–2004) by open-ended conversations with special focus on histories of deaths and near-miss cases.

We define maternal death as a woman dying during pregnancy or within 6 weeks after termination of pregnancy. Perinatal death includes stillbirths and babies dying within 1 week after delivery, abortions (fetus less than 22 weeks) excluded. The Confidence Interval Analysis (CIA) program was used for analysis of proportions and their differences.⁹ We consider the difference between two proportions to be insignificant if the 95% confidence interval for the difference contains zero.

Back to Article Outline

Results 

The community-based survey registered a total of 4482 deliveries in year 2004. Of these, 20 mothers died during delivery or within 42 days after termination of the pregnancy indicating a maternal mortality rate of 446/100,000 (95% confidence interval from 0.25% to 0.64%). Of 4180 babies reported born, 209 died giving a perinatal mortality rate of 5.0% (95% confidence interval 4.3–5.7%). The household survey reported 14 dead mothers in a total of 3152 deliveries, a maternal mortality rate of 444/100,000 (95% confidence interval from 0.24% to 0.74%). We found in the household survey that 270 out of 3150 babies died, a perinatal mortality rate of 8.6% (95% confidence interval 7.6–9.6%).

Comparing the results of the two methods, there was no significant difference for maternal mortality registration (95% confidence interval for difference of proportions ranging from −0.3% to +0.3%). However, the direct household method reported significantly higher newborn mortality comparing to the community-based method (95% confidence interval for difference between 2.4% and 4.8%).

During the household survey we spent time talking to villagers. This gave us new information about the hardships mothers face in remote villages. Several mothers with severe bleeding after delivery had been carried in hammocks at night to get to the local health center. During the rainy season such transport could take hours due to muddy tracks and risk of land mines, and mothers died on the way from pregnancy-related complications. In some villages the mothers cross illegally to Thailand to deliver as they did not trust the local health service due to lack of skilled attendants and equipment. Several villagers also claimed that most trained health workers were arrogant, “rich people looking down upon us poor people”. Officially, the health centers are said to always have trained midwives on call. However, several informants reported that this was often not the case, instead the midwives took the mothers to private high-price clinics for assistance. The poorest families were hesitant to carry mothers with complicated deliveries to health centers and hospital because the “under table” charge for medical treatment was so high that they had to take up loans from village leaders or “rich” neighbors. For this reason they preferred to stay in the village even if the mother was in a critical condition, hoping that an experienced traditional birth attendant would be able to take her through. The information indicates that referral of risky deliveries is an additional economical burden to the already poor.

Equivalent information about hardships and sensitive political matters was not gained by the community-based survey method.

Back to Article Outline

Discussion 

Both types of survey under study gave similar results, the MMR estimates being equivalent to WHO's national estimates for Cambodia which are based on sisterhood surveys. We interpret the findings – not as a confirmation of the reliability of the MMR estimates, but rather as an indication of survey method failure; in circumstances where life is hard and mortality is high there seems to be facts that are hard to get access to regardless survey method. The assumption of inadequate access is confirmed when we study the findings of the perinatal mortality surveys. Firstly, we registered fewer babies than would be expected given the reported number of deliveries. The disproportion indicates inaccurate estimates, probably more so for the community-based method where at least 400 babies (9.5%) went unregistered. If we assume a twin birth rate at 2% in the survey area, at least 50–150 babies also went unregistered by the household survey. We doubt that the discrepancy between perinatal mortality rates reported for the two surveys is true; the social context, health conditions, widespread poverty, and access to delivery assistance is more or less the same in both areas, so probably the real perinatal mortality rate is approximately equal. Secondly, if we assume for reasons of estimation that the direct household survey gave correct rate of perinatal mortality, then 150 perinatal deaths went unreported in the community-based survey. That indicates a possible detection failure in the range of 30–40% for perinatal deaths for surveys based on secondary sources. There are definitely perinatal death detection failures also in the household survey, the range of which cannot be estimated.

Reasons for both unreported deliveries and fatalities may vary. Where migration is common, mothers, newborn and also village birth attendants may have left the study area during the survey period. For cultural reasons family members and birth attendants may feel ashamed of the fatalities and therefore keep stories untold to outsiders. Traditional birth attendants have been trained to refer critical cases according to national standards and lists of “danger signs”, but why refer if there is no money to pay for the treatment? So, there are clearly hidden painful stories of deaths that are not easily told. And if that is so for perinatal mortalities, we would expect that a number of maternal fatalities are also not recorded. This experience should not be regarded as specific for rural Cambodia. Songane and Bergström report detection failures as high as 86% in public surveys of maternal deaths in Mozambique.¹⁰ The WHO also assumes around 50% underreporting of fatalities in public registries in many low-income countries.²

Differences between the two survey populations under study make survey accuracy comparisons unfair. We conducted the household survey in populations probably poorer and with less access to health facilities (Sector 3) as compared to the community-based survey (Sectors 1 and 2). Graham's quality assessment of MMR studies illustrates that there may be substantial differences inside one country as the MMR is strongly stratified by the poverty factor. One reason why significantly higher rates of baby deaths were reported in Sector 3 may thus be the local poverty and not necessarily a better survey method.

Back to Article Outline

Conclusion 

Any population-based survey method seems to have inborn detection failures, especially so in remote and low-resource communities where mortality rates may be highest. Detection failures leave us with false low proportions of mortality. In addition, mortality point estimates are so inaccurate that they are useless as indicators of maternal care intervention effect. However, for policy making purposes household surveys may be useful; qualitative information gathered direct from primary informants yields sensitive information about poor mothers’ hardships enabling us to discover why and where the mothers die, and consequently design interventions fitted to the specific scenario.

To monitor short-term intervention effects, our experience from the actual MMR survey study made it mandatory to work out statistical methods better suited for comparisons of small proportions. We are at present analyzing the outcome of the Delivery Life Support program in Cambodia based on geometric probability distributions (waiting-time analysis) which hopefully will be more useful in estimation intervention effect. The results of that study are pending.

Back to Article Outline

Acknowledgements 

We want to thank National Center for Maternal and Child Care, Phnom Penh for advice and support. The study protocol is approved by the Cambodian and Norwegian Committees for Research Ethics.

The intervention is funded by the Norwegian Research Council and the Norwegian Ministry of Foreign Affairs and coordinated by Tromsoe Mine Victim Resource Center at the University Hospital Northern Norway.

Back to Article Outline

References 

1. Maternal Mortality in 2000. Estimates developed by WHO, UNICEF and UNFPA. Geneva: WHO; 2002.
   • View In Article
2. AbouZahr C, Donnay F, Filippi V, et al. The sisterhood method for estimating maternal mortality: guidance notes for potential users. WHO/RHT/97.28.
   • View In Article
3. Songane F. Lessons learnt from applying “the cemetery approach” and alternative methods in maternal mortality research [proceeding]. In: Enhancing Maternal Survival. A research priority in low-income countries. Karolinska Institute, Stockholm. 2002;
   • View In Article
4. Fanthahun M, Fottrell E, Berhane Y, et al. Assessing a new approach to verbal autopsy interpretation in a rural Ethiopian community: the interval model. Bull World Health Org. 2006;84:204–210
   • View In Article
5. Fottrell E, Byass P, Ouegraogo TW, et al. A probabilistic model for determining pregnancy-related causes of death from verbal autopsies. Popul Health Metr. 2007;5:1;http://www.pophealthmetrics.com/content/5/1/1, last accessed Oct 14, 2007
   • View In Article
6. Graham WJ, Fitzmaurice AE, Bell JS, Cairns JA. The familial technique for linking maternal death with poverty. Lancet. 2004;363:23–27
   • View In Article
   • Full Text
   • Full-Text PDF (80 KB)
   • CrossRef
7. Husum H, Heger T, Sundet M. Postinjury malaria: a study of trauma victims in Cambodia. J Trauma. 2002;52:259–266
   • View In Article
   • MEDLINE
8. Husum H, Gilbert M, Wisborg T, Heng YV, Murad M. Land mine injuries: a study of 708 victims in North Iraq and Cambodia. Mil Med. 2003;168:934–940
   • View In Article
   • MEDLINE
9. Confidence Interval Analysis [statistical computer program]. London: BMJ; 1992.
   • View In Article
10. Songane FF, Bergström S. Quality of registration of maternal deaths in Mozambique: a community-based study in rural and urban areas. Soc Sci Med. 2002;54:23–31
    • View In Article
    • MEDLINE
    • CrossRef