Good morning. This morning I am going to share some experiences we have had using verbal autopsy methods to estimate maternal mortality using three different platforms. I would like to acknowledge my two co-authors, Emily Weaver who is here today, and Robert Mswia, who is unable to be here. This was very much a team effort with each of us contributing different expertise and perspectives.
Read study objectives.
As many people here probably know, maternal mortality is a high priority global public health issue. Each year over a quarter of a million women die from complications of pregnancy and childbirth. MDG5 aims to reduce the MMR by ¾ by 2015. However, we lack good data to measure progress toward this goal. Ideally, we would have complete, high quality vital registration data to track maternal mortality, but such systems are rare, and are particularly weak in settings with the highest maternal mortality.
In the absence of complete vital registration systems, several interim methods have been proposed to estimate maternal mortality. Some of these are facility-based, which has some advantages, but many deaths occur outside of facilities so here we are focusing on population-based methods. Several platforms exist to identify deaths in the population including censuses, sample vital registration systems, DSS, and household surveys. Once deaths have been identified, verbal autopsy can be used to assign a cause of death. There is an extensive literature on verbal autopsy methods that I will not go into here, but for those unfamiliar with the methods, this is essentially an interview with a close relative or caregiver of the deceased concerning the symptoms and circumstances of the death. This information is then used to assign a probable cause of death. Standard verbal autopsy tools and procedures have been developed by WHO in recent years.
In this study we review three community-based platforms for measuring maternal mortality using verbal autopsy: a sample vital registration system with verbal autopsy, or SAVVY system in Zambia, a post-census mortality survey in Mozambique, and a household survey in Bangladesh.
The household survey in Bangladesh was conducted in 2010. It was specifically designed to measure change in maternal mortality compared to an earlier survey in 2001. All deaths in selected households in the reference period October 2006 to the interview date were identified through a household survey. Verbal autopsy interviews were then conducted for all deaths identified among women age 13-49. There are two points to note here; first, the reference period is a little over 3 years, which is somewhat longer than is typically used for VA, and the VA focused only on WRA because of the survey focus was on maternal mortality. The post-census mortality survey in Mozambique was conducted in 2007. It was the first such survey ever conducted in Africa. Its goal was to obtain cause-specific mortality fractions at the provincial level for all causes of death and ages. In this survey, deaths in the household were identified from the 2007 census. All households in the census were asked about deaths in the household in the last year. A sample of census enumeration areas was selected for the PCMS (144 total) and all deaths identified in those households were selected for a follow up VA interview. This means that the second level sampling unit for the survey are deaths, not households per se. The data from Zambia come from the pilot phase of a new SAVVY system that is being established in Zambia. The SAVVY systems is designed to provide ongoing, nationally representative data on mortality and causes of death in Zambia. A baseline census was conducted in selected census enumeration areas in January 2010. In that census, all households were asked about deaths in the household in the last 12 months. Follow up VA interview were conducted for those deaths. In addition, deaths in the subsequent year were identified on an ongoing basis through community key informants and a follow up VA interview conducted. This study uses data in the pilot sites from both the census (retrospective) and the ongoing community surveillance (prospective) for a total reference period of 2 years. The Bangladesh survey included 168,629 households from which 18,609 deaths were identified including 878 deaths to WRA and 132 maternal deaths. The Mozambique PCMS identified a total of 10,080 deaths in the 12 month reference period of which 1,643 were among WRA and 259 were classified as maternal deaths. Note that the number of households from which these deaths were drawn is the number of households in the selected CEAs in the 2007 census. This information is not included in the data file we have so is not reported here but could in principle be obtained from the census. The Zambia SAVVY pilot sample is much smaller: 17,000 households from which 1,063 deaths were identified, of which 171 were among WRA and 18 were classified as maternal deaths. Additional notes: Table 1. Comparison of Sample Characteristics, (unweighted) a The sampling units for the Mozambique survey were deaths identified from the 2006 census not households (144 census segments; 64 single segments and 80 double segments). The relevant number of households from which deaths were identified is the total number of households in the selected CSA segments, which is unavailable. b Fieldwork was conducted Jan – Aug 2010. Only deaths 1-36 months before the household interview are included in all subsequent analyses (15,857 household deaths; 768 deaths to WRA; 108 maternal deaths). c Not all deaths occurring in the latter part of 2010 are expected to be included due to the lag time between a death being identified by a key informant and a verbal autopsy being conducted. d This table includes all deaths identified. Subsequent tables exclude deaths with missing information on age (0 in Bangladesh, 4 in Mozambique, and 46 in Zambia) or incomplete verbal autopsy data (2 in Bangladesh). e Maternal death statistics include late maternal deaths (1 in Bangladesh, 46 in Mozambique, 0 in Zambia) and maternal deaths with an underlying cause of HIV/AIDS (0 in Bangladesh, 33 in Mozambique, 3 in Zambia).
The analysis is primarily descriptive and addresses the four areas listed here.
Now turning to some results. In the interests of time we present a few selected findings. Further analyses can be found in the paper which is available on the conference web site.
The causes of death were determined by physician review of the verbal autopsy responses in all three countries. In general, less than 10% of deaths had an undetermined cause of death. The percentage was lower in Zambia and a bit higher in Bangladesh, but overall these levels of undetermined COD are relatively low.
Here we present three key maternal mortality indicators from the three platforms. The percentage of deaths among WRA that are classified as maternal deaths is available from all three platforms and ranges from 8.8% in Zambia to 17.3% in Mozambique. The MMRate is only available for Bangladesh and Zambia. This is because this rate requires the number of WRA for the denominator. For Mozambique, this denominator comes from the census and was not available directly in the PCMS data set. In principle, it is possible to calculate the MMRate by linking the PCMS data with the census data, but in practice, there is a lag in the processing of the census data and the denominators may not be available until some time after the PCMS data are available. We plan to explore linking with the census data in future work. The MMR was only available for Bangladesh, where it is 197 per 100,000 live births. Again, the MMR could be calculated for Mozambique if the PCMS can be linked to the census data since the census contains information on live births. In Zambia, data on live births was not collected in the pilot phase of the SAVVY so it is not possible to calculate the MMR. However, data on live births is now being collected in the scale up of the system based on the pilot experience so it should be possible to calculate the MMR from the SAVVY system in the future.
This graph shows the distribution of maternal deaths by type from the three countries. For this analysis we chose to include late maternal deaths, although only Mozambique reported a significant number of such deaths (46), the majority of which were AIDS-related. The percentage of indirect maternal deaths looks a little low in Zambia but this is based on a very small number of deaths so should be interpreted cautiously.
Here we present the distribution of deaths to WRA by age for the countries. Mozambique and Zambia show quite similar age patterns of deaths among WRA with a peak among women age 25-29. Bangladesh exhibits a different pattern with a sharp increase in the percentage of deaths occuring after age 35. Bangladesh has lower fertility and mortality, much lower HIV prevalence, and an older population age structure which all influence the distribution of deaths among WRA.
So, turning now to the discussion. I will just hit a few highlights here in the interests of time but the paper includes further details. First, not all common maternal mortality indicators could be obtained from all the platforms without linking to other sources (Mozambique) or additional data collection (Zambia). This is a consideration when choosing a platform and also illustrates the need to plan ahead with any data collection platform; there have been cases where rates could not be calculated from HH surveys because the need to link to denominators in the HH component of the survey was not adequately anticipated. The Zambia and Mozambique platforms were not specifically focused on maternal mortality so when using general mortality platforms, any special considerations for maternal mortality (such as collection of data on live births) need to be anticipated.
Second, even with very large sample sizes and using recall periods of up to 3 years, the number of maternal deaths observed was small. The number of maternal deaths observed will depend on several factors including mortality levels, fertility levels, and age structure, and the sample size requirements will change as these change, typically towards ever-increasing sample sizes as fertility and mortality decline. Third, the periodicity of these platforms varies. The PCMS has the advantage of piggy-backing onto the census to identify deaths, but this means it can only be done when a census is done about every 10 years. However, in practice sample size requirements means that doing household surveys more frequently than every 10 years is unlikely to be feasible; the Bangladesh survey was done 9 years after the previous one to detect a 20% decline over that period which required a sample of almost 170K households. The SAVVY system is the only one to provide data on an ongoing basis but the number of maternal deaths each year is small so data will have to be cumulated over several years to obtain stable estimates (although the system will provide more frequent estimates for the most common COD such as AIDS and malaria).
To conclude, our experiences with these platforms suggest that all three are feasible options for obtaining estimates of maternal mortality using VA methods. Each has pros and cons (which are discussed in more detail in the paper) which need to be weighed in selecting a platform. However, none of the platforms are likely to provide estimates of change in MM over short time periods. They are therefore a compliment nor a substitute for long term investment in vital registration systems. However, we know that investment in VR is very long term endeavor. While these methods are viable interim options there is no panacea for tracking progress in reducing MM.
Applications for Measuring Maternal Mortality: Three Case Studies Using Verbal Autopsy Methodology
Applications for measuringApplications for measuring
maternal mortality:maternal mortality:
three case studies using verbalthree case studies using verbal
autopsy methodologyautopsy methodology
Siân Curtis, University of North Carolina at Chapel Hill
Robert Mswia, Futures Group
Emily Weaver, University of North Carolina at Chapel Hill
XXVII IUSSP International Population Conference, 2013
Study objectiveStudy objective
To review and contrast three community-based
platforms for measuring maternal mortality using
verbal autopsy with:
(1) a post-census mortality survey
(2) a large-scale demographic household survey
(3) a sample vital registration system
Maternal mortality in the spotlightMaternal mortality in the spotlight
~287,000 women die
each year from
pregnancy or childbirth
MDG 5 aims to reduce the
maternal mortality ratio by
three quarters by 2015
Lack of good quality data
toward progress of this
Interim methods - several optionsInterim methods - several options
Population-based interim methods identify deaths
■ population census
■ sample registration systems
■ demographic surveillance sites
■ household surveys
Need additional methods to identify deaths due to
maternal causes Verbal autopsy
Three case studies using verbalThree case studies using verbal
autopsy methodologyautopsy methodology
Photo credit: Photobucket.com
Sample CharacteristicsSample Characteristics
168,629 -- 17,000
Oct 2006 –
Aug 2006 –
Feb 2009 –
Compare and contrast plaforms with regard to:
1. death identification and classification
2. estimating maternal mortality ratios and
3. sample sizes and periodicity of estimates
4. data quality
Maternal Mortality IndicatorsMaternal Mortality Indicators
Proportion of deaths that are maternal for
14.2 17.3 8.8
MMRate per 100,000 women of WRA 17.0 na 69.1
MMR for WRAa
(per 100,000 live births) 197 na na
Maternal Deaths by TypeMaternal Deaths by Type
Distribution of deaths among women aged 15-Distribution of deaths among women aged 15-
49 by age group and country, weighted49 by age group and country, weighted
Not all common
available from all
platforms without further
Estimating maternal mortality ratios and rates
Sample sizes and periodicity of estimates
■ Number of maternal deaths observed is small even
with large sample sizes
■ Number of maternal deaths will depend on
mortality level, fertility level and age structure
■ Sample size requirements will change as these change
■ Periodicity of platforms varies
■ PCMS approx every 10 years with census
■ SAVVY continual data collection
■ 3 platforms studied are feasible for use with verbal
autopsy to estimate maternal mortality indicators
■ Weigh pros and cons of each platform with
required information and resources available
■ None of the platforms will provide estimates of
change in maternal mortality indicators over short
periods of time
■ Compliment not substitute for long term investment
in vital registration systems
Support for this activity has been provided by the U.S.
Agency for International Development through MEASURE
Evaluation. We are grateful to the Carolina Population
Center (R24 HD050924) for general support. This study was
also partially supported by Award Number T32NR008856
from the National Institute of Nursing Research. The content
is solely the responsibility of the authors and does not
necessarily represent the official views of the National
Institute of Nursing Research or the National Institutes of
Health. We are grateful to the Bangladeshi National Institute
for Population Research and Training (NIPORT),
Government of Bangladesh, Mozambican National Institute
of Statistics (INE) and Ministry of Health (MISAU), and the
Zambian Central Statistical Office (CSO) for use of their
MEASURE Evaluation is funded by the U.S. Agency for
International Development through Cooperative Agreement
GHA-A-00-08-00003-00 and is implemented by the
Population Center at the University of North Carolina at
Chapel Hill, in partnership with Futures Group, ICF Macro,
John Snow, Inc., Management Sciences for Health, and
Tulane University. The views expressed in this presentation
do not necessarily reflect the views of USAID or the United
Visit us online at http://www.cpc.unc.edu/measure.