Approaches to improve and adapt maternal mortality estimations in low‐ and middle‐income countries: A scoping review

In the absence of robust vital registration systems, many low‐ and middle‐income countries (LMICs) rely on national surveys or routine surveillance systems to estimate the maternal mortality ratio (MMR). Although the importance of MMR estimates in ending preventable maternal deaths is acknowledged, there is limited research on how different approaches are used and adapted, and how these adaptations function.


| INTRODUC TI ON 1.| Background
Maternal death is defined by the World Health Organization (WHO) as "the annual number of female deaths from any cause related to or aggravated by pregnancy or its management (excluding accidental or incidental causes) during pregnancy and childbirth or within 42 days of termination of pregnancy, irrespective of the duration and site of the pregnancy". 1 In 2017, approximately 295 000 women died during and following pregnancy and childbirth. 2 The majority of these deaths (95%) occurred in low-and lower-middle-income countries. 2stainable Development Goal 3.1.aims to achieve a global maternal mortality ratio (MMR) of fewer than 70 deaths per 100 000 live births by 2030.As of 2017, however, the global MMR was 211 deaths per 100 000, representing only a 38% reduction from the level in 2000.As many countries do not have reliable civil registration and vital statistics (CRVS) systems to report on maternal deaths, the United Nations uses modeled estimates to report country-level MMRs. 3 Countries also rely on special studies including the Demographic Health Surveys (DHS), which are typically conducted every 5 years to generate national MMRs.Although these country-level estimates are useful to monitor national trends and relative performance across countries, they do not produce data with sufficient frequency or granularity to inform internal monitoring or to guide subnational programming-posing a further challenge for achieving Goal 3.1.Commonly used approaches for estimating MMR were extracted from Graham et al. 4 and are shown in Table 1.
In many low-and middle-income countries (LMICs), both empirical and analytical approaches are used to estimate MMR.A 2017 systematic review by Mgawadere et al. 5 described methods to estimate the MMR in LMICs.Methods used were grouped into two categories: those that calculate MMRs using existing data from sources such as civil registration, health facilities, and the census; and those that rely on special studies including population or household surveys, reproductive age mortality studies, and the sisterhood method (both direct and indirect).The strengths and limitations of these methods were described and the authors concluded that reproductive age surveys were a good option in LMICs until CRVS systems were able to routinely produce reliable MMR estimates. 5However, the review by Mgawadere et al. 5 did not cover studies assessing the completeness of these MMR estimates, adaptations to existing methods, or novel estimate approaches.
Recognizing the importance of more explicitly linking information with action, most countries have expanded on their existing maternal death reporting and prevention efforts to implement comprehensive maternal death surveillance and response systems as introduced by WHO in 2013 and later expanded to include perinatal deaths. 6Maternal and perinatal death surveillance and response (MPDSR) is the overall framework adopted by most LMICs to end preventable deaths. 7The framework explicitly links approaches to identify deaths (e.g.maternal death surveillance) with efforts to understand the context of a woman's death to analyze, summarize, and share the information, to take action that addresses the challenges or missed opportunities contributing to the maternal death, and to monitor and refine the response.Identification and reporting of all maternal deaths require a fully scaled and optimized MPDSR system. 8,9As countries vary in the maturity of their CRVS and MPDSR systems and related enabling characteristics, approaches to improve the quality and utility of maternal mortality estimation continue to be developed and piloted. 7

| Objectives
The present review was designed to assess methods to estimate MMRs in LMICs and to understand the rationale for the approach used and the gaps the study implementers hoped to address.

| MATERIAL S AND ME THODS
The reporting of this scoping review is in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses, Extension for Scoping Reviews (PRISMAScR)-statement (see Data S1). 10

| Eligibility criteria
Studies were considered if their primary focus was maternal mortality estimation or measurement.We included studies reviewing, adapting, maternal deaths, provided local ownership, and was expected to reduce implementation costs.However, most studies did not include a cost-effectiveness analysis.Also, entries describing MPDSR implementation, modeled data, or death reviews were considered to be outside the scope of this review.

Conclusion
We only considered studies published after January 1, 2013, as we aimed to expand on the maternal mortality surveillance technical guidance by the WHO. 6Inclusion and exclusion criteria are listed in Table 2.

| Search strategy
The search included controlled terms and free text terms for synonyms of "maternal death", "surveys", and "low-and middle-income countries" (see Data S2 All articles that were rated relevant or that could not be excluded based on abstract review were independently reviewed by two reviewers against the inclusion criteria.Differences in judgment during full-text review were resolved through discussion with co-authors.

| Data extraction
Data were extracted on variables of interest including mortality measurement methods used, context, and results.Review and synthesis of extracted information was completed by all authors.

| RE SULTS
The initial search returned a total of 24 689 references (Figure 1).Africa, four from Asia, two from Central America, and one each from the Caribbean and Oceania (Table 3).Sixteen studies provided an MMR and compared their estimation method with (sub-)national routine estimation efforts (Table 4).Studies were categorized as efforts to assess and/or improve the completeness of MMR estimates, generating timely subnational MMR estimates or incorporating community engagement.

| Addressing completeness of maternal mortality estimates
Understanding the extent to which and the reasons behind whether routine reporting methods capture all maternal deaths, i.e. reporting evaluating the completeness of death reporting, was the focus of five publications included in Table 3. [14][15][16][17][18] This involves understanding the extent to which routine reporting methods capture all maternal deaths, and reasons for this.
5][16] The reasons for underreporting of maternal deaths fell into two general categories: (1) challenges with finding a maternal death, e.g.administrative delays in reporting deaths, missing deaths that occurred outside a health faclity, poor coordination between reporting and recording entities; 15 or (2) difficulties with assigning a cause of death, e.g.misclassification or failure to correctly identify a maternal death given a lack of information on pregnancy status. 16war et al., 14  Anwar et al. 14 implemented a more expansive approach and focused on prospectively identifying and following the outcomes of all pregnant women using multiple information streams in both the community, and public and private sectors. Alhough MMR accuracy and timeliness were improved, the method relies heavily on community workers and its feasibility is unknown in areas with low community worker coverage.
Mola and Kirby 18 compared international modeled MMR estimates to data from a health facility-based information system and a community-based survey in Papua New Guinea.They estimated a national MMR based on the population distribution with women living in urban areas, rural areas with access to care, and rural areas inaccessible to care.They found international modeled MMRs to be much lower, probably because these models included wealth as an explanatory factor for maternal mortality, but did not account for the inequity in wealth distribution and access to health care.Despite only registering facility-based deaths, the authors recommend using local Health Information System data instead of modeled data.
Lougue et al. 17 found the MMR generated by the 2006 population census in Burkina Faso to be acceptable, as cleaning data from duplicated cases, adjustments of the number of live births, and estimation technique provided a similar MMR. 17

| Estimating subnational maternal mortality
[21][22][23][24][25][26][27] Citing wide variation in MMR levels, the need for localized information, and challenges with CRVS data, three papers from Nigeria used the indirect sisterhood method to calculate MMRs at administrative level 1 (i.e.state) or administrative level 2 (i.e.local government areas). 22,23,27In Onoja et al. 22 and Sharma et al., 23   subnational calculations (Table 4). 22,23In Aminati et al. 27 the estimated MMR was lower than the national DHS estimate, which was explained by the authors as the result of regional differences, with the north of Nigeria having higher MMRs.
In Western Guatemala, Perry et al. 19 collected data prospectively using volunteers and community members, such as traditional birth attendants and village health committee members.The subnational MMR was found to be three times the national MMR reported by DHS and twice the regional MMR reported by local authorities.Data quality was increased by having experienced local nurses conducting verbal autopsies of the identified deaths in 2 weeks from their report.
Three papers published from Ethiopia explored empirical approaches for subnational MMR estimation given the need for subnational information to facilitate local health programming and the interest in more frequent estimations. 20,21,24Kea et al. 20,21 used two different methods for calculating regional MMRs: a household survey supplemented by verbal autopsy and the indirect sisterhood method.Both methods provided MMR estimates that were higher than DHS estimates.However, in the same region the indirect sisterhood method identified more deaths than were confirmed by the WHO verbal autopsy tool, which was attributed to the lack of exact information obtained by the indirect sisterhood method. 21Godefay et al. 24 also used a community-based cross-sectional household survey for a straightforward and feasible approach to estimate subnational MMRs, plus acquiring local information important for policy-making.
Moseson et al. 25 implemented a modified direct sisterhood method in Liberia, adding one additional question regarding the location of the maternal death.The addition of this one question allowed for the calculation of subnational MMR estimates, although confidence intervals were wide.

| Engaging the community with maternal mortality estimates
9][30][31][32]  Making data collection tools more user friendly and acceptable by communities was described by Roggeveen et al. 29 and Adomako et al. 30 Roggeveen et al. 29 developed "the Pictorial Sisterhood Method", a modification of the indirect sisterhood method, to estimate maternal mortality in a closed-off Maasai community in Tanzania.The indirect sisterhood method involves four questions about the sisters of a woman in question, regarding the death of women of reproductive age during pregnancy, during childbirth, or in the 6 weeks after pregnancy (Table 1).The adapted method includes a form depicting sisters and their children, so that illiterate community to identify what fraction of maternal deaths identified by the second group was also identified by the first.By doing so, the number of maternal deaths missed by both groups could be estimated using the "capture-recapture method". 28In the Banten Province of Indonesia it was estimated that routine maternal death surveillance identified105 out of 184 (57%) maternal deaths, whereas MADE-IN/ MADE-FOR captured 169/184 (92%). 28In the rural Chackwal district in Pakistan, the MMR generated from the MADE-IN/MADE-FOR method was 309 maternal deaths per 100 000 live births, which was comparable to MMR in rural areas of 319 from the last Pakistan DHS. 31 Alam and Townend 32 implemented a community-based survey directly involving adult women, asking them about nearby vital events in recent years among their neighbors in a rural area of Bangladesh.The reported maternal deaths were compared with the health and demographic surveillance system in the same area.
Although the method was not able to provide absolute values of MMR because this requires the assumption that overreporting and underreporting are similar for live births and deaths, it provided an inexpensive option to monitor MMR relative differences between and change within an area.There were several maternal deaths reported by the adult women that could not be identified in the health and demographic surveillance system.These deaths could be unnoticed maternal deaths by routine surveillance or the result of a different definition of maternal death between the two methods (e.g.

pregnancy-related death versus maternal death).
Table 5 provides an updated version of the previously discussed maternal mortality estimate approaches.Some existing methods may be used in a different way (indirect sisterhood method, modeling) or were adapted (census, direct sisterhood method, RAMOS).
Also, new methods have been added (RAMOS 4 + 2, community surveys with verbal autopsy, Pictorial Sisterhood Method).

TA B L E 5
Measuring and estimating maternal mortality-updated, enhanced or alternative approaches.tors for a population of five million people. 24The authors argue that $0.012 per capita is a small price to provide such valuable information for running an effective healthcare system.Mir et al. 31

| DISCUSS ION
In this scoping review, we presented different estimation practices to arrive at the MMR, their completeness, adaptations, and rationale behind such practices.Methods were implemented and adaptations were made for various reasons, such as increasing data accuracy and improving community awareness.No one method can be stated as best in every situation.The wide variety of methods and their success in specific settings highlight the need for local level and locally managed information.Our review indicated three areas of focus of the included studies: reporting on completeness of estimates, the need for calculating (sub)national MMR estimates, and integration of community engagement.
We reported active surveillance efforts combined with triangulating information from multiple data sources, including the community, public health facilities, and private health facilities, assisted with improvement in the completeness in maternal death reporting.Analyzing multiple data streams and introducing active data collection (which proactively seek to identify cases in contrast to passive efforts, which rely on routine reporting systems to identify cases) may be an option to improve the completeness of MMR estimates and to make use of existing data systems when faced with the alternative of waiting multiple years for a DHS survey or modeled estimate. 18e importance of local level data to help inform and guide Safe Motherhood efforts was frequently stated as the rationale for calculating subnational MMR estimates.22][23][24][25]27 For these countries, including Nigeria and Ethiopia, an understanding of regional maternal mortality variation in combination with an understanding of local conditions, including Ideally, the DHS would be powered to provide these estimates, but the cost to generate subnational estimates with sufficient precision is currently prohibitive. 359][30][31][32] Both the Pictorial Sisterhood Method used in Tanzania and the MADE-IN/ MADE-FOR used in Indonesia and Pakistan employed approaches adapted for community-level use and generated information that may inform community advocacy efforts for maternal health and support blame-free mortality reporting efforts. 28,29,31A separate scoping review has delved further into the topic of community surveillance for maternal death reporting in LMICs and described the added value of this approach to capture maternal deaths and to complement nascent civil registration systems, while noting the unrealized potential of fully engaging community health workers with the death reporting, review and response process. 36While engaging community members for informal and formal reporting and surveillance roles may enhance the completeness of maternal mortality estimates and community ownership, it is also important to consider the sustainability of such efforts and, ultimately, the formal integration of these individuals and their efforts into the mainstream health system. 37fferent estimation methods have their own benefits and area of application, but the biggest limitation to their use is their costs.
As maternal deaths are rare, a large sample of the population is needed to provide MMR estimates with meaningful confidence intervals.Detecting differences in MMR after an intervention is even more cumbersome.The current review is limited in its analysis into the costs of the proposed new methods, because most authors did not include an economic analysis.It is, however, very important to consider costs of the proposed methods, especially in low-resource settings, as $60 000 spent on a survey could have been invested in the healthcare system that it is trying to supplement.Methods that could reduce costs are use of existing data sets (e.g.census, local authorities), predictive models using more prevalent variables, or the use of existing community networks.Future studies should explore the relative cost implications of these approaches, i.e. triangulating multiple, existing data sources versus implementing a national Demographic and Health Survey, relative to the precision of the MMR estimates.This will assist policy-makers to prioritize resources to strengthen mortality reporting and estimations.per 100 000 live births, providing an actual MMR of 11.8. 38Although there are many lessons to be learned from these studies, absolute MMRs are low compared with LMICs.We aimed to identify methods that could be distributed widely in LMICs and, by doing so, provide reliable estimations to reduce the largest burden of maternal deaths. 2 Given that ending preventable maternal deaths is the ultimate goal, MMR estimates help stakeholders to understand where progress has been made and where additional emphasis is needed.
National-level estimates and internationally generated modeled metrics inform global monitoring, but are less informative for locally led efforts to understand and address maternal mortality.Ensuring that MMR estimates derived from routine systems are as complete as possible, engaging the community to facilitate data completeness as well as local ownership, and generating information at a frequency and geographic level relevant for programming, all present pragmatic options to help strengthen estimation approaches.Additionally, to address challenges with the precision of maternal mortality estimates, regular surveillance of the primary causes of maternal mortality (e.g.postpartum hemorrhage, eclampsia, and sepsis) and morbidities may help to address needs in subnational or resourcelimited settings and assist with efforts to end preventable maternal deaths. 34Progress towards Goal 3.1 can only be measured accurately when MMR estimates are precise and reliable.

:
Household surveys with community involvement and RAMOS can be used to increase data validity, improve local awareness of maternal mortality estimates, and reduce costs in LMICs.K E Y W O R D S community participation, data collection, developing countries, maternal death, maternal mortality, quality indicators or comparing maternal mortality estimation methods, or aimed at strengthening their quality.Studies describing trends or associated factors without focusing on the data collection method were excluded.

4 Facility-based 128 --
Abbreviations: CI, confidence interval; CRVS, civil registration vital statistics; DHS, Demographic Health Surveys; HDSS, health and demographic surveillance system; MADE-IN/MADE-FOR, Maternal Death from Informants/Maternal Deaths Follow-on Review; MDSS, maternal death surveillance system; MMR, maternal mortality ratio; RAMOS, reproductive-age mortality surveys; a As no subnational estimates are available, comparison was made with national maternal mortality estimates.
stakeholders could be involved.This increased the local awareness of maternal death and gave community members ownership of information.Similarly, Adomako et al.30 conducted a reproductive-age mortality survey using a simplified questionnaire involving six questions regarding whether a woman was pregnant when she died, the location, and presumed cause of death (RAMOS 4 + 2).The questions were used by community-based volunteers to identify pregnancyrelated deaths in the Bosomtwe district in Ghana.The instrument was considered an effective option for maternal death surveillance in countries lacking complete vital registration.Both studies were conducted with a relatively small sample size and require further evaluation in larger populations.Both Qomariyah et al.28 and Mir et al. 31 used the Maternal Deaths from Informants/Maternal Death Follow on Review (MADE-IN/ MADE-FOR) method in Indonesia and Pakistan, respectively.The MADE-IN/MADE-FOR method consists of two phases for data collection, with phase one involving identification of pregnancyrelated deaths by community informants (MADE-IN) and phase two confirming these deaths through verbal autopsy and collecting additional information.Both studies used two separate groups of community-based informants covering the same geographical area,

3. 4 |
CostsOnly 3 out of 19 studies mentioned reducing MMR estimation costs as a motivation for their proposed MMR estimation method.Godefay et al.33 estimated that their survey in the Ethiopian Tigray region, including training of staff and field work, cost approximately $60 000.They covered 900 000 people and found 54 pregnancyrelated deaths, providing a meaningful MMR and contributing fac- infrastructure and healthcare availability, may help inform and guide resource allocation and relevant programming.The importance of subnational efforts, alongside national and global efforts, to realize change in maternal mortality outcomes is reflected in WHO's 2015 report, Ending Preventable Maternal Mortality (EPMM), indicating that subnational information is critical for programming and resource allocation, as well as accountability.34The EPMM Report includes 10 milestones to monitor progress, including 'Milestone 4: Date for Action', which highlights the need for national and subnational data.
the current review may be its focus on new or adapted estimation methods in LMICs, as it excludes attempts to improve maternal death reporting in high-income countries.Despite their CRVS, maternal deaths in high-income countries are also underreported.As an example, in five Italian regions between 2000 and 2007 only 37% of deaths were included in the official MMR of 4.4

4 2.2 | Information sources
Measuring and estimating maternal mortality-approaches, uses, strengths, and limitations.a a Adapted from Graham et al.
Inclusion and exclusion criteria.criteria Measurement or estimation methods are the papers focus: e.g.sisterhood method, vital registration, communitybased surveillance Methods to estimate maternal mortality are compared, reviewed, adapted, modified, piloted, assessed, or used as surrogates Reliability, discrepancies, or challenges with measurement of maternal mortality are discussed Interventions or approaches to strengthen the quality (e.g.completeness, reliability) of maternal mortality data Conditions or complications associated with maternal mortality (e.g.abortion, cesarean section) Death reviews, verbal, or social autopsy as primary study focus Abbreviations: DHS, Demographic and Health Surveys; MPDSR, Maternal and perinatal death surveillance and response.

Identification of studies via databases and registers Identification Screening Included
Its feasibility is unknown in areas with low community worker coverage.Women living in very remote areas could not be completely covered.
15ouchadi et al.,15and McCaw-Binns et al.16enhanced data collection by including different sources (e.g.data from private facilities, police reports, and local authorities) and demonstrated that routine reporting does not capture all deaths.McCaw-Binns et al.16and Abouchadi et al.15both worked with multiple data sources over fixed time periods to retrospectively identify and classify all maternal deaths.McCaw-Binns et al. 16 started with all reported deaths from vital registration records and incorporated data sources including police reports (e.g.trnsport accidents, suicides) F I G U R E 1 PRISMA flow diagram for search and selection process; *Medline, Embase, APA PsycINFO, CINAHL, IBSS, ERIC, Web of Science, and Scopus. MPnal and perinatal death surveillance and response.Records identified from multiple databases*: (n = 24 689) Records removed before screening: Maternal mortality trends (n = 6) Death reviews (n = 5) Modeled data (n = 1) Full text not in English (n = 3) Full text inaccessible (n = 5) TA B L E 3 Summary table of included studies with study methodology and conclusion (N = 19).Author Country Study methodology ConclusionAddressing completeness of maternal mortality estimates (n = 5)Anwar et al. (2018) 14Pakistan Prospective analysis through enhancement of a surveillance system capturing data from public health facilities, now expanded to include information from private facilities and the community.Data collection from multiple sources and follow up of all pregnancies improves MMR accuracy and timeliness at (sub)district levels.Used census data from two countries and estimated differential MMRs based on residence and educationWhile feasible, the authors concluded that estimate differential MMRs using census data was not recommendedAminati et al. (2013)27 Nigeria Cross-sectional study using the indirect sisterhood method to estimate MMR in Suleja of Niger state.Interest in capturing community-level data given low facility birth rate. (Cntinues) and Ministry of Health records (maternal mortality surveillance) to refine the MMR estimate.Abouchadi et al. 15 identified deaths of women of reproductive age using civil and health registration offices, local authorities, and data from public and private hospitals.Qualitative information was obtained through household interviews to identify maternal deaths among deceased women of reproductive age.Both studies show that data from surveillance systems were incomplete and could be improved by implementing registries from local authorities instead of the larger health offices, and by questioning communities.
28gaging the community with maternal mortality estimates (n = 5) Qomariyah et al.28Indonesia MADE-IN/MADE-FOR approach using two groups of community informants (heads of neighborhood units, health volunteers) to identify maternal deaths in Banten Province.Maternal deaths were confirmed by verbal autopsy.Deaths were compared with routine community-based reporting (district health offices).Involvement of existing community networks improves data completeness and provide the possibility for subnational MMR estimates.Intermittent implementation of MADE-IN/MADE-FOR may be used to evaluate completeness of routine maternal death reporting systems.No MMRs were provided.
the calculated MMRs were substantially higher than the most recent national DHS estimate, which further justified the rationale for Abbreviations: CI, confidence intervals; CRVS, civil registration and vital statistics; DHS, Demographic Health Survey; MADE-IN/MADE-FOR, Maternal Death from Informants/Maternal Deaths Follow-on Review; MMR, maternal mortality ratio; RAMOS, reproductive-age mortality surveys.
Several existing methods were applied for community use, including the sisterhood method by Roggeveen et al., 29 the reproductive-age mortality survey by Adomako et al., 30 and the MADE-IN/MADE-FOR approach by Mir et al. 31 and