British Medical Bulletin, 2017, 121:121–134

doi: 10.1093/bmb/ldw056
Advance Access Publication Date: 19 January 2017

Invited Review

Measuring maternal mortality: a systematic

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

review of methods used to obtain estimates
of the maternal mortality ratio (MMR)
in low- and middle-income countries
Florence Mgawadere*, Terry Kana, and Nynke van den Broek
Centre for Maternal and Newborn Health, Liverpool School of Tropical Medicine, Pembroke Place,
Liverpool L3 5QA, UK
*Correspondence address: Florence Mgawadere, Centre for Maternal and Newborn Health, Liverpool School of Tropical
Medicine, Pembroke Place, Liverpool L3 5QA, UK. E-mail: [email protected]
Editorial Decision 19 December 2016; Accepted 5 January 2017

Abstract
Background: The new global target for maternal mortality ratio (MMR) is a
ratio below 70 maternal deaths per 100 000 live births by 2030. We under-
took a systematic review of methods used to measure MMR in low- and
middle-income countries.
Sources of data: Systematic review of the literature; 59 studies included.
Areas of agreement: Civil registration (5 studies), census (5) and surveys
(16), Reproductive Age Mortality Studies (RAMOS) (4) and the sisterhood
methods (11) have been used to measure MMR in a variety of settings.
Areas of controversy: Middle-income countries have used civil registration
data for estimating MMR but it has been a challenge to obtain reliable data
from low-income countries with many only using health facility data (18
studies).
Growing points and areas for further research: Based on the strengths and
feasibility of application, RAMOS may provide reliable and contemporan-
eous estimates of MMR while civil registration systems are being intro-
duced. It will be important to build capacity for this and ensure
implementation research to understand what works where and how.
Key words: measuring, maternal mortality ratio, RAMOS, low- and middle-income countries

© The Author 2017. Published by Oxford University Press.

This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (http://creativecommons.org/
licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited.
For commercial re-use, please contact [email protected]
122 F. Mgawadere et al., 2017, Vol. 121

Introduction method, lack of in-country capacity to use the meth-

od and requirement for large sample sizes to be able
Reducing maternal mortality is one of the priority
to estimate MMR with reasonable accuracy.
goals on the international agenda—the new global
Although some of these methods have been used in
target is to reduce the maternal mortality ratio
a number of developing countries, there is a lack of
(MMR) to <70 maternal deaths per 100 000 live
knowledge and guidance regarding which method(s)
births and country should reduce their MMR by at
are the be most appropriate and feasible to use in
least two-thirds from the 2010 baseline and no
which settings (e.g. large or small population,
country should have an MMR higher than 140

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

national or sub-national application and type of
deaths per 100 000 live births by 2030.1,2 A cross-
data required to estimate MMR).
cutting priority for the post-2015 agenda is to move
We, therefore, conducted a systematic review of
toward counting every birth, maternal death and
the literature to identify which methods have been
perinatal death through the establishment of effect-
used to estimate MMR and reviewed their use and
ive national registration and vital statistics systems
applicability to low- and middle-income countries.
in every country, as stated within the recommenda-
The specific research questions included: what are
tions of the Commission for Information and
the type of data and data sources required, what
Accountability.3
are the strengths and weaknesses of each method;
Reliable data are needed so that adequate
and which method(s) would be useful and applic-
resources can be allocated to maternal health pro-
able in low- and middle-income settings and able to
grammes for countries (or regions in countries) that
provide reasonably accurate and contemporaneous
are not yet accelerating the annual reduction in
data.
maternal deaths. These data are also needed to
monitor progress toward the targets set for the new
Sustainable Development Goals (SDG). Assessing Methods
progress has been a challenge because <40% of
We used the following databases SCOPUS, PUBMED
countries currently have complete civil registration
and Institute for Scientific Information (ISI) and
(CR) systems in place or other methods to provide
MEDLINE to search for studies that measured MMR
accurate and contemporaneous MMR data.
in low- and middle-income countries. Publications of
Similarly, although Maternal Death Surveillance
organizations and programmes such as the United
and Review is promoted and being implemented in
Nations Population Fund (UNFPA), United Nations
many settings, attribution and reporting of cause of
Children’s Fund (UNICEF), World Bank, WHO and
maternal death is not yet systematically in place.4,5
the Initiative for Maternal Mortality Programme
Only 2 of the 49 least developed countries have
Assessment (IMMPACT) were included. Internet
>50% coverage with regard to death registration.1
searches using the Google search engine were con-
The World Health Organization (WHO) pub-
ducted to identify relevant literature not published in
lishes global estimates of MMR based on United
peer-reviewed journals and the references of all identi-
Nations statistical models, including estimates for
fied, relevant papers were hand-searched.
countries without reliable data.6 Most of these
estimates are subject to greater or lesser degrees
of uncertainty and this is a recognized limitation. Search terms used
There are, in addition to modelling, a variety of Medical Subject Headings (MeSH) were searched to
methods available to measure MMR including identify all relevant terms used to describe maternal
via censuses, household surveys, Reproductive mortality and measuring. Boolean operators such as
Age Mortality Studies (RAMOS) and using the ‘OR’ were used to join keywords and MeSH terms
Sisterhood methods. Each method has strengths and defining the same concepts and different concepts
weaknesses. This may include cost of application of were searched with ‘AND’ to arrive at the final
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 123

result. We used the search terms ‘maternal mortality discussion with a third researcher. A summary table
OR maternal death OR pregnancy death OR was developed and agreed by all authors before
motherhood death OR women deaths’ in combin- full-text review was conducted and all included
ation with ‘measure OR estimate’ OR ‘estimation’. studies were then summarized. (Supplementary
These were then combined using the Boolean oper- Table S1: Summary Table of included studies)
ator ‘AND’ with the following search terms: civil
registration data, Census, Surveys, health facility
data, sisterhood methods, RAMOS and low- and
Results

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

middle-income countries. Star truncation (*) was
used where multiple endings of terms were possible. In total, 60 studies meet the inclusion criteria
(Fig. 1). Six methods by which MMR or relevant
data are obtained in low- and middle-income coun-
Inclusion and exclusion criteria
tries were identified. These include use of existing
We included papers published between 2000 and opportunities which include (i) CR data (5 studies),
October 2016 from low- and middle-income coun- (ii) health facility data (18 studies), (iii) population
tries as defined by the World Bank income categor- census (5 studies), use of special studies which
ization.7 This time period was selected as many included (iv) population or household surveys (16
countries undertook to assess the MMR to evaluate
the burden of maternal mortality and effect of
implementation of interventions to achieve MDG5
at this time.8
Studies were included for estimates of MMR
obtained at either national or sub-national level
regardless of method used. We excluded studies
assessing impact of one or more interventions on
MMR; demographic health surveys (DHS) as they
are included in the direct sisterhood methods and
global estimates by WHO, UNICEF, UNFPA and
the World Bank (published by the WHO and not
countries that use statistical models that may have
errors, use unreliable data sources and, in some
cases, countries do not use them). We also excluded
reviews, posters, editorials and discussion papers,
which did not include methodologies and estimates
of MMR. DHS were excluded as these use the dir-
ect sisterhood method which is already included in
the review. Global estimates were excluded when
they employed statistical modelling.

Data extraction
Two reviewers independently screened all titles and
abstracts. When the information provided by title
and abstract was insufficient to decide on inclusion
or exclusion, full-text versions were retrieved and
evaluated. All included papers were reviewed in Fig. 1 PRISMA diagram for identification of included
full. Any discrepancies were resolved through studies.
124 F. Mgawadere et al., 2017, Vol. 121

studies), (v) direct and indirect sisterhood method on the death certificates for studies conducted in
(11 studies), and (vi) RAMOS Studies (4 studies). China, the Dominican Republic, Brazil and Egypt.
The authors note that this resulted in misclassifica-
tion and possibly an underestimation of the number
CR and vital statistics data of maternal deaths.12–14,16 Deaths among women
CR is defined as the continuous, permanent, com- living in villages accessible only by foot were not
pulsory and universal recording of the occurrence registered in the study in the Dominican Republic.14
and characteristics of vital events pertaining to the

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

population as provided through decree or regula- Health facility surveys
tion in accordance with the legal requirements of a
Health facility data remain the main, routine source
country.9 The data retrieved from CR systems are
of data on MM for many developing countries.
referred to as vital registration (VR) data. Complete
Currently, health facility data are not used by aca-
coverage, accuracy and timeliness of CR are essen-
demics and by agencies for compiling global mor-
tial for quality vital statistics and are the ideal data
tality estimates, but they are widely used in many
to count maternal deaths. CR is carried out pri-
countries as they are locally generated and continu-
marily for the purpose of establishing the legal
ously available. Data sources include routinely
documents provided by the law. Additionally, regis-
reported records in health facilities or sentinel sites,
tration of births and deaths generates information
reports from healthcare providers and health facility
that has substantial policy utility, especially when
surveys.
the age of the mother giving birth, age and sex of
Eighteen papers reviewed used health facility
the decedent and underlying cause of death are cor-
data to estimate MMR.17–34 Most studies were con-
rectly specified.10 Ideally, CR systems with high
ducted in low- and middle-income African countries
coverage and good attribution of cause of death
such as Nigeria, Cameroon, Malawi and Zambia.
provide accurate data on the level of MM and the
However, middle-income countries such as India,
causes of maternal deaths. The drawback, however,
Pakistan and Turkey also estimated MMR using
relates primarily to the availability, reliability, com-
health facility data.18,23,27 It was noted that 15 stud-
pleteness and coverage of the CR data.11 The num-
ies were conducted in tertiary or teaching hospitals,
ber of maternal deaths and number of live births
which are expected to have a significant proportion
recorded are used to calculate MMR (number of
of high-risk obstetric cases although this proportion
MD per 100 000 live births).
was not reported.17,19–23,27–29,31–34 Maternal deaths
Five papers reported using CR data to estimate
were identified from maternity ward records in 16
maternal mortality. These studies were conducted in
out of 18 facilities. Only two studies identified cases
China, the Dominican Republic, Brazil, Egypt and
from other wards including the female or gynaecol-
Guatemala, all middle-income countries.12–16 There
ogy ward and from operating theatre registers.21,31
were no studies from low-income countries using
Case notes for women who had died were noted to
this method.12–16 In Guatemala, VR data were sup-
have been missing in some facilities and there were
plemented with additional information from med-
considerable inaccuracies in routine registers noted
ical charts and from public healthcare centres which
in most retrospective studies.20,25
improved the quality of data obtained. The study
detected three times the number of maternal deaths
compared to using the civil registry data only.15 Population census
However, to establish whether the death of a Greater interest has been shown in using data from
woman of reproductive age (WRA) is a maternal population censuses to measure maternal mortality.
death, information on pregnancy status at time of A national census, with the addition of a small
death is required. It was noted that information on number of additional questions, can be used to
pregnancy status was often either missing or unclear obtain estimates of maternal mortality.35 This is a
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 125

result of the endorsement of this method by the per 100 000 (95% CI: 295, 411) in Orangey in
United Nations Principles and Recommendations Burkina Faso.41 Similarly, there was a greater
for Population and Housing Censuses.36 The UN number of reported pregnancy-related deaths
principles recommend two follow-up questions in using census data than obtained via sample sur-
cases where the household being interviewed veys in Latin America.
reports a death during the past 12 months. After
ascertaining the name, age and sex of the deceased
Population or household surveys
person and date of death, the interviewer should
Population or household surveys are one of the

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

enquire:
most important data capturing methods for mater-
(1) Was the death due to an accident, violence,
nal deaths in settings where routine information
homicide or suicide?
systems are weak or non-existent. These surveys
(2) If the deceased was a woman aged 15–49, did
are administered at the household level to collect
the death occur while she was pregnant, during
information about maternal deaths. Names and
childbirth or during the six weeks after the end
residences are cross-checked to avoid double
of pregnancy?
counting. Sometimes, they are complemented with
As a result, in the 1990s several countries verbal autopsies where the family members or
included questions intended to ascertain if any other people with knowledge about the death
WRA had died during pregnancy or within a could be asked to describe the situation surround-
defined period postpartum, usually 6 weeks. In ing the death relatives. The WHO has devised a
principle, a census allows the identification of standard verbal autopsy tool to collect informa-
deaths in a household in a relatively short refer- tion on signs, symptoms, medical history and cir-
ence period (1–2 years) and thereby provides esti- cumstances preceding death,42 which countries
mates of recent maternal mortality. can adapt according to their situation. In both
Population census data were used to estimate population or household surveys and verbal aut-
maternal mortality in five studies.37–41 Questions opsies, names and residences are cross-checked to
regarding the time of circumstance of death avoid double counting. These methods are only
among WRA who died during pregnancy, labour appropriate for settings in which the sampling
and in the postpartum period (usually 6 weeks unit is a complete village and the geographical
after delivery) were included during census data scope of the study is quite limited. Surveys, how-
collection. The studies were conducted in Latin ever, require a relatively large sample size to obtain
America (Honduras, Nicaragua and Paraguay), statistically significant findings for occurrences that
South Africa, Burkina Faso, Honduras (only) and are relatively rare such as maternal deaths.35 Sixteen
Indonesia. A study conducted in Burkina Faso studies included in this review estimated the number
used this approach and obtained an estimate of of maternal deaths using population and/or house-
the MMR, the results of which were similar to a hold surveys.38,43–57 Out of the 16 studies, only 5
previous study which had used the direct sister- were conducted in Africa (Ethiopia, Kenya, Malawi,
hood method.41 In Latin America, there was a Senegal and Tanzania).44,52,56,57 Six were conducted
greater number of reported pregnancy-related prospectively (Colombia, Sri Lanka, Ethiopia,
deaths in census data when compared with the Indonesia, Jamaica and Pakistan).38,43,48,50,51,53 For
number reported during a household survey which cultural reasons, family members and birth atten-
was conducted at the same time.39 Similarly, in dants in Cambodia were reported to have felt
the Republic of South Africa, an increase in mater- ashamed of deaths that had occurred and, there-
nal death was observed.40 Narrow confidence fore, did not report all deaths. There were sam-
intervals were obtained: MMR: 519 per 100 000 pling problems in some of the studies and very
(95% CI: 454, 584) in Damage and MMR: 353 wide confidence intervals were obtained for the
126 F. Mgawadere et al., 2017, Vol. 121

MMR estimates.55,57 For example, in India, only a non-maternal deaths. However, the extent of the com-
small area was covered because it was considered pensation is unknown. Finally, estimates obtained
too expensive to conduct a household survey cov- using the indirect sisterhood method relate to the
ering a bigger, geographical area. previous 10–12 years and are, therefore, not contem-
poraneous and cannot be used for evaluating the
effectiveness or impact of an intervention programme.
Direct and indirect sisterhood methods Rutenburg and Sullivan proposed the direct sis-
In sisterhood surveys, adult respondents report on terhood method,59 which has been widely used in

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

the aggregate numbers of surviving sisters and of DHS programmes. This is a variant of the indirect
sisters who have died.58 There are two types of sisterhood method based on a detailed sibling
sisterhood methods, the indirect and the direct history obtained from each respondent. The four
method. questions listed above for the original indirect sis-
The original (indirect) sisterhood method was terhood method are expanded to 11 questions. In
developed in the late 1980s by Graham et al.58 In addition, the formulation of the original questions
the indirect sisterhood method, adult respondents was altered. For example, the fourth question on
are asked four questions pertaining to the survival timing of death in relation to pregnancy, childbirth
(or not) of all their adult sisters born to the same and the postpartum period was changed to include
mother. By enquiring about female siblings in a three separate questions (Box 2).
high fertility setting, one effectively expands the The data requirements for the direct sisterhood
sample size with very little additional cost. The method are considerably more demanding than those
method also reduces the need for large sample sizes for the indirect approach. In the direct approach, a
because there may be more than one respondent per respondent is asked to provide the birth history of
household and more than one sister per respondent. her mother, including the current age of all living sib-
The questions for which responses are required in lings and the age at death and years since death for
the indirect sisterhood method are listed in Box 1. all deceased siblings. These data allow deaths and
As this method identifies any death that occurs births to be placed in calendar time and, therefore,
during pregnancy, childbirth or the postpartum permit the calculation of sex and age-specific death
period; the indirect sisterhood approach identifies rates for reference periods.59 Unlike the indirect sis-
pregnancy-related deaths rather than true maternal terhood method, the direct sisterhood method targets
deaths. Overestimation of maternal mortality due a more limited reference period for sister deaths:
to the inclusion of deaths that are coincidental the previous 0–6 years compared with the previous
and/or non-maternal deaths is likely. Conversely, 10–12 years for the indirect sisterhood method.
abortion-related maternal deaths are often not cap- Point estimates for maternal mortality are obtain-
tured. It has been suggested that the omission of able. The approach also allows for the calculation of
induced abortions a compensate for the inclusion of rates/ratios for the reference period of interest and

Box 1 Indirect sisterhood

(1) How many sisters (born to the same mother) have you ever had who reached age 15 (who were
ever-married) including those who are now dead?
(2) How many of these ever-married sisters are alive now?
(3) How many of these are dead?
(4) How many of these dead sisters died while they were pregnant, or during childbirth, or during the
six weeks after the end of the pregnancy?
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 127

Box 2 Direct sisterhood

(1) How many children did your mother give birth to?
(2) How many of these births did your mother have before you were born?
(3) What was the name given to your oldest (next oldest) brother or sister?
(4) Is (NAME) male or female?
(5) Is (NAME) still alive?
(6) How old is (NAME)?

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

(7) In what year did (NAME) die? OR how many years ago did (NAME) die?
(8) How old was (NAME) when she died?
For dead sisters only:
(9) Was (NAME) pregnant when she died?
(10) Did (NAME) die during childbirth?
(11) Did (NAME) die within two months after the end of pregnancy or

Source: World Health Organization (1997). The Sisterhood method for estimating maternal mortality:
guidance potential users. Available on http://apps.who.int/iris/bitstream/10665/64007/1/WHO_RHT_97.28.
pdf. Accessed on November 21, 2016.

monitor trends. The direct sisterhood method is cur- or incidental causes (i.e. not maternal deaths).
rently used during DHS. This method requires larger Cause of death is not determined and data collected
sample sizes than the indirect method. It also requires refers to the previous 10–12 years.
an additional 8–10 min per interview on average and
additional training and supervision in the field.
Both methods measure the ICD-10 concept of Reproductive age mortality studies
pregnancy-related mortality rather than maternal A RAMOS has been identified as a relatively robust
mortality on the grounds that respondents would method, which uses both active and passive data
not be easily able to distinguish between maternal collection methods to estimate the MMR in coun-
and pregnancy-related deaths.39 tries without VR data and are often considered to
We did not find any peer-reviewed studies that be the gold standard.35 The approach involves
use the direct sisterhood method apart from the retrospective or prospective identification and inves-
DHS. An analysis of the quality of maternal health tigating the causes of all deaths of WRA in a
indicators for DHS studies is not part of this defined area/population by using multiple sources
review and has been described elsewhere.60 Eleven of data such as existing records (CR and health
included studies used the indirect sisterhood me- facility data), census, surveys and surveillance.
thod to estimate MMR. Ten studies were con- RAMOS are conducted in two phases. The first
ducted in Africa (Liberia, Nigeria (2 studies), Mali, phase, involves identification of all deaths among
Tanzania (3 studies), Swaziland, Uganda and WRA in a population. In the second phase, all
Ghana).61–69 One study was conducted in India.70 deaths are investigated (using verbal autopsy, health
In Ghana and Uganda, the MMR estimates iden- facility reports or medical record reviews death cer-
tified through the indirect sisterhood method were tificates with medical cause and interview with
higher than those obtained as national estimates household members and relatives) to ascertain if
(modelling, UN global estimates) conducted at the there are pregnancy-related or maternal deaths.71
same time.68,69 All studies registered pregnancy- Four studies conducted in Malawi, Sudan,
related deaths and include death due to accidental Jordan and Ghana estimated MMR using the
128 F. Mgawadere et al., 2017, Vol. 121

RAMOS method.72–75 Three studies were prospect- was recorded and may not be reported as a mater-
ive and one study retrospective.73 A list of all deaths nal death even if the woman was pregnant. Even in
of WRA was collected using data collated at health countries where routine registration of deaths is in
facilities (e.g. admission and discharge books, death place, maternal deaths may be unidentified due to
certificate books, death registers and mortuary log- misclassification of ICD-10 coding and identifica-
books) and individual case notes when necessary, tion of the true numbers of maternal deaths may
available census data or any other relevant data e.g. require special investigations into the causes of
the number of births from the most recent DHS and deaths.59,77 This review shows that even in high-

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

from the Health Management Information System middle income countries such mechanisms are only
(HMIS). Deaths that occurred in the community now in process of being fully developed. A variety
were identified by local key informants, traditional of other methods are used in low- and middle-
birth attendants and community workers. A study income settings. Very often the only contemporan-
in Malawi used the existing health staff (nurses, eous data available are health facility-based MMR
doctors, medical assistants and community health estimates which do not apply to the whole popula-
workers known as health surveillance assistants) at tion or estimates obtained via the sisterhood meth-
both health facility and community level to iden- od which are not contemporaneous and report
tify and report all deaths of WRA. One study in pregnancy-related rather than maternal deaths
Pakistan included interviews with graveyard care- per se.
takers as an additional source of data. In most
studies, verbal autopsies where family members or
other people with knowledge about the death were Birth and death registration
interviewed to describe the situation surrounding For birth and death registration systems to provide
the death. data on the number of maternal deaths among all
All RAMOS studies highlight that this method deaths of WRA, it is important that pregnancy sta-
identifies more maternal deaths than obtained via tus is known. Although a tick box has been
any one of the existing reporting mechanisms included on the death notification from, underre-
alone (e.g. HMIS and facility death reports). porting of the number of maternal deaths and mis-
Underreporting of maternal deaths (by 44 and reporting (misclassification of death of a WRA as a
43%) documented via survey and CR was maternal death or not) has been identified as a
reported in Ghana and Malawi, respectively.72,75 problem of CR data.78,79 In this review, informa-
In Sudan and Jordan, the RAMOS study was con- tion on pregnancy status was often either missing
ducted at state level, while in Malawi and Ghana or unclear in the identified deaths.12–16 Although
the studies were conducted at district and city countries such as Sweden, the Netherlands, the UK
level, respectively.73,74 In Malawi, verbal autopsy and USA, which have documented reduction in
was only done for deaths that were identified as MMR over several decades, have relied on adequate
maternal deaths. Maternal deaths were identified CR systems, misclassification and underreporting
using the ICD-10 version 10 definition of MD.76 exist.4,79 Revision of the death certificate to include
information on pregnancy status improves the qual-
ity of data and helps to reduce misclassification of
Discussion maternal deaths. CR data can be compared with
Accurate levels of maternal mortality are difficult to data obtained via other systems specific to the
measure in a population for it is challenging to evaluation or audit of maternal deaths; countries
identify maternal deaths precisely, particularly in such as the UK and South Africa have used the
settings where routine recording of deaths is not Confidential Enquiry into Maternal Deaths
complete within CR systems.6 The woman’s preg- (CEMD) to ensure any death missed by the CR sys-
nancy status is usually missed even if such a death tem are captured.2,80,81,82
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 129

Health facility data identify pregnancy-related deaths (not maternal

deaths). Early pregnancy deaths may remain under-
Valuable information can be obtained when mater-
reported if pregnancy status was not known and
nal deaths that occur in a health facility are
maternal mortality can be over-estimated where
reviewed specifically to identify where the health
death was incidental and not due to the pregnancy.
system needs to improve.22,83 However, in low- and
This is illustrated in a study conducted in the
middle-income countries unless >95% of women
Republic of South Africa where an increase in
give birth in a health facility (as opposed to at
maternal deaths was identified when census data
home), findings from hospital-based studies cannot

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

were used as compared with a previous MMR esti-
be generalized to the entire population. However,
mate which was obtained using a survey method.40
the majority of the studies from developing country
Despite these limitations, census data still offer the
settings reported hospital-based MMR. These only
opportunity to measure pregnancy-related mortality
apply to the hospital itself and reflect the type of
as a proxy for maternal mortality in countries with
services provided; large referral hospitals with a
poor or no death registration systems in place.
large proportion of referred and complicated cases
When specific, planned surveys are used, captur-
(rather than uncomplicated deliveries) can expect
ing deaths and births is more complete than with
the MMR to be higher than for lower level hospi-
routinely gathered statistics. However, survey meth-
tals (from where patients who are severely ill will be
ods require prohibitively large sample sizes to
referred out). Thus, facility-based MMR can only
obtain statistically significant findings. Such surveys
be used at the facility level to monitor trends over
could, however, be used to estimate MMR in
time and if the proportion of women with poten-
resource-limited countries in smaller subsets of
tially life threatening obstetric complications is
populations where the other data sources are not
taken into consideration. It is also crucial that all
available and/or RAMOS cannot be conducted.
deaths of WRA are identified and an assessment is
made to classify them as maternal deaths or not.
Unless a country has a healthcare system like Sisterhood methods
Saudi Arabia, where almost all maternal deaths
The sisterhood method is cost-effective and easier
take place in the hospital or where all women are
to perform than prospective population-based
brought into hospital soon after death outside the
methods. Specifically, with the indirect sisterhood
facility, hospital-based data cannot be used to pro-
method, the number of households that need to be
vide accurate estimates of MMR for the
visited in order to obtain information on large num-
population.84
bers of women who have reached reproductive age
is relatively small.58 Given that questions are asked
Census and population or household about the deaths of adult sisters, both methods
surveys actually measure pregnancy-related deaths rather
The United Nations recommend the use of a popula- than maternal deaths, on the grounds that respon-
tion censuses for estimating MMR, without consid- dents (sisters) would not easily be able to distin-
ering this a substitute for VR.85 Use of census data guish between maternal and non-maternal deaths
to calculate MMR is cost-effective as the data can and/or usually unable to assign cause of death with
be obtained as part of an already agreed country certainty. Both methods provide estimates of mater-
census. Census data should provide a complete pic- nal mortality in orders of magnitude rather than
ture of the whole population and results in an esti- precise ratios since both can have wide margins of
mate with relatively normal confidence indicators error (wide confidence intervals). Neither method
due to the large sample sizes. However, a census is provides a current estimate for the year of the sur-
usually only conducted every 10 years and cannot vey. For these reasons, sisterhood studies cannot be
be used for routine monitoring. Furthermore, they used to monitor changes in maternal mortality or to
130 F. Mgawadere et al., 2017, Vol. 121

assess the impact of safe motherhood programmes also assist in monitoring any trends in MMR over
in the short term. The sisterhood method has been time. Many low- and middle-income countries are
recommended by the WHO for countries without in the process of introducing CR systems for births
other reliable source of data and this method is fre- and deaths. For countries without reliable systems
quently used as part of the 5-year DHS in low- and in place, a RAMOS can be an effective method that
middle-income countries. can be used to obtain recent data and provides bet-
ter estimates of MMR.
Reproductive Age Mortality Studies A RAMOS approach can also help illustrate

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

(RAMOS) what is needed to support the introduction of a full-
scale Maternal Death Surveillance and Response
In the absence of a CR system with/without add- (MDSR) process. The MDSR builds on the princi-
itional data collection mechanisms such as a ples of public health surveillance and response by
CEMD, the RAMOS approach probably provides collecting accurate information on cause of mater-
the most complete and contemporaneous estima- nal deaths so lessons can be learnt and actions
tion of MMR because information regarding the taken to prevent similar deaths in the future and to
number of maternal deaths is obtained from a var- improve quality of care.
iety of sources and each death among WRA is eval-
uated to assess whether the death is a maternal
death or not. However, the RAMOS approach is
difficult in the absence of a reasonably complete ini- Author Biographies
tial list of deaths. Inadequate identification of all Florence Mgawadere PhD Florence, a nurse-midwife
deaths among WRA results in an underestimation from Malawi with over 12 years of international
of maternal mortality levels. For example, Surinam experience in teaching, research and technical assist-
had a reliable registration system for deaths which ance. Florence’s current research interests include
made identification of deaths of WRA relatively quality of care, application of the ICD-MM cause
easy.86 Similarly, during the prospective RAMOS in classification for maternal deaths, maternal death
Pakistan, good population-based systems were in audit or review and strengthening of Maternal Death
place for tracking deaths.48 This meant that the Surveillance and Response (MDSR). Florence has
number of maternal death among deaths of WRA carried out research on maternal health, including at
could be assessed. RAMOS studies can be expen- the community and facility levels and programme
sive and time consuming when conducted on a evaluations. Florence works as a Senior Research
larger scale.14 A RAMOS may, therefore, be con- Associate at the Centre for Maternal and Newborn
sidered to provide accurate MMR data for a sub- Health at the Liverpool School of Tropical Medicine.
national population. Terry Kana MSc Terry, an experienced teacher in
midwifery and public health from community to
postgraduate level, has worked in a wide variety of
Conclusion hospital and community settings with over 10 years
To end preventable maternal deaths, it is crucial of international programme management and re-
that countries develop systems and processes to search experience. Her current research interests
ensure the ability to count every maternal death include assessment of the effectiveness of competency
and identify the cause of death and contributing based ‘skills and drills’ training in emergency obstet-
conditions. This will help identify where and how ric care, the role, scope of work and workload of
the availability or coverage as well as quality of nurse-midwives working in low- and middle-income
care need to be improved. Ideally, MMR estimates settings. Terry works as a Senior Research Associate
should be obtained from CR data, which provide at the Centre for Maternal and Newborn Health at
both numerator and denominator data. This would the Liverpool School of Tropical Medicine.
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 131

Nynke van den Broek MBBS, DTMH, PhD, 5. Ameh CA, Adegoke A, Pattinson R, et al. Using the
FRCOG Professor van den Broek is a recognized new ICD-MM classification system for attribution of
international expert in global maternal and newborn cause of maternal death-a pilot study. Brit J Obstet
Gynaecol 2014;121:32–40.
health who established and leads the Centre for
6. WHO, UNICEF, UNFPA, World Bank, United Nations
Maternal and Newborn Health (CMNH), one of the
Population Division. Trends in Maternal Mortality 1990
largest academic groups in Europe with an inter- to 2015: Estimates by the WHO, UNICEF, UNFPA,
nationally recognized portfolio of work that incorpo- The World Bank and the United Nations Population
rates priority interventions for reducing maternal Division. Geneva: World Health Organization, 2015.

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

and newborn mortality and morbidity. Four key the- Available from: http://www.who.int/reproductivehealth/
matic areas include skilled birth attendance, emer- publications/monitoring/maternal-mortality-2015/en/
7. World Bank. New Country Classifications.
gency obstetric care, quality of care and maternal
Washington, DC: World Bank, 2015. Available from:
morbidity. Professor van den Broek has designed
http://data.worldbank.org/news/new-country-classifi
and conducted large population-based randomized cations-2015
controlled trials of single interventions for improved 8. United Nations. United Nations Millennium
maternal and newborn outcomes. She has used this Declaration, vol. 37, UN Chronicle, 2000;38.
experience to develop complex packages of interven- 9. Phillips DE, Abou Zahr C, Lopez AD, et al. Counting
tions and to design and conduct operational research births and deaths: Are well functioning civil registration
programmes in multi-country settings. and vital statistics systems associated with better health
outcomes? Lancet 2015;386:1386–94.
10. Setel PW, MacFarlane SB, Szreter S, et al. A scandal of
Supplementary material invisibility: making everyone count by counting every-
one. Lancet 2007;370:1569–77.
Supplementary material is available at BRIMED
11. Graham WJ, Ahmed S, Stanton C, et al. Measuring
Journal online.
maternal mortality: an overview of opportunities and
options for developing countries. BMC Med 2008;6:12.
12. Zhu L, Qin M, Du L, et al. Comparison of maternal
Conflict of interest statement mortality between migrating population and permanent
The authors have no potential conflicts of interest. residents in Shanghai, China, 1996–2005. Brit J Obstet
Gynaecol 2009;116:401–7.
13. Alves SV. Maternal mortality in Pernambuco, Brazil:
References what has changed in ten years? Reprod Health Matters
1. WHO. Strategies towards Ending Preventable Maternal 2007;15:134–44.
Mortality (EPMM). Geneva: World Health Organization, 14. Westoff W, Calcagno E, McDermott R, et al.
2015. Available from: http://who.int/reproductivehealth/ Estimating maternal mortality in Monseñor Nouel
topics/maternal_perinatal/epmm/en/ Province, Dominican Republic. Matern Child Health J
2. United Nations. Sustainable Development Goals. 2009;13:707–714.
GenevaWorld Health Organization, 2015. Available 15. Kestler E, Ramírez L. Pregnancy-related mortality in
from: http://www.un.org/sustainabledevelopment/susta Guatemala, 1993–1996. Rev Panam Salud Publ 2000;
inable-development-goals/ 7:41–6.
3. WHO. Commission on Information and Accountability 16. Hamza S. The maternal mortality. Egyptian Natl
for Women’s and Children’s Health. Keeping Promises, Matern Mortality Study 2005;2:306.
Measuring Results. Geneva: World Health Organization, 17. Agan TU, Archibong EI, Ekabua JE. Trends in maternal
2011 Available from: http://www.who.int/topics/millenni mortality at the University of Calabar Teaching
um_development_goals/accountability_commission/ Hospital, Nigeria, 1999–2009. Int J Womens Health
Commission_Report_advance_copy.pdf?ua=1 2010;2:249–54.
4. WHO. Maternal Death Surveillance and Response. 18. Aggarwal AR, Pandey A, Kar R. Estimates of maternal
Geneva: World Health Organization, 2016 Available mortality ratio and the associated medical causes in
from: http://www.who.int/maternal_child_adolescent/ Orissa and Rajasthan states - A cross sectional study.
epidemiology/maternal-death-surveillance/en/ Ind J Comm Health 2015;27:18–24.
132 F. Mgawadere et al., 2017, Vol. 121

19. Omo-Aghoja LO, Aisien OA, Akuse JT, et al. Maternal 35. WHO. WHO Guidance for Measuring Maternal
mortality and emergency obstetric care in Benin City, Mortality from Census Data. New York: World Health
South-south Nigeria. J Chin Clin Med 2010;5:164–7. Organization, 2013. Available from: http://apps.who.
20. Bergsjø P, Vangen S, Lie RT, et al. Recording of mater- int/iris/bitstream/10665/87982/1/9789241506113_eng.
nal deaths in an East African university hospital. Acta pdf
Obstet Gynecol Scand 2010;89:789–93. 36. United Nations Statistics Division, Department of
21. Malik FR, Swati AA, Akhter S, et al. Retrospective Economic and Social Affairs. Principles and
Analysis of Maternal Mortality at a Tertiary Care Recommendations for Population and Housing
Hospital of Peshawar, Pakistan. Khyber Med Univ J Censuses. New York: United Nations, 2007.

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

2015;7:25–9. 37. Queiroz BL. Estimating maternal mortality differentials
22. Gumanga SK, Kolbila DZ, Gandau BBN. Trends in using census data: experience in Honduras. J Popul Res
maternal mortality in Tamale teaching hospital, Ghana. 2011;28:75–87.
Ghana Med J 2011;45:105–10. 38. Qomariyah SN, Braunholtz D, Achadi EL, et al. An
23. Iftikhar R.. A study of maternal mortality. J Surg option for measuring maternal mortality in developing
Pakistan 2009;14:177. countries: a survey using community informants. BMC
24. Igwegbe AO, Eleje GU, Ugboaja JO, et al. Improving Pregnancy Childbirth 2010;10:74–81.
maternal mortality at a university teaching hospital in 39. Hill K, Queiroz BL, Wong L, et al. Estimating
Nnewi, Nigeria. Int J Gynecol Obstet 2012;116: pregnancy-related mortality from census data: experi-
197–200. ence in Latin America. Bull World Health Organ 2009;
25. Lema VM, Changole J, Kanyighe C, et al. Maternal mor- 87:288–95.
tality at the Queen Elizabeth Central Teaching Hospital, 40. Garenne M, McCaa R, Nacro K. Maternal mortality in
Blantyre, Malawi. East Afr Med J 2005;82:3–9. South Africa in 2001: from demographic census to epi-
26. Li N, Matchi E, Spiegelman D, et al. Maternal mortality demiological investigation. Popul Health Metr 2008;6:
among HIV-infected pregnant women in Tanzania. 1–13.
Acta Obstet Gynecol Scand 2014;93:463–8. 41. Bell JS, Byass P, Fitzmaurice AE, et al. The epidemi-
27. Malatyalioglu E, Kokcu A, Cetinkaya MB, et al. ology of pregnancy outcomes in rural Burkina Faso.
Maternal mortality rates in the last eight years: a uni- Trop Med Int Health 2008;13:31–43.
versity hospital-based study from Turkey. J Matern 42. World Health Organization. Verbal Autopsy Standards:
Fetal Neo Med 2006;19:353–6. The 2012 WHO Verbal Autopsy Instrument. Geneva:
28. Mundkur A, Rai L. Prepare and prevent rather than WHO, 2012.
repair and repent: Study of maternal mortality in ter- 43. Agampodi S, Wickramage K, Agampodi T, et al.
tiary care hospital. Int J Med Public Health 2013;3: Maternal mortality revisited: the application of the
163–7. new ICD-MM classification system in reference to
29. Okeh UM. Statistical analysis of the maternal death rate maternal deaths in Sri Lanka. Reprod Health 2014;11:
at the Ebonyi State University Teaching Hospital, 17.
Abakaliki for the year ending 31 December 2007. Afr J 44. Ba MG, Kodio B, Etard JF. [Verbal autopsy to measure
Prim Health Care Fam Med 2009;1:118–20. maternal mortality in rural Senegal]. J Gynecol Obstet
30. Olopade FE, Lawoyin TO. Maternal mortality in a Biol Reprod (Paris) 2003;32:728–35.
Nigerian Maternity Hospital. Afr J Biomed Res 2010; 45. Barnett S, Nair N, Tripathy P, et al. A prospective key
11:267–73. informant surveillance system to measure maternal
31. Onakewhor JUE, Gharoro EP. Changing trends in mortality – findings from indigenous populations in
maternal mortality in a developing country. Nigerian J Jharkhand and Orissa, India. BMC Pregnancy
Clin Pract 2008;11:111–20. Childbirth 2008;8:6.
32. Rulisa S, Umuziranenge I, Small M, et al. Maternal near 46. Chandy H, Heng YV, Samol H, et al. Comparing two
miss and mortality in a tertiary care hospital in survey methods for estimating maternal and perinatal
Rwanda. BMC Pregnancy Childbirth 2015;15:203. mortality in rural Cambodia. Women Birth 2008;21:
33. Sule-Odu AO. Maternal deaths in Sagamu, Nigeria. Int 9–12.
J Gynaecol Obstet 2000;69:47. 47. Farooq N, Jadoon H, Masood TI, et al. An assessment
34. Tebeu PM, Ngassa P, Kouam L, et al. Maternal mortal- study of maternal mortality ratio databank in five dis-
ity in Maroua Provincial Hospital, Cameroon (2003– tricts of North Western Frontier Province Pakistan. J
2005). West Indian Med J 2007;56:502–7. Ayub Med Co Abbottabad 2006;18:64–8.
Methods to estimate the maternal mortality ratio (MMR), 2017, Vol. 121 133

48. Jafarey RN, Rizvi T, Koblinsky M, et al. Verbal autopsy Inc., 1997. Available from: https://dhsprogram.com/
of maternal deaths in two districts of Pakistan – filling pubs/pdf/AR4/AR4.pdf
information gaps. J Health Popul Nutr 2009;27: 61. Moseson H, Massaquoi M, Bawo L, et al. Estimation of
170–83. maternal and neonatal mortality at the subnational level
49. Karabulut A, Çalişkan A, Özcan. Maternal mortality in in Liberia. Int J Gynecol Obstet 2014;127:194–200.
Denizli region: three years’ evaluation. Turkiye 62. Adegoke AA, Campbell M, Thomson AM, et al.
Klinikleri Jinekoloji Obstetrik 2010;20:29–34. Community study of maternal mortality in south west
50. McCaw-Binns A, Lindo JL, Lewis-Bell KN, et al. Nigeria: How applicable is the sisterhood method?
Maternal mortality surveillance in Jamaica. Int J Matern Child Hlth J 2013;17:319–29.

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

Gynaecol Obstet 2008;100:31–6. 63. Oye-Adeniran BA, Odeyemi KA, Gbadegesin A, et al.
51. Mohammed AA, Elnour MH, Mohammed EE, et al. The use of the sisterhood method for estimating mater-
Maternal mortality in Kassala State - Eastern Sudan: nal mortality ratio in Lagos state, Nigeria. J Obstet
community-based study using reproductive age mortal- Gynaecol 2011;31:315–9.
ity survey (RAMOS). BMC Pregnancy Childbirth 2011; 64. Aa I, Grove MA, Haugsjå AH, et al. High maternal
11:102. mortality estimated by the sisterhood method in a rural
52. Olsen BE, Hinderaker SG. Estimates of maternal mor- area of Mali. BMC Pregnancy Childbirth 2011;11:56.
tality by the sisterhood method in rural northern 65. Mbaruku G, Vork F, Vyagusa D, et al. Estimates of
Tanzania: a household sample and an antenatal clinic maternal mortality in western Tanzania by the sister-
sample. Brit J Obstet Gynaecol 2000;107:1290–7. hood method. Afr J Reprod Health 2003;7:84–91.
53. Prata N, Gerdts C, Gessessew A. An innovative 66. Font F, Alonso González M, Nathan R, et al. Maternal
approach to measuring maternal mortality at the com- mortality in a rural district of south Eastern Tanzania:
munity level in low-resource settings using mid-level an application of the sisterhood method. Int J
providers: a feasibility study in Tigray, Ethiopia. Epidemiol 2000;29:107–12.
Reprod Health Matters 2012;20:196–204. 67. Lech MM, Zwane A. Survey on maternal mortality in
54. Saleem S, McClure EM, Goudar SS, et al. A prospective Swaziland using the sisterhood method. Paediatr
study of maternal, fetal and neonatal deaths in low- and Perinat Epidemiol 2002;16:101–7.
middle-income countries. Bull World Health Organ 68. Orach CG. Maternal mortality estimated using the
2014;92:605–12. Sisterhood method in Gulu district, Uganda. Trop Doct
55. Singh P, Pandey A, Aggarwal A. House-to-house survey 2000;30:72–4.
vs. snowball technique for capturing maternal death in 69. Smith JB, Fortney JA, Wong E, et al. Estimates of the
India: a search for a cost-effective method. Indian J maternal mortality ratio in two districts of the Brong-
Med Res 2007;125:550–6. Ahafo region, Ghana. Bull World Health Organ 2001;
56. van den Broek NR, White SA, Ntonya C, et al. 79:400–8.
Reproductive health in rural Malawi a population- 70. Bhat PNM. Maternal mortality in India: an update.
based survey. Brit J Obstet Gynaecol 2003;110:902–8. Stud Family Plann 2002;33:227–36.
57. Ziraba AK, Madise N, Mills S, et al. Maternal mortality 71. AbouZahr C. Global burden of maternal death and dis-
in the informal settlements of Nairobi city: what do we ability. Brit Med Bull 2003;67:1–11.
know? Reprod Health 2009;22:6. 72. Mgawadere F, Unkels R, Adegoke A, et al. Measuring
58. Graham W, Brass W, Snow RW. Estimating maternal maternal mortality using a Reproductive Age Mortality
mortality: the sisterhood method. Stud Family Plann Study (RAMOS). BMC Pregnancy Childbirth 2016;16:291.
1989;20:125–35. 73. Mohammed AA, Elnour MH, Mohammed EE, et al.
59. Rutenberg N, Sullivan J. Direct and indirect esti- Maternal mortality in Kassala State - Eastern Sudan:
mates of maternal mortality from the sisterhood community-based study using Reproductive age mortal-
method, Proceedings of the Demographic and ity survey (RAMOS). BMC Pregnancy Childbirth 2011;
Health Surveys World Conference, 5–7 August 16:102.
1991. Vol. 3. Washington, DC, USA: Institute for 74. Amarin Z, Khader Y, Okour A, et al. National maternal
Resource. 1991. mortality ratio for Jordan, 2007–2008. Int J Gynecol
60. Stanton C, Abderrahim N, Hill K. DHS Maternal Obstet 2010;111:152–56.
Mortality Indicators: An Assessment of Data Quality 75. Zakariah AY, Alexander S, van Roosmalen J, et al.
and Implications for Data Use. Calverton, MD: Reproductive age mortality survey (RAMOS) in Accra,
Demographic and Health Surveys, Macro International, Ghana. Reprod Health 2009;6:7.
134 F. Mgawadere et al., 2017, Vol. 121

76. WHO. ICD-10, the ICD-10 Classification of Mental 81. Moodley J, Pattinson RC, Fawcus S, et al. The con-
and Behavioural Disorders: Clinical Descriptions and fidential enquiry into maternal deaths in South
Diagnostic Guidelines. Geneva: World Health Africa: a case study. Brit J Obstet Gynaec 2014;
Organization, 1992. 121:S53–60.
77. Hogan MC, Foreman KJ, Naghavi M, et al. Maternal 82. World Health Organisation. Beyond the Numbers:
mortality for 181 countries, 1980–2008: a systematic Reviewing Maternal Deaths and Complications to
analysis of progress towards Millennium Development Make Pregnancy Safer. Geneva: Department of
Goal 5. Lancet 2010;375:1609–1623. Reproductive Health and Research, WHO, 2004.
78. Deneux-Tharaux C, Berg C, Bouvier-Colle MH, et al. 83. Yego F, D’Este C, Byles J, et al. Risk factors for mater-

Downloaded from https://academic.oup.com/bmb/article/121/1/121/2926155 by guest on 05 November 2023

Underreporting of pregnancy-related mortality in the nal mortality in a Tertiary Hospital in Kenya: A case
United States and Europe. Obstet Gynecol 2005;106: control study. BMC Pregnancy Childbirth 2014;14:38.
684–92. 84. al-Meshari A, Chattopadhyay SK, Younes B, et al.
79. Hoyert DL, Danel I, Tully P. Maternal mortality, United Trends in maternal mortality in Saudi Arabia. Int J
States and Canada, 1982–1997. Birth 2000;27:4–11. Gynecol Obstet 1996;52:25–32.
80. MBRRACE-UK. Saving Lives, Improving Mothers’ 85. United Nations Statistical Division. Principles and
Care Surveillance of maternal deaths in the UK 2011- recommendations for population and housing censuses /
13 and lessons learned to inform maternity care from Department of Economic and Social Affairs, Statistics
the UK and Ireland Confidential Enquiries into Division. New York: United Nations Publications,
Maternal Deaths and Morbidity 2009-13. Oxford: 2008.
National Perinatal Epidemiology Unit, 2015. Available 86. Mungra A, Van Bokhoven SC, Florie J. Reproductive
from: https://www.npeu.ox.ac.uk/downloads/files/mbrra age mortality survey to study under-reporting of mater-
ce-uk/reports/MBRRACE-UK%20Maternal%20Report nal mortality in Surinam. Eur J Obstet Gynecol Reprod
%202015.pdf Biol 1998;77:37–9.