Can women's groups reduce maternal and newborn deaths?

Maternal and Child Health Intervention Research Group, Department of Population Health, London School of Hygiene and Tropical Medicine, London WC1E 7HT, UK. Electronic address: .
The Lancet (Impact Factor: 45.22). 05/2013; 381(9879):e12-e14. DOI: 10.1016/S0140-6736(13)60985-X
Source: PubMed


Available from: Betty Rosamund Kirkwood, Mar 24, 2016
Comment Vol 381 May 18, 2013 e
Can women’s groups reduce maternal and newborn deaths?
Evaluation of interventions to reduce the risk of
maternal, newborn, or child mortality in sub-Saharan
Africa is a priority because of the lack of progress
towards the Millennium Development Goals (MDGs)
and the high burden of newborn, infant, and maternal
deaths borne by this region.
In The Lancet, Sonia Lewycka and colleagues
report the
results of their ambitious trial in Malawi of simultaneous
evaluation of the eff ect of two somewhat diff erent
interventions: women’s groups with facilitated action-
learning cycles to tackle maternal, newborn, and child
health problems; and home visits by female volunteer
peer counsellors to promote breastfeeding and infant
care. The investigators postulated that women’s groups
would reduce the rates of maternal, newborn, and
infant mortality, and that volunteer peer counselling
would improve exclusive breastfeeding rates in the fi rst
6 months and reduce infant mortality rates.
The interventions were assessed over 3 years in
a factorial cluster-randomised controlled trial with
48 village clusters randomly assigned as four groups
of 12 clusters receiving no intervention (controls),
women’s groups only, volunteer peer counselling only,
or both interventions. This is an effi cient design that
allows for simultaneous evaluation of two separate
interventions, but its complexity can lead to diffi culties
in interpretation and has done so in this study.
A major issue is choosing the appropriate model to
use in regression analyses, and, in particular, deciding
whether or not to assume that the two interventions
act independently of each other. The model with the
assumption of independent eff ects is the norm for
analysing factorial randomised controlled trials with the
odds ratio for the eff ect of women’s groups assumed
to be the same whether or not the volunteer peer coun-
selling is also taking place, as shown in the table; it is
based on a combined comparison of the two groups
with the women’s groups versus the two groups without
this intervention. The odds ratio for the group with both
interventions is simply the odds ratios for women’s groups
and volunteer peer counselling multiplied together.
Also shown in the table are the odds ratios fi tted by the
alternative model that allows for interaction between the
eff ects of the interventions. The estimate obtained for
the eff ect of women’s groups in the interaction model is
based only on data from the stratum with no volunteer
peer counselling, and indicates the diff erence between
the women’s group only and the control group; it will be
diff erent from the estimate obtained with the factorial
model assuming independence unless the interaction
term is 1, in which case the two models are equivalent.
Similar considerations apply to the volunteer peer coun-
selling odds ratio that is based solely on a comparison of
volunteer peer counselling only with the control group.
The odds ratio for the group given both interventions
equals the product of the women’s group and volunteer
peer counselling odds ratio and the interaction term.
The interaction term is greater than 1 if the women’s
group and volunteer peer counselling delivered together
have a synergistic eff ect (combined eff ect greater than
the product of their individual eff ects). It will be less
than 1 if their combined eff ect is less than the product of
their individual eff ects, which might occur, for example,
if the interventions were saving the same lives, or if their
combined delivery led to decreased effi ciency in either or
both of them.
The table shows that if the interaction term is not
equal to one then the interaction model gives a diff erent
Factorial model Interaction model
ORs fi tted WG eff ect in each stratum ORs fi tted WG eff ect in each stratum
Stratum without VPC WG
no VPC
Control 1 ·· 1 ··
WG only WG
·· WG
no VPC
Stratum with VPC WG
VPC areas
= WG
no VPC
× IT
VPC only VPC
·· VPC
no WG
WG and VPC
no VPC
no WG
× IT
OR=odds ratio. WG=women’s groups. VPC=volunteer peer counselling. IT=interaction term.
Table: Comparison of odds ratios fi tted in the factorial and interaction models and the resulting stratum-specifi c WG eff ects
See Correspondence page e15
See Articles pages 1721 and 1736
Page 1
e13 Vol 381 May 18, 2013
eff ect of women’s groups in the stratum where the
volunteer peer counselling intervention was being
implemented as compared with the stratum without
volunteer peer counselling; the diff erence between the
two depends on the nature of the interaction with:
where WG is women’s groups, OR is the odds ratio, VPC is
volunteer peer counselling, and IT is the interaction term.
The main fi ndings are based on data from years 2 and 3,
allowing a year for embedding of the women’s groups,
and adjustment for clustering, baseline diff erences
between the four treatment groups, socioeconomic
quintile, and parity. The odds ratio obtained from the
standard factorial regression model for the eff ect on the
neonatal mortality rate was 0·85, corresponding to a
15% reduction but with a very wide 95% CI (0·59–1·22)
and a large p value (0·39). There was a suggestion that
women’s groups reduced the maternal mortality rate; the
odds ratio was 0·48 (0·26–0·91; p=0·0238), although the
investigators note that after allowing for multiple testing
this p value was no longer signifi cant at the 5% level.
The stratum-specifi c odds ratios obtained in the
interaction model seem to tell a diff erent story, with
women’s groups leading to sizeable reductions in
both neonatal mortality rate (0·59, 95% CI 0·40–0·86;
p=0·0058) and maternal mortality rate (0·26; 0·10–0·70;
p=0·0074) in areas with no volunteer peer counselling.
By contrast, the odds ratios, based on the stratum with
volunteer peer counselling, are greater than 1 with wide
95% CIs and large p values for both neonatal mortality
rate (1·38, 0·75–2·54; p=0·31) and maternal mortality
rate (1·09, 0·40–2·98; p=0·86), leading the authors to
conclude that the women’s groups had no eff ect in areas
with volunteer peer counselling, but that they did in
areas without volunteer peer counselling.
The reason that the odds ratios from the two models
are so diff erent is because of large and highly signifi cant
negative interaction terms in the model allowing for
stratum-specifi c eff ects; negative eff ects were seen in
the group with both women’s groups and volunteer peer
counselling, and positive eff ects were seen in each of
the women’s group and the volunteer peer counselling.
Unfortunately, as the investigators point out, the
interaction eff ects seem to be due to major diff erences
between the four groups at inception or baseline and
not to any true negative interaction between the
interventions, which does not seem to be biologically
plausible. It is noteworthy that participation was
higher in the group receiving both interventions than
in the group receiving only women’s groups with 57%
compared with 50% of pregnant women participating,
The issue is how to interpret the fi ndings in view
of the very diff erent values for the three odds ratios
, WG
no VPC
, and WG
VPC areas
). Despite a well done trial
and comprehensive analyses, we are left wondering
what the results show. Can women’s groups save lives
in sub-Saharan Africa? We think the most appropriate
model to be the one assuming independent eff ects,
which is the standard for factorial designs. Doing
stratum-specifi c analyses is akin to doing subgroup
analyses, which are not recommended if no overall
eff ect is seen because they can lead to misleading
Our conclusion, therefore, is that
unfortunately the results of the MaiMwana trial from
Malawi have not convincingly shown an eff ect of
women’s groups in reducing the rate of newborn or
maternal deaths in this setting.
Although this trial provides the only evidence so far
concerning the eff ect of women’s groups in an African
setting, there have been several previous trials in
Asia. Also appearing in The Lancet is a meta-analysis
by the same research group of seven trials.
Four of
the trials have been previously reported, the fi fth is
the MaiMwana trial, and the other two are in press
and not yet accessible. The meta-analysis showed
substantial heterogeneity between the trial fi ndings,
and Audrey Prost and colleagues
investigated how
much of this could be due to the diff ering participation
rates in women’s groups in the diff erent trials. They
conclude that women’s groups have saved maternal
and newborn deaths in the four trials (including the
Malawi trial) in which participation rates have been
greater than 30%, but not in the three trials that did
not achieve this. It should be noted that the meta-
analysis only includes data for the Malawi trial from
the stratum with no volunteer peer counselling and
does not include data from the other stratum; in our
view, it should use the overall odds ratios, or include
the odds ratios from both strata; omitting 50% of the
data is misleading. Inclusion of all data might change
the conclusions of the meta-analysis.
VPC areas no VPC
Page 2
Comment Vol 381 May 18, 2013 e
The authors also use the meta-analysis to estimate
that women’s groups could save 283 000 newborn
deaths and 41 100 maternal deaths if implemented in
rural areas of the 74 Countdown countries, provided
they achieved participation of greater than 30% of
pregnant women. But is this estimation realistic since
three of seven trials undertaken in research conditions
with intensive input did not achieve this? Important
questions to be addressed are why not, what was
diff erent between the successful and non-successful
trials, and what lessons can be learnt about achieving
high participation?
In conclusion, although women’s groups might have
a part to play in saving maternal and newborn deaths in
sub-Saharan Africa, the data from the MaiMwana trial
do not provide conclusive support. The meta-analysis
results show large variability in fi ndings of included trials
and issues of how to achieve widespread participation at
scale remain to be addressed.
*Betty Kirkwood, Rajiv Bahl
Maternal and Child Health Intervention Research Group,
Department of Population Health, London School of Hygiene and
Tropical Medicine, London WC1E 7HT, UK (BK); and Department
of Maternal, Newborn, Child and Adolescent Health, World Health
Organization, Geneva, Switzerland (RB)
We declare that we have no confl icts of interest.
1 WHO. Countdown to 2015: maternal, newborn and child survival.
Accountability for maternal, newborn and child survival: an update on
progress in priority countries. Geneva: World Health Organization, 2012.
2 Lewycka S, Mwansambo C, Rosato M, et al. Eff ect of women’s groups and
volunteer peer counselling on rates of mortality, morbidity, and health
behaviours in mothers and children in rural Malawi (MaiMwana):
a factorial, cluster-randomised controlled trial. Lancet 2013; 381: 1721–35.
3 Kirkwood BR, Sterne JAC. Essential medical statistics, 2nd edn. Malden, MA:
Blackwell Science, 2003.
4 Prost A, Colbourn T, Seward N, et al. Women’s groups practising
participatory learning and action to improve maternal and newborn health
in low-resource settings: a systematic review and meta-analysis.
Lancet 2013; 381: 1736–46.
© 2013. World Health Organization. Published by Elsevier Ltd/Inc/BV.
All rights reserved.
Page 3
  • Source
    • "Remote districts of the northern mountainous regions and southern areas of the country have the highest perinatal and neonatal mortality rates [4]. Despite some progress in improving perinatal and neonatal mortality through communitybased interventions of maternal and neonatal care packages [5,6], relatively few large-scale community-based projects have delivered neonatal interventions within local health systems and through public sector health workers with perinatal and neonatal mortality as a defined outcomes7891011. Available evidence suggests that promotion of maternal and newborn care practices through implementation of community-based packages, including promotion of essential newborn care and community mobilization, are effective in improving neonatal survival in low income settings12131415. "
    [Show abstract] [Hide abstract] ABSTRACT: There is limited evidence from community-based interventions to guide the development of effective maternal, perinatal and newborn care practices and services in developing countries. We evaluated the impact of a low-cost package of community-based interventions implemented through government sector lady health workers (LHWs) and community health workers (CHWs) of a NGO namely Aga Khan Health Services on perinatal and neonatal outcomes in a sub-population of the remote mountainous district of Gilgit, Northern Pakistan. The package was evaluated using quasi experimental design included promotion of antenatal care, adequate nutrition, skilled delivery and healthy newborn care practices. Control areas continued to receive the routine standard health services. The intervention areas received intervention package in addition to the routine standard health services. Outcome measures included changes in maternal and newborn-care practices and perinatal and neonatal mortality rates between the intervention and control areas. The intervention was implemented in a population of 283324 over a 18 months period. 3200 pregnant women received the intervention. Significant improvements in antenatal care (92% vs 76%, p < .001), TT vaccination (67% vs 47%, p < .001), institutional delivery (85% vs 71%, p < .001), cord application (51% vs 71%, p < .001), delayed bathing (15% vs 43%, p < .001), colostrum administration (83% vs 64%, p < .001), and initiation of breastfeeding within 1 hour after birth (55% vs 40%, p < .001) were seen in intervention areas compared with control areas. Our results indicate significant reductions in mortality rates in intervention areas as compared to control areas from baseline in Perinatal Mortality Rate (from 47.1 to 35.3 per 1000 births, OR 0.62; 95% CI: 0.56-0.69; P 0.02) and neonatal mortality rates (from 26.0 to 22.8 per 1000 live births, 0.58; 95% CI: 0.48-0.68; P 0.03). The implementation of a set of low cost community-based intervention package within the health system settings in a mountainous region of Pakistan was found to be both feasible and beneficial. The interventions had a significant impact in reduction of the burden of perinatal and neonatal mortality. This study is registered, NCT02412293 .
    Full-text · Article · Apr 2015 · BMC Pregnancy and Childbirth
  • No preview · Article · May 2013 · The Lancet