Mobilizing communities to improve maternal health: Results of an intervention in rural Zambia

Objective: To determine whether a complex community intervention in rural Zambia improved understanding of maternal health and increased use of maternal health-care services. Methods: The intervention took place in six rural districts selected by the Zambian Ministry of Health. It involved community discussions on safe pregnancy and delivery led by trained volunteers and the provision of emergency transport. Volunteers worked through existing government-established Safe Motherhood Action Groups. Maternal health indicators at baseline were obtained from women in intervention (n = 1775) and control districts (n = 1630). The intervention’s effect on these indicators was assessed using a quasi-experimental difference-in-difference approach that involved propensity score matching and adjustment for confounders such as education, wealth, parity, age and distance to a health-care facility. Findings: The difference-in-difference comparison showed the intervention to be associated with significant increases in maternal health indicators: 14-16% in the number of women who knew when to seek antenatal care; 10-15% in the number who knew three obstetric danger signs; 12-19% in those who used emergency transport; 22-24% in deliveries involving a skilled birth attendant; and 16-21% in deliveries in a health-care facility. The volunteer drop-out rate was low. The estimated incremental cost per additional delivery involving a skilled birth attendant was around 54 United States dollars, comparable to that of other demand-side interventions in developing countries. Conclusion: The community intervention was associated with significant improvements in women’s knowledge of antenatal care and obstetric danger signs, use of emergency transport and deliveries involving skilled birth attendants.

We adopted a quasi-experimental approach to evaluating the effect of a complex community-based intervention that was devised to reduce barriers to the use of maternal health-care services and to increase deliveries involving a skilled birth attendant. The intervention was novel because it involved the whole community and emphasized social approval and its ability to bring about changes in behaviour. The involvement of men, older women and community leaders is vital in places where behaviour is dependent on their approval. In our intervention, men were encouraged to become emergency transport drivers, community leaders were invited to train as community volunteers and older women, who are often traditional birth attendants, were encouraged to become mother’s helpers and were trained to recognize obstetric danger signs. This reduced the likelihood that older women would pressure younger women to conform to dangerous practices. The intervention involved improving the effectiveness of Safe Motherhood Action Groups by training volunteers and developing and strengthening systems in the community that help women get to heath-care facilities, such as existing arrangements for emergency transport. In practice, training was cascaded across communities: core trainers trained lead volunteers at district health offices and lead volunteers, in turn, trained volunteers within each community. Since the literacy level of volunteers and communities was poor, we adopted a largely paper-free approach to training volunteers and to communicating with communities. Each trained volunteer coordinated a cycle of four community discussions on safe pregnancy and delivery, with additional discussions on neonatal care. Both women and men were encouraged to participate. In addition, communities were encouraged to develop plans for helping women access maternal health services, which included ensuring that transport was accessible and that people (i.e. “mother’s helpers”) were available to assist with child-minding or to accompany women to health-care facilities. Communities were also asked to think about other actions that could be taken to ensure that pregnant women were taken to a health centre in an emergency (e.g. the introduction of a local “law”), to consider social issues that could have a bearing on maternal health, such as violence against women, and to ensure that women without family or other support were included in any initiatives. Although there was some investment in equipment and supplies at health-care facilities in the intervention areas, most was provided towards the end of the intervention period and was unlikely to have had a substantial impact on the use of maternal health-care services. The intervention was also intended to reduce the delay many women experience in reaching health-care facilities. The distances involved and the poor road conditions mean that, for rural communities, travel is often slow and may be dangerous. Moreover, emergency transport to hospitals is seldom available at health centres. As part of the intervention, community transport, appropriate to the terrain, was provided for groups of two or three villages. In most cases, bicycle ambulances were provided but, in areas where the terrain was particularly uneven or sandy, communities were given ox or donkey carts. One community, which is situated next to a river, was provided with a boat. In addition, health centres were provided with motorcycle ambulances. Each community identified a group of volunteers who were willing to operate and maintain the vehicles and some basic training was provided through the intervention. Follow-up reviews revealed that, in most cases, the vehicles had been well maintained by the communities and that additional vehicle operators were being recruited and trained. Information on the use of emergency transport during the intervention period was obtained by asking women if they had used community transport or transport provided by a health-care facility to reach a health-care facility or hospital for delivery. Additional information on transport was available through a community monitoring system, which was established as part of the intervention to report on women’s use of emergency transport and of health services, such as antenatal care. The intervention was implemented in six districts selected by the Zambian Ministry of Health largely because they were not already receiving substantive assistance from donors to improve maternal health. Three districts – Serenje, Mongu and Choma – took part in phase one, which started in July 2011, and three – Chama, Kaoma and Mkushi – took part in phase two, which started in January 2012. As a result of the experience gained in phase one, implementation of the intervention – the procurement of vehicles, for example – was slightly quicker in phase two. Otherwise the interventions were identical. Two catchment areas were selected for the intervention in Serenje, Mongu, Choma and Chama, three were selected for Kaoma and one was selected for Mkushi. Each district had a basic emergency obstetric care centre to which smaller health centres or posts were most likely to refer patients. The intervention covered about 25% of the population of each district and included a total of 250 000 inhabitants. In addition, data were collected in five control districts with similar maternal health indicators to monitor changes over time. Any spill-over effects of the intervention were minimized by ensuring that control districts were not adjacent to intervention districts. A baseline survey of recent births was conducted between December 2010 and May 2011 in both intervention and control districts to assess: (i) current use of maternal and neonatal health-care services; (ii) knowledge of maternal care, including when antenatal care should first be received, and of obstetric danger signs; and (iii) use of community systems that help women obtain care, such as savings schemes and emergency transport (Table 1). A list of all births in the three months before the survey in each area was made using information provided by Safe Motherhood Action Groups. The sample size of women who had recently given birth was 3405 at baseline: 1775 in intervention districts and 1630 in control districts. In the final survey, which was carried out in October 2012, the sample size was 2788: 1445 in intervention districts and 1343 in control districts. For any intervention for improving maternal health care, it is important to know whether it is sustainable, scalable and cost effective. Although a full cost-effectiveness analysis was outside the scope of this study, we examined some of the costs involved. The main start-up costs were for training volunteers and providing vehicles. We assumed that 20% of volunteers would drop out of the programme each year. The main recurrent costs were for maintenance of vehicles and transportation. As a summary measure, we used the incremental cost per additional delivery involving a skilled birth attendant. Ethical approval for the study was obtained from the University of Zambia Biomedical Research Ethics Committee. All study participants were provided with, or had read out to them, an informed consent form approved by the ethics committee. The form provided information on the study, on how data would be used and on the participant’s right to withdraw at any time. The indicators used to assess the effect of the intervention on maternal health care were, before delivery, knowledge that antenatal care should first be received in the first trimester, receipt of antenatal care, knowledge of obstetric danger signs (i.e. fever, discharge, blood loss, severe headache and retained placenta) and use of emergency transport and, subsequently, delivery involving a skilled birth attendant, delivery at a health-care facility, receipt of postnatal care within 6 days and use of modern contraceptives after giving birth. Changes in these indicators due to the intervention were assessed using a difference-in-difference approach that involved a proxy counterfactual drawn from households in control districts. This approach controls for confounding factors and for any general changes that would have occurred over time in the absence of the intervention. The basic equation used in the difference-in-difference approach was: where Y is the binary response variable (e.g. delivery at a health-care facility or knowledge of three obstetric danger signs), T is a time variable (baseline = 0, final survey = 1), I is the intervention variable (intervention district = 1, control district = 0), X is a vector of covariates, including education (i.e. highest level achieved), household wealth, the woman’s age and parity and the distance of the woman’s home from the health centre. The impact of the intervention is expressed by the parameter β3, β1 is a measure of the general change in the response variable between baseline and final surveys, β2 represents the general difference in the response variable between intervention and control areas, β0 is a constant, γ is a vector representing the impact of covariates and e is an error term. Since the accuracy of the difference-in-difference approach is greater if time trends in the control and intervention groups are similar and the time trends are more likely to be similar if the characteristics of the individuals in the groups are broadly similar, we used propensity score matching to match individuals in intervention districts with similar individuals in control districts. Individuals without a close match were excluded from the analysis. Three types of propensity score matching were used: (i) nearest-neighbour matching, which provides a one-to-one match between an observation in the intervention group and the nearest observation in the control group; (ii) calliper matching, which matches observations within a defined distance (i.e. the difference in propensity score between an observation in the intervention group and the nearest observation in the control group); and (iii) radius matching, which takes a weighted average of all observations within a defined distance. Matching reduced the average difference between control and intervention areas to below 10% for most key characteristics. The remaining differences were controlled for by including the relevant characteristics in the difference-in-difference specification.

N/A

• Dima Health