Community-based approaches for neonatal survival: meta-analyses of randomized trial data

Abstract Objective To analyse the impact of community approaches to improving newborn health and survival in low-resource countries. Methods We updated previous meta-analyses of published cluster randomized trials of community-based interventions for neonatal survival. For each study we extracted baseline data on the context: geographical area; available facilities and staffing; immediate breastfeeding and facility births; and neonatal mortality. We also extracted data on the primary outcome (neonatal survival) and intermediate outcomes of the interventions (changes in immediate breastfeeding and facility births). We used forest plots and pooled sub-group analysis to seek patterns in associations between the effect size and the context or type of intervention (home-based counselling or women’s groups). Findings We included 17 trials, spanning years from 2001 to 2013. A 25% reduction in neonatal mortality (relative risk, RR: 0.75; 95% confidence interval, CI: 0.69–0.80) was found when pooling six studies in settings with 44 or more deaths per 1000 live births. In lower-mortality settings (pooling six studies with 32 or fewer deaths per 1000 live births) there was no evidence of an effect. We observed some evidence that community approaches had a stronger effect in south Asia than in sub-Saharan Africa. Community approaches had a lower impact on neonatal mortality in settings where at least 44% of women delivered in a facility. Conclusion As neonatal mortality declined, the impact of community approaches on survival appeared to be lower, and the role of these approaches in supporting newborn care in weak health systems may need to be re-examined.


Introduction
Despite progress in reducing child deaths in the past 25 years, an estimated 2.6 million neonatal deaths occurred globally in 2015. 1 Sustainable development goal (SDG) 3 included the target of no more than 12 deaths per 1000 live births in the first 28 days of life. 2 To reach the target, more effective ways of delivering quality preventive and curative care need to be identified and monitored.
Approaches based on health promotion and on community empowerment and participation have long been promoted as part of formal health-care systems in low-and middleincome countries. 3,4 Trials to improve maternal and newborn health through community approaches have focused on two approaches: (i) home-based counselling 5 and (ii) participatory women's groups. 6 Both approaches promote appropriate care-seeking as well as improved home practices in newborn care. Home-based counselling focuses on health education and behaviour change to improve newborn care practices by mothers, such as immediate breastfeeding, dry cord care and appropriate health care (e.g. delivering in a health-care facility and seeking care for sick newborns). Women's groups use an empowerment and problem-solving approach aiming similarly to improve care practices and care-seeking by mothers of newborns. The mechanisms of the effect of the home-based counselling strategies are backed by an analysis using the Lives Saved tool. 7 Previous meta-analyses have reported moderate effects on neonatal mortality of both home-based counselling and women's groups. A meta-analysis of five proof-of-principle trials of home-based counselling in south Asia in 2010 found an almost 40% reduced risk of neonatal death (relative risk, RR: 0.62; 95% confidence interval, CI: 0.44-0.87). 5 In response, the World Health Organization (WHO) recommended home visits to improve neonatal health in high neonatal mortality settings. 8 However, trials of home-based counselling conducted in a larger population and in programme settings 9,10 showed a smaller risk reduction for neonatal mortality (RR: 0.93; 95% CI: 0.85-1.01). 9 A review of seven trials of women's groups based on participatory learning and action cycles published in 2013 reported a 20% reduction in neonatal mortality (RR: 0.77; 95% CI: 0.65-0.90). 6 The evidence prompted WHO to recommend community mobilization with women's groups to improve maternal and neonatal health. 11 Factors reported to have the greatest impact on neonatal mortality include how successfully the intervention was implemented, as reflected by the proportion of pregnant women participating in women's groups; 6 the inclusion of injectable antibiotics for treatment of possible severe bacterial infection; 6 and home management of asphyxia. 5 However, it is not clear how the women's group approach works, 12 or what is the interaction between community approaches and contextual factors, such as the characteristics of the healthcare system.
In this paper we updated previous searches and metaanalyses of trials of home-based counselling and women's groups in low-resource countries. The aim was to generate and test hypotheses about which factors may lead to weaker or stronger effects on neonatal survival. We examined associations between reductions in neonatal mortality and the context in which the trial took place or the characteristics of the local health system. We also assessed associations between reductions in mortality and the characteristics of the implementation.
Objective To analyse the impact of community approaches to improving newborn health and survival in low-resource countries. Methods We updated previous meta-analyses of published cluster randomized trials of community-based interventions for neonatal survival. For each study we extracted baseline data on the context: geographical area; available facilities and staffing; immediate breastfeeding and facility births; and neonatal mortality. We also extracted data on the primary outcome (neonatal survival) and intermediate outcomes of the interventions (changes in immediate breastfeeding and facility births). We used forest plots and pooled sub-group analysis to seek patterns in associations between the effect size and the context or type of intervention (home-based counselling or women's groups). Findings We included 17 trials, spanning years from 2001 to 2013. A 25% reduction in neonatal mortality (relative risk, RR: 0.75; 95% confidence interval, CI: 0.69-0.80) was found when pooling six studies in settings with 44 or more deaths per 1000 live births. In lowermortality settings (pooling six studies with 32 or fewer deaths per 1000 live births) there was no evidence of an effect. We observed some evidence that community approaches had a stronger effect in south Asia than in sub-Saharan Africa. Community approaches had a lower impact on neonatal mortality in settings where at least 44% of women delivered in a facility. Conclusion As neonatal mortality declined, the impact of community approaches on survival appeared to be lower, and the role of these approaches in supporting newborn care in weak health systems may need to be re-examined.

Research
Community-based approaches for newborn health Claudia Hanson et al.

Inclusion criteria and search methods
We reviewed cluster randomized trials evaluating community approaches to enhancing neonatal survival in low-and middle-income countries in April 2016, covering all studies published to this date. All trials compared neonatal mortality in pregnant women receiving the intervention with those receiving the local standard care (Table 1). We included trials of both home-based counselling and facilitated women's groups delivered during pregnancy. Our starting point was two previously published reviews 5,10 of five trials of home-based counselling interventions, 13,14,26,27 and another five published between 2010 and 2013. 9,10,[15][16][17] One trial was excluded from the review as it was only quasi-experimental. 28 We also included a review published in 2013 6 covering seven trials of women's groups. [18][19][20][21][22][23][24] To identify the most recently published trials we conducted a literature search of the PubMed and Web of Science online databases using the following search string (((("newborn" OR "neonatal" OR "maternal")) AND mortality) AND trial), and identified trials of home-based counselling or women's group interventions published between January 2013 and May 2016 in low-or middle-income countries (Fig. 1). We screened 1481 titles and identified one additional cluster randomized trial that examined women's groups in rural eastern India. 25 Another identified trial 29 lacked a randomized design and was not included. Although they had been included in earlier metaanalyses by other authors, we excluded two non-randomized trials 26,27 from our meta-analysis after an assessment of the risk of bias.

Data processing
Two authors independently assessed the risk of bias (allocation concealment, and method of data collection for neonatal mortality data) for each study included in the review using the Cochrane Collaboration tool. 30 For each trial we extracted data on the study context (geographical area; baseline neonatal mortality rate; baseline proportion of births with infant breastfed immediately after delivery; baseline proportion of births in a facility); health system characteristics (. . .continued) Colbourn  All studies were cluster randomized trials comparing neonatal mortality in the population receiving the intervention with mortality in a comparison population receiving the local standard care.
b Immediate breastfeeding was defined in most studies as the percentage of births in which the infant was breastfed within 1 hour of delivery (mother's report), except Bhutta et al. 16 who defined breastfeeding within 30 minutes, and Tripathy et al. 19 who defined breastfeeding within 4 hours of birth. Facility birth was defined in all studies as the percentage of births in a health-care facility. Baseline was the value at the trial baseline in the intervention and comparison groups. Change was the change in values between the trial baseline and endline separately for intervention and comparison groups (the difference-in-differences).
c ThermoSpot™ (Camborne Consultants, Dorset, England) is a non-invasive liquid crystal indicator for hypothermia. d For trials that did not report on newborn practices at baseline and endline we did not calculate the difference-in-difference change but the simple difference between estimates from intervention and comparison group. e The intervention group included all 24 clusters with women's groups (with and without additional peer counselling). The comparison group included the 12 clusters without any intervention. However, the data on increases in breastfeeding and facility births were calculated with the comparison group of all clusters without women's groups as no data were available separately for the clusters with no intervention. Note: This in the trial area (number of nurses and midwives per 1000 population; number of health facilities per 100 000 population); and type of intervention (home-based counselling or women's groups). We also extracted data on the strength of the implementation (proportion of pregnant women visited in home-based counselling or attending women's groups). Not all the variables were reported in all trials. The data were obtained from the published papers and through contacting authors. One author extracted data, which were subsequently checked by another author. We performed all analysis in Stata, version 13.0 (Stata Corp, College Station, United States of America). The primary outcome for all studies was neonatal mortality. We also used immediate breastfeeding and facility births as tracer indicators for good newborn care practices. We calculated the changes in the proportions of women breastfeeding immediately after delivery and giving birth in a facility between baseline and endline separately for intervention and comparison groups (the differencein-differences). When baseline figures were not available, we calculated the differences between the intervention and control groups at endline.

Analysis
We used the metan command in STATA to compute forest plots calculating the RR for neonatal mortality for each study based on the number of deaths and births reported in intervention and comparison groups at the end of the trial period. Heterogeneity was assessed and I 2 and P-values were tabulated together with the summary estimates to provide measures of heterogeneity. We used the forest plots to examine patterns in the effect size on neonatal mortality according to the study context or health system characteristics in the trial area. We also investigated patterns in neonatal survival according to features of the implementation. For the analysis of associations between intervention characteristics and the effect size we chose equal-sized groups. For example, we categorized the 17 trials into three groups based on neonatal mortality rates in the trial area: very high mortality settings (≥ 44 deaths per 1000 live births), high mortality settings (33-43 deaths per 1000 live births) and moderately high mortality settings (≤ 32 deaths per 1000 live births).

Included trials
We included 15 articles 9,10,13-25 reporting 17 trials (two papers 13,22 reported two studies each). Eight studies reported interventions using home-based counselling and nine were interventions based on women's groups.
The trials took place in sub-Saharan Africa (Ghana, 1 trial; Malawi, 3 trials; United Republic of Tanzania, 1 trial) and in south Asia (Bangladesh, 4 trials; India, 6 trials; Nepal, 1 trial; Pakistan, 1 trial). All the trials were done in poor rural societies, except for the trial in an urban slum in India 21 (Table 1).
All packages aimed to improve home-based newborn care by mothers, such as immediate and exclusive breastfeeding, thermal care, and safe and dry cord care; the home care arm from one study 13 encouraged home treatment with antibiotics if referral was not possible (Table 1). Most trials reported coverage of these newborn practices as intermediate outcomes. Home-based behaviour change counselling involved visits to pregnant women at home by a community health worker or volunteer and sometimes also included community meetings. Women's participatory groups took place in the community and were facilitated by trained community members who used problem-solving methods, such as action cycles. Both approaches included education and be-haviour change communication to overcome challenges in health-care seeking and home newborn care practices.
All trials reported neonatal mortality as the main outcome, defined as the number of deaths in the first 28 days of life per 1000 live births in both sexes. Neonatal mortality data were obtained either from surveys or continuous surveillance in the target population. The trials were done in diverse contexts where the neonatal mortality rate ranged from 11 deaths per 1000 live births in an Indian urban slum 21 to 84 deaths per 1000 live births in India. 14 While the reported trials from Asia were from a period spanning the years 2001 to 2012, the reported trials from sub-Saharan Africa were from the years 2004 to 2013 (Table 1).

Context characteristics
We observed the largest reduction of neonatal mortality in settings with very high neonatal mortality. We calculated a 25% reduction in neonatal mortality   Fig. 2; Table 2). The pattern of the largest reductions in settings with the highest neonatal mortality was observed for both home-based counselling and women's group approaches (Fig. 3).
The pooled analysis suggested that the effects of the community approaches on neonatal mortality were stronger in the 12 pooled studies in south Asia (RR: 0.82; 95% CI: 0.78-0.86; P < 0.001 for heterogeneity), while there was no evidence of an effect in five studies in sub-Saharan Africa (RR: 0.95; 95% CI: 0.88-1.02; P = 0.193 for heterogeneity). None of the African studies, however, were done in a setting with very high neonatal mortality ( Fig. 4; Table 2).
The effect size of the community approaches was somewhat higher (RR: 0.77; 95% CI: 0.71-0.85) in pooled data from five studies (P = 0.001 for heterogeneity) where the baseline level of facility births was low (≤ 10%). The effect was lower when pooling six studies with 11-43% births in a facility (RR: 0.85; 95% CI: 0.80-0.91; P < 0.001 for heterogeneity) and five studies with ≥ 44% facility births (RR: 0.90; 95% CI: 0.83-0.97; P < 0.001 for heterogeneity; Fig. 6; Table 2).  16 who defined breastfeeding within 30 minutes, and Tripathy et al. 19 who defined breastfeeding within 4 hours of birth. b Baseline was the value at the trial baseline (in the intervention and comparison groups). c Facility birth was defined in all studies as the percentage of births in a health-care facility. d Change was the change in values between the trial baseline and endline separately for intervention and comparison groups (the difference-in-differences). e Percentage of pregnant women visited at home by a community health worker. f Percentage of pregnant women attending their local women's group.

Research
Community-based approaches for newborn health Claudia Hanson et al.
Only nine trials reported the health-system characteristics of facilities in the trial area. The meta-analysis suggested a lower effect of the community-based approaches on neonatal mortality in settings with more health facilities (Table 2). No evidence of community approaches was observed (RR: 0.95; 95% CI: 0.88-1.04) when pooling four studies (P = 0.121 for heterogeneity) with a density of > 9 facilities per 100 000 population. However, we found a 16% reduction (RR: 0.84; 95% 0.78-0.90) when pooling five studies (P < 0.001 for heterogeneity) in areas with ≤ 8 facilities per 100 000 population. Only six trials reported on the number of nurses and midwives in the area and we observed no difference in the effect on neonatal mortality in settings with higher or lower number of nurses and midwives per population ( Table 2).
We observed no evidence that the effect of the community-based approaches on neonatal mortality was associated with improvements in immediate breastfeeding and facility births. The analysis pooling five trials which achieved improvement in immediate breastfeeding of 25% or more suggested a reduction of neonatal mortality of 18% (RR: 0.82; 95% CI: 0.76-0.89; P < 0.001 for heterogeneity). Similarly, the pooled analysis of four trials achieving only marginal improvement (≤ 5%) in immediate breastfeeding suggested a 19% reduction in neonatal mortality (RR: 0.81; 95% CI: 0.74-0.89; P < 0.001 for heterogeneity; Table 2).
All home-based counselling interventions reached more than 40% of pregnant women and the size of the effect of the intervention on neonatal mortality did not differ in relation to the proportion of women reached. However, a difference was seen when running a sub-analysis of the women's group interventions. Pooling four trials that reached 37-66% of pregnant women we found a 26% reduction in neonatal mortality (RR: 0.74; 95% CI:  Table 2).

Discussion
Our analysis suggests that large gains in neonatal survival can be achieved using community approaches in settings with very high neonatal mortality and very low rates of facility births. Where mortality is lower, although still moderately high, no evidence of an effect of community approaches on neonatal mortality was found. The observed effect size of the community approaches was larger in south Asia, while there was no evidence of an effect when pooling the studies done in sub-Saharan Africa. This might be partly explained by the fact that the trials in Ghana, Malawi and the United Republic of Tanzania were done in settings with moderately high neonatal mortality.
The large effect of a 45% reduction of neonatal mortality which was previously reported 5 could be because these early trials were done in settings with high mortality and unhealthy home-care practices. Except in one trial, 25 subsequent meta-analysis 9,10 included trials done in places where neonatal mortality was considerably lower.
As neonatal mortality in an area decreases, the relative importance of infectious diseases and other more easily addressable risk factors, such as cold injuries, reduces. The latest work of the Global Burden of Disease group clearly highlights the increasing importance of intrapartum complications, including neonatal encephalopathy, as causes of death. 31 As non-infectious causes of neonatal mortality become more prominent, health system constraints to prevent intrapartum-related complications and mitigate the effect of prematurity might become more important. 32 One study concluded that part of the reason their intervention did not result in mortality reduction -despite improved neonatal care and facility coverage -was the failure to address birth asphyxia and prematu-rity. 15 A similar argument was raised by others reporting on community and participatory women's group approaches. 21,33 Lower neonatal mortality is likely to reflect recent or ongoing trends in health service uptake, household wealth education and health literacy. One study reported a decrease in neonatal mortality in both intervention and comparison groups, accompanied by increases in newborn care practices and healthservice uptake, suggesting underlying trends that had a larger impact than the trial intervention itself. 20 Others reported a doubling of facility deliveries during the trial period that was possibly due in part to increased transportation and better communications in the area. 9 A third study suggested that a reduction in neonatal mortality in both intervention and control groups was likely related to improvements in the living environment in the slum areas, such as covering gutters and better sanitation and electricity supplies. 21 These welcome investments in the health system and overall development reflect a rapidly changing context in which it is inherently more difficult to show

Mortality decreases Mortality increases
CI: confidence interval; RR: relative risk. a Baqui et al. 13 Note: For references with more than one trial, each trial is presented separately and denoted with a letter after the date.
large mortality reductions from specific interventions. 34 Many of the trials in our analysis reported only a very modest improvement in the numbers of women delivering in a health-care facility. This is in contrast to the most recent large increases in facility births observed in many low-and middle-income countries; these have occurred because of multiple factors, both within and outside their health systems. 35 Going forward, the strategies and content of community approaches to neonatal survival might need to be reexamined. Still many potential benefits of community approaches to enhancing health literacy, reducing delays in careseeking and improving linkages between the community and health facilities for emergency referral exist. Community approaches can also encourage accountability measures that could support facility strengthening. 36 However, the effect on neonatal mortality would depend on the quality of services available, and the two effects could not be separated.
Our approach of examining effects of community approaches in relation to context and health-system factors has to be interpreted with caution. We hypothesize that in settings with lower neonatal mortality, more facility births and improved newborn care practices, these interventions may have less effect. However, our stratified meta-analysis cannot prove such an association. Our findings are plausible against the background that the present community approaches target neonatal sepsis and complications of prematurity, while in a context of reduced neonatal mortality, intrapartum-related complications leading to asphyxia become more important. 37 Reducing intrapartum complications and birth asphyxia will demand quality intrapartum services which the present community approaches do not address.
We combined the two different approaches of home-based counselling and women's groups in our analysis, which strictly speaking prohibits any meta-analysis. Nevertheless, both approaches aimed to improve mothers' newborn care practices at home and health-seeking behaviours, and thus the mediators through which they affect neonatal mortality are expected to be similar. Our main aim was not to present summary estimates of the mortality effect to guide policy changes. Rather, we hope to contribute to the development of a theory underpinning the opportunities and limitations of community approaches and the role these approaches might play in the development of care packages to address neonatal mortality in the SDG era.
We computed summary baseline rates of immediate breastfeeding and facility births as well as percentage point changes due to the interventions. However, some authors did not include such information in their papers. While some reported baseline data, others only reported comparisons at endline. As a result the difference-in-difference between intervention and comparison groups could not be calculated. Thus improvements in newborn care practices are not adjusted for differences in baseline values between intervention and comparison groups for some studies. Moreover, our analysis was constrained by the lack of reporting on health system factors such as availability of health facilities or health providers. This reminds us of the value of a careful description of the context in which interventions are implemented to enable an understanding of the transferability of results.
In conclusion, the findings suggest that beyond a certain mortality threshold, community approaches alone might not lead to marked improvements in survival. This finding supports the recent trend in the SDG era towards increasing investment in the quality of facility care. ■

Mortality decreases Mortality increases
CI: confidence interval; RR: relative risk. a Baqui et al. 13 Notes: For references with more than one trial, each trial is presented separately and denoted with a letter after the date. Immediate breastfeeding was defined in most studies as the percentage of births in which the infant was breastfed within 1 hour of delivery, except Bhutta et al. 16 who defined breastfeeding within 30 minutes, and Tripathy et al. 19 who defined breastfeeding within 4 hours of birth.

Mortality decreases Mortality increases
CI: confidence interval; RR: relative risk. Notes: For references with more than one trial, each trial is presented separately and denoted with a letter after the date. Coverage of home-based counselling was the percentage of pregnant women visited at home by community health workers. Coverage of women's groups was the percentage of pregnant women attending their local women's group.