What is the role of community capabilities for maternal health? An exploration of community capabilities as determinants to institutional deliveries in Bangladesh, India, and Uganda

Background: While community capabilities are recognized as important factors in developing resilient health systems and communities, appropriate metrics for these have not yet been developed. Furthermore, the role of community capabilities on access to maternal health services has been underexplored. In this paper, we summarize the development of a community capability score based on the Future Health System (FHS) project’s experience in Bangladesh, India, and Uganda, and, examine the role of community capabilities as determinants of institutional delivery in these three contexts. Methods: We developed a community capability score using a pooled dataset containing cross-sectional household survey data from Bangladesh, India, and Uganda. Our main outcome of interest was whether the woman delivered in an institution. Our predictor variables included the community capability score, as well as a series of previously identified determinants of maternal health. We calculate both population-averaged effects (using GEE logistic regression), as well as sub-national level effects (using a mixed effects model). Results: Our final sample for analysis included 2775 women, of which 1238 were from Bangladesh, 1199 from India, and 338 from Uganda. We found that individual-level determinants of institutional deliveries, such as maternal education, parity, and ante-natal care access were significant in our analysis and had a strong impact on a woman’s odds of delivering in an institution. We also found that, in addition to individual-level determinants, greater community capability was significantly associated with higher odds of institutional delivery. For every additional capability, the odds of institutional delivery would increase by up to almost 6 %. Conclusion: Individual-level characteristics are strong determinants of whether a woman delivered in an institution. However, we found that community capability also plays an important role, and should be taken into account when designing programs and interventions to support institutional deliveries. Consideration of individual factors and the capabilities of the communities in which people live would contribute to the vision of supporting people-centered approaches to health. The FHS project research teams have begun implementing their interventions in 2011. The teams began their work under the cross-cutting “Unlocking Community Capabilities” theme by developing approaches to measure community capability, in order to understand social relations and resources within and across the communities they worked in [5]. In each setting, a baseline, cross-sectional household questionnaire was conducted in 2012, to which the teams added a module of questions to assess the presence or absence of key community capability. Each of the survey components are explained in greater detail below. Sampling and study area description: For this article, we pooled survey data collected from Bangladesh, India, and Uganda. In Bangladesh, all of the data was collected from the Chakaria Health and Demographic Surveillance System (CHDSS). The CHDSS includes 20,036 households that are eligible to participate in the survey. The UCC questionnaires and the household data were collected from the same sample, but at two different points in time. For the UCC survey, the team selected only one household member from each household, randomly, using a sampling fraction of 400/population of the age group with the lowest population size, for men and women separately. The total number of individuals selected was 5152 (2188 men and 2964 women). Data on safe motherhood practices was collected from mothers of infants through three monthly household visits, as part of the CHDSS [25]. The data from the CHDSS and the UCC questionnaire were combined and matched by household ID. The final sample was comprised of 1238 women of reproductive age (15–49 years old), whose last child was born with the past 5 years. In India, data on community capability, maternal and child health and demographic questions were collected as part of the baseline survey. The survey included 1200 households in both deltaic and non-deltaic regions of the Patharpratima Block in West Bengal, India. The sample was selected using probability proportionate to size (PPS), where the sampling unit was the village (equivalent to a community). Thirty out of 87 villages were selected. Within each village, households were selected at random, based on whether any children under 5 lived in a particular household. For the purpose of this analysis, the data on institutional deliveries reflects the information provided by mothers about their most recent birth. The final sample consisted of 1199 women of reproductive age (15–49 years old). In Uganda, data on the community capability were collected as a cross-sectional study annexed to the household baseline survey for MAternal and NEwborn care practices STudy (MANEST) among women who had delivered within 1 year prior the survey from 3 districts. For the community capability data, 369 household heads (both men and women) out of 2011 households selected for the MANEST survey were selected across 3 districts in the North-Eastern Part of Uganda. One parish from each of the 17 sub-counties that make up the 3 districts were randomly selected and 369 household heads out of the 2011 households were randomly selected to complete the community capability questionnaire. The data from the community capability questionnaire was linked to the MANEST data using a unique household identifier in order to obtain the institutional delivery (main outcome) and the individual level predictor variables used in this analysis. In all countries, the sample was restricted to women of reproductive age (15–49 years old), whose last child was born in the past 5 years. Outcome variable: The outcome of interest in this paper was “institutional delivery”, whether or not a woman delivered in an institution. In the survey, women were asked “Whether or not a woman delivered in an institution” (Bangladesh), “Where was the child born” (India), or “Where did you deliver from” (Uganda). The delivery counted as an institutional delivery if the child was born at a hospital, regardless of ownership by government or non-government sector. The delivery was not counted as institutional if the child was born at home or elsewhere (e.g. with a traditional birth attendant or en route to the hospital or health facility). The FHS team developed a series of quantitative community capability questions based on a thorough literature review conducted at the beginning of the project (see Additional file 1). These questions were intended to be used across countries, in household surveys, exit surveys, or any other quantitative surveys, though they could also be used in mixed methods research. The community capability questions developed by the FHS team spanned several conceptual domains: community/village assets (inclusive of 9–13 services, such as schools, that were offered in a community), community organizations (examining both the general existence of organizations, as well as the extent to which community members participate and/or benefit), civic voice actions (e.g. voting in a local election), community coherence and decision-making (e.g. about commitment to collective goals), and health system problems (e.g. absence of doctors). Because these concepts are multi-dimensional and highly contextualized, each of the country teams had the flexibility to select the domains and questions that were most relevant to their research projects. For the analysis presented here, we selected to use only the domains and questions that were common across all of the three countries, in order to ensure that the same level of details was available. The common domains specifically considered for this analysis include: community/village assets (the physical and organizational resources of a community to which the community members should have access, and the ability of communities to mobilize resources for collective use), group participation (the community capacity to engage its members in collective action, and the degree to which members are active in group functions), and community cohesion (the forces that act on members of a community to retain and actively contribute to the community or the degree to which members want to be part of a group and are loyal and united in pursuit of group goals). A total of 13 common community capability variables were identified and are listed in Table 1. While the original survey recorded respondents’ answers on a 5-point Likert scale (except in India, where community coherence was measured on a 4-point Likert scale), for the purposes of the analysis reported here, we dichotomized answers (i.e. 0 = if Likert scale denoted poor structures or disagreement (including here the mid-range value of the scale) and 1 = if Likert scale denoted fair or better structures or agreement). Summary of common community capability variables Individual level predictor variables: For this analysis, we extracted information that was found in all three data sets, such as maternal age, education, parity, and antenatal care participation. In the pooled dataset, maternal age was classified as: 15–24 years old, 25–34 years old, and 35–49 years old. Maternal education was categorized as no formal education; primary education; lower secondary education; higher secondary education; and university or higher. Parity was categorized as one child, two to four children, and 5 or more children. Antenatal care participation was classified as: none; one or two visits; and three or more visits. Statistical analyses: All analyses were performed using Stata 14 [26]. In the first phase, we used exploratory factor analysis to identify the key factors behind community capability across the three country setting. We selected factors if they had eigenvalues exceeding 1.0, verifying the percent variance explained, cumulative percent of variance explained, and their scree plots [27]. Our main factor analysis was based on polychoric correlations, an analysis which is best suited for data that are not normally distributed, and we used varimax rotation to facilitate interpretation of identified factors. Three factors had eigenvalues >1, with and the loadings for each variable are summarized in Table 2. Internal consistency reliability was measured with the Cronbach’s alpha coefficient, for which a value of 0.7 is considered acceptable [28]. Factor loadings for community capabilities items from the pooled dataset The Cronbach’s alpha coefficient was 0.68, suggesting that the reliability of our scale was marginally acceptable. The pooled data showed that the community capability variables grouped into logically cohered latent components. Because all community capability variables represented essential aspects of community capability and for ease of interpretation, we developed a simple summative and unweighted community capability index score. We generated at the individual respondent level as the number of items declared per respondent out of all the 13 possible items. Given the limited number of items, we chose a single community capability score rather than three separate scores. The score we obtained was then translated into percentages, so that the final variable was a continuous variable from 0 to 100. We tried to aggregate this score at the community level, commonly identified across the surveys as the village level. However, this was not possible due to very small sample sizes. In the second phase of the analysis, we combined the community capability score with the household survey data. Individual-level determinants of institutional deliveries included maternal age, education, parity, number of visits to ante-natal care, and country of origin. The community-level determinant of institutional deliveries was the community capability score. We produced descriptive summary statistics for all of the variables of interest. We further conducted bivariate logistic regression analyses and multi-collinearity tests, to verify which variables to include in our model. Our final marginal model consisted of a logistic regression using generalized estimating equations (GEE) with institutional delivery as the outcome using the community capability score and demographic measures as explanatory variables [29]. The GEE model allows for non-independence in responses to produce a population-averaged or marginal model. Therefore, the marginal model allows the estimation of the odds of institutional deliveries as averaged over the entire sampled population. We further ran a series of mixed effects models, using the generalized linear latent and mixed models (GLLAMM) to identify the contribution of community level effects [30]. The GLLAMM method helped us to account for the intra-class correlation (ICC) between multiple observations within the same geographic areas. The mixed effects model we estimated with GLLAMM allows us a deeper exploration and quantification of the within sub-national unit (cluster) variability. Thus, it allows us to investigate the cluster-specific odds of institutional deliveries. Our final mixed effects model contained two levels, which assumed that households were nested within sub-national units. Because our data did not contain sufficient sample size at the village level, the sub-national unit was selected to represent the union level in Bangladesh; the sub-county level in Uganda; and the gram-panchayat in India. In addition to producing the odds ratio of interest and their respective confidence intervals, we also conducted an analysis of the various variance estimators. For the mixed effects model we calculated the intra-class correlation coefficient, which is a measure of the proportion of the total variation attributable to variation between clusters. We also explored the goodness of fit of these models.