Effect of a participatory multisectoral maternal and newborn intervention on maternal health service utilization and newborn care practices: A quasi-experimental study in three rural Ugandan districts

Background: The MANIFEST study in eastern Uganda employed a participatory multisectoral approach to reduce barriers to access to maternal and newborn care services. Objectives: This study analyses the effect of the intervention on the utilization of maternal and newborn services and care practices. Methods: The quasi-experimental pre- and post-comparison design had two main components: community mobilization and empowerment, and health provider capacity building. The primary outcomes were utilization of antenatal care (ANC), delivery and postnatal care, and newborn care practices. Baseline (n = 2237) and endline (n = 1946) data were collected from women of reproductive age. The data was analysed using difference in differences (DiD) analysis and logistic regression. Results: The DiD results revealed an 8% difference in early ANC attendance (p < 0.01) and facility delivery (p < 0.01). Facility delivery increased from 66% to 73% in the intervention area, but remained unchanged in the comparison area (64% vs 63%, p < 0.01). The DiD results also demonstrated a 20% difference in clean cord care (p < 0.001) and an 8% difference in delayed bathing (p < 0.001). The intervention elements that predicted facility delivery were attending ANC four times [adjusted odds ratio (aOR) 1.42, 95% confidence interval (CI) 1.17- 1.74] and saving for maternal health (aOR 2.11, 95% CI 1.39-3.21). Facility delivery and village health team (VHT) home visits were key predictors for clean cord care and skin-to-skin care. Conclusions: The multisectoral approach had positive effects on early ANC attendance, facility deliveries and newborn care practices. Community resources such as VHTs and savings are crucial to maternal and newborn outcomes and should be supported. VHT-led health education should incorporate practical measures that enable families to save and access transport services to enhance adequate preparation for birth. This study employed a quasi-experimental pre- and post-comparison study design. It was implemented in the districts of Kamuli, Kibuku and Pallisa in eastern Uganda, with a total population of 1,075,242 in 2014 [30]. This population mostly practises subsistence farming, crop farming, petty trading and small-scale animal rearing. The whole of Kibuku district was an intervention area, because it has only one administrative zone, referred to as a health subdistrict. Kamuli and Pallisa have three administrative zones, and so one health subdistrict in each of these two districts was selected as an intervention area and one as a comparison area. The district team selected the intervention and comparison areas. The selection was purposive and determined based on maternal and newborn service indicators for the district. The health service infrastructure comprised a total of 104 health facilities, 33 in Pallisa, 17 in Kibuku and 54 in Kamuli. The project had two main components: a community mobilization and empowerment component to stimulate demand for services, and a health provider and management capacity-building component to strengthen the delivery of quality maternal and newborn health services. The community mobilization and empowerment component comprised several strategies, including: (1) home visits by CHWs, also referred to as village health teams (VHTs); (2) health education through radio spots, talk shows and quarterly community dialogues; (3) promotion of saving through savings groups and other methods; and (4) promotion of partnerships with local transporters to ease geographical access to care. The capacity-building component included: (1) emergency obstetric and newborn care refresher training; (2) mentorship and support supervision of primary health workers; (3) a certificate course in health services management for health managers and a postgraduate diploma in project planning and management for district health officers; and (4) recognition of best performing facilities and managers. This supply-side package of interventions aimed to improve the skills of health workers in the provision of maternal and newborn care services, in addition to improving skills in leadership for maternal and newborn health care, to provide an enabling environment for service delivery. A detailed description of the intervention is provided in the design paper that is part of this supplement [29]. This intervention was provided in line with Susman’s participatory action research approach [31]. This approach comprised five main stages: (1) diagnosing, during which problems are identified; (2) action planning, during which alternative courses of action are considered and the best options selected; (3) taking action, during which selected courses of action are implemented; (4) evaluation, during which the actions taken and consequences are evaluated; and (5) specifying learning, during which key lessons are identified. Tetui et al. [32] provide a detailed description of the participatory approach used in this paper. The primary outcomes for this paper were early ANC attendance (defined as ANC attendance in the first trimester); attending ANC at least four times; delivery in a health facility; postnatal care (PNC) attendance within 6 weeks; and newborn care practices, such as clean cord care (putting nothing on the umbilical cord), delayed bathing (bathing the newborn 24 h after birth) and skin-to-skin care. The independent variables included VHT home visits (visits by a VHT at home while pregnant or after delivery); community dialogue meeting attendance; receipt of health education about maternal and newborn health on the radio; saving for maternal health (saving money to meet maternal health-related needs); wealth (measured using a wealth asset index); and sociodemographic characteristics such as age, gender, marital status, educational level and occupation. The wealth quintiles were generated using principal components analysis based on the information collected on assets and household structure. The sample size was determined using a two-sided Z test of the difference between proportions (Equation (1)) with 80% statistical power, a 5% significance level, 1.5 design effect and a non-response rate of 10%. The major quantifiable outcome of the study used in the calculation of the sample size was the proportion of women who delivered in a health facility with a skilled provider. We therefore assumed that after 3 years (2013–2015) of implementation, skilled deliveries would increase from 38% to 58%, from 62% to 72% and from 68% to 78% in the intervention areas of Kibuku, Pallisa and Kamuli districts, respectively [21]. The assumptions resulted in a sample size of 2293 women. A two-stage sampling technique was applied per district for each of the study areas. We estimated that we required 119 villages to realize our sample size. Therefore, 52 out of 514 villages were selected for Kamuli, 46 out of 346 for Pallisa and 21 out of 244 for Kibuku using probability proportionate to size sampling techniques. Thereafter, all households were listed to identify eligible study participants. During listing, 3456 and 3199 women were identified as having delivered in the 12 months preceding the baseline and endline, respectively. The inclusion criteria comprised all women of reproductive age, who were residents and had delivered in the past 12 months, irrespective of birth outcomes (only pregnancies which lasted at least 28 weeks were considered). Women aged less than 18 years who met the inclusion criteria and provided informed consent were included as emancipated minors. Women who were severely ill at the time of the survey and those who had not lived in the community for at least 1 year were excluded from the study. Of the women listed in the 119 villages, a total of 2237 (1101 in the comparison area and 1136 in the intervention area) were interviewed during the baseline survey and 1946 during the endline (920 in the comparison and 1026 in the intervention). A detailed description of the data collection methods has been presented in the design paper [29]. The data were collected using interviewer-administered structured questionnaires in 2013 and 2015. The questionnaires were translated into local languages used in the respective districts to obtain data from the study participants in a language easily understood by them. Before data collection, the tools were pre-tested and adjusted according to the suggestions made by the pre-testing team. The data collection team comprised 24 research assistants, two editors and two field supervisors. They were trained and divided into two teams. All members of the data collection team were fluent in the local language and had completed secondary level education. The information collected included information about sociodemographic characteristics, places of birth, number of ANC and PNC attendances, pregnancy gestational age at the first ANC visit, birth preparedness, area of residence, newborn care practices, home visits by CHWs and participation in community dialogue meetings. A data collection manual outlined the procedures to be followed during data collection, storage and entry. To ensure that the data were collected accurately, the field supervisors reinterviewed randomly selected respondents, while the data editors checked for errors in the data collection forms. Any errors identified were verified and corrected immediately by the field staff. In addition, an independent quality control team visited the field every week to ensure that the data were being collected according to the set protocol. The data were entered into Epi info 7. To check the consistency of data entry, 10% of the questionnaires were double entered. The entered data were transferred into STATA 13.0 for analysis, and backed up. Descriptive statistics of the independent and dependent variables are presented using frequencies. Difference in differences (DiD) analyses (Equation (2)) were used to understand the contribution of the intervention package towards health facility utilization and maternal and newborn care practices. The treatment and time variables were dummy variables: 1, treatment group; 0, non-treatment group; and 0, before intervention; 1, after intervention, respectively. yis represents the study outcomes, which included health facility delivery, ANC attendance, PNC attendance and newborn care practices. β 3 is the DiD estimator that tells us whether the expected mean change in outcomes before the intervention and after the intervention were different in the intervention and control groups. xit represents covariates such as age, education and occupation, while λi represents the covariates’ estimators. We ran the model separately for each of the study outcomes by considering all the covariates that we thought had an effect on the outcome variables. A significant coefficient of the interaction term implies that the outcomes differed by group over time. Multivariate analysis was performed using logistic regression to understand the predictors of the study outcomes (newborn care practices, early ANC attendance, fourth ANC attendance and health facility delivery). We performed univariate analysis using ulogit command in STATA to seek the likelihood of covariate variables in affecting the study outcomes. Variables with p values ≤ 0.25 were considered for multivariate analysis. Multicollinearity was assessed using the collin command in STATA, where variables with large values of the variance inflation factor (> 2.0) were considered as strongly correlated factors and subsequently dropped from the final model. Hosmer–Lemeshow and Pregibon tests were used to test the goodness of fit of the model. A model was considered a good fit if the linktest (hatsq) under Pregibon’s test and p value under the Hosmer–Lemeshow test were non-significant. We introduced interaction terms between the VHT and area of study, and between saving for health and study area to assess how the VHT home visits and saving for health affected health utilization differently in the intervention and comparison areas. Similarly, we introduced the interaction between health facility delivery and study area to assess how health facility delivery affected newborn care practices differently in the intervention and comparison areas.