MWHs have existed in Zambia for decades with generally low quality and no specific policy to keep them at a particular standard . The Maternity Home Alliance (MHA), a collaboration of two implementing partners, two academic partners, and the Government of Zambia implemented MWHs using a Core MWH Model with specific standards and policies . The MWH parent study was conducted in seven primarily rural districts: Nyimba, Lundazi, Choma, Kalomo, and Pemba, Mansa and Chembe. Characteristics of these districts as well as the core MWH model figure are thoroughly explained elsewhere (20).
One implementing partner (Africare-Zambia), operating in Lundazi, Mansa, and Chembe districts, also implemented SILCs from the beginning of January 2016, within their MWH intervention sites. By the end of October 2017, there were more than 310 active SILCs with 6711 participants from the 10 different communities with the core MWH model. The core MWH models were implemented between June 2016 and August 2018 .
Of the seven districts included in the overarching parent study, Kalomo, Mansa, Nyimba, and Lundazi were part of the first phase of Saving Mothers Giving Life (SMGL) initiative . SMGL is a 5-year initiative that was implemented from 2012 to 2016 as a multi-lateral initiative to reduce maternal and newborn mortality . The SMGL approach included a variety of interventions such as training community health workers responsible for improving the knowledge and access to RHSs within their local communities, and mentoring health facility staff to increase quality of care, improving the referral system, and investing in supply chain and facility equipment [10, 25]. The baseline Household Survey (HHS) data were collected in April and May of 2016, overlapping with the SMGL initiative which ended December of 2016 .
A secondary analysis was conducted on two cross-sectional samples of recently delivered women surveyed at baseline (March to May 2016) and endline (August to September 2018) for the MHA impact evaluation. MWHs aim to improve maternal and neonatal health outcomes for the most rural women, who live far from health services by increasing access to facility-based delivery services with a SP . The MHA evaluated the impact of MWH on RHS access, assessed primarily through delivery at a HF. Both baseline and endline HHS data were collected from the communities surrounding 40 rural health centers in seven rural districts of Zambia. Each community had at least one health center capable of managing basic emergency obstetric and neonatal complications (BEmONC) where the core MWH model was implemented nearby . The MWH core model was implemented in 20 of the communities and the remaining 20 communities were used as a control, with a health facility present but no MWH model implemented. The details of the MWH parent study design and data collection process are described elsewhere [20, 21].
Written informed consent was sought from the original study participants and this study was conducted using the de-identified dataset. Ethical approvals for the MWH project were obtained from the authors Institutional Review Boards (IRBs), as well as from the ERES Converge Research IRB, a private local ethics board in Zambia.
The parent study used a multistage random sampling procedure for both baseline and endline HHS data (goal of 2400 women) with a probability for village selection proportionate to population size . A household was defined as a group of people who regularly cook together. HHS data were collected from two cross-sectional samples within the sample villages at baseline and endline. Eligibility criteria for women to participate in the HHS included: 1) delivered a baby within the past 12 months, 2) 15 or older (if aged 15–17, a legal guardian had to consent), and 3) resident of the community identified for sampling. If the women who gave birth was deceased, a proxy participant who is 18 or older, took the HHS . To capture the community level changes, different women from the same community were followed at baseline and endline.
The total sample was separated into three CGs: CG1) communities with neither the core MWH model nor SILC (20 communities), CG2) communities with only the core MWH model (10 communities), and CG3) communities with both the core MWH model and SILC (10 communities). All communities included in the study had a BEmONC health facility.
Of the 2381 participants from baseline HHS, 1031participans were from CG1, 597 participants from CG2, and 756 participants from CG3. Of the 2330 participants from endline, 1113 participants were from CG1, 610 participants from CG2, and 598 participants from CG3.
Our primary outcomes of interests are: 1) household wealth, 2) financial preparedness for birth, and 3) utilization of RHSs. Variables for demographics, household wealth, saving for delivery, and utilization of RHSs were extracted from a de-identified HHS dataset.
Demographic variables included women’s age, marital status, number of pregnancies, number of livebirths, and education level.
Household wealth was assessed by using the comprehensive list of wealth indicator variables. A total of 57 dichotomized variables included ownership of household assets and quality of housing and water supply that are similar to the variables used in the Demographic and Health Survey (DHS) . Principal component analysis (PCA) was used to assign weights to each of the wealth indicator variables, summed, and created into quintiles – poorest, poor, middle, rich, and richest [26, 27]. PCA is a data reduction procedure where a set of correlated variables are replaced with a set of uncorrelated variables representing unobserved characteristics of the sample . Therefore, wealth indicator variables that are more unequally distributed across the sample will have higher weight. While PCA has its own limitations, using PCA to develop wealth quintiles is one the most frequently used methods by the World Bank and is used in more than 76 countries [26, 27]. We excluded observations that was missing any of the 57 wealth indicator variables and created the wealth quintiles twice, once for the baseline sample and once for the endline sample. This allowed us to understand the wealth distribution between the CGs at baseline and endline.
Financial preparedness for birth was determined by whether women saved any money for their most recent delivery or not.
Utilization of RHSs was examined by the number of ANC and PNC visits, utilization of MWH, HF, and SP delivery. The five variables were dichotomized as ‘utilized’ versus ‘not utilized’. Women who attended four or more ANC contacts were categorized as ‘utilized’ for ANC visits. Even though the 2016 WHO ANC model recommends a minimum of eight ANC contacts, the guideline was not yet widely implemented in rural Zambia . Therefore, the previous guideline of four or more ANC visits was used for the analysis. Similarly, if a woman attended all four PNC visits, first within 24 hours of delivery, second within 3 days postpartum, third between 7 and 14 days postpartum, and fourth before 6 weeks postpartum, she was categorized as having utilized PNC visits . If a woman stayed at a MWH at any point of her pregnancy, she was categorized as having a MWH. If a woman delivered her most recent baby at a health post, HF, or a hospital, she was categorized as having utilized a HF and if she delivered with a doctor, clinical officer, nurse, or midwife she was categorized as having delivered with a SP. Each of the RHSs variables were examined individually.
One may argue that utilization of MWHs often increases delivery at HF with SP, and that delivery at HF and delivery with SP are interchangeable. However, because of the limited number of SP, women delivering at a HF does not always lead to delivery with SP [31, 32]. Similarly, in many sub-Saharan African countries, SP travel to women’s homes for delivery in cases of emergency, which means that sometimes women can deliver with a SP without delivering at a HF . Hence, both variables were included as part of the utilization of RHSs.
To compare the changes in the outcome variables over time between the communities that had access to SILCs and those that did not, interaction effects of the stratified CGs and timepoints (baseline versus endline) were used. This study hypothesized that women from CG3 compared to women from CG1 and women from CG2 will have higher household wealth, higher likelihood to be financially prepared for birth, and higher utilization of RHSs – ANC visits, PNC visits, MWH, HF delivery, and SP delivery – at endline.
Descriptive statistics were analyzed with the means and standard deviation (SD) provided for both the baseline and endline samples as well as the stratified sample between the CGs at baseline and endline. A set of Chi-square tests of independence and independent sample t-tests were implemented to examine the differences in demographic and outcome variables between the baseline and endline participants and participants from the three CGs at baseline and endline.
Interaction effects of CGs and timepoint (i.e., baseline versus endline) were used to assess the relationships between the independent and dependent variables since CGs and timepoint combined have an effect on each of the dependent variables. Linear or logistic regression models without the interaction effect assumes that the effect of each independent variable on the outcome is separate from the other independent variable in the model. Hence, using the interaction effects of CGs and timepoint on outcome variables provides a more accurate understanding of how the inclusion of SILCs in communities influences wealth and maternal health. Key outcome variables were 1) household wealth (wealth index), 2) financial preparedness for birth (saving for most recent delivery), and 3) utilization of RHSs (ANC visits, PNC visits, MWH utilization, HF delivery, and SP delivery). All adjusted models included age, marital status, number of pregnancies, number of live births, and education level. Wealth was also added to the adjusted model when exploring financial preparedness for birth and utilization of RHSs. All analyses accounted for the clustering at the community level by using the vce(cluster) command in Stata. In addition, coefficient (b), standard error (SE), adjusted odds ratios (AORs), and 95% confidence intervals (95%CI) were provided. All statistical analysis was conducted in Stata 17.0 (StataCorp, College Station, TX, USA).