Menu
Menu
Menu
Menu
Objective: To assess whether the implementation of a results-based financing (RBF) project in The Gambia resulted in (1) improved national vaccination coverage (2) higher coverage in intervention than non-intervention areas, and (3) improvement in rural–urban coverage inequality. Methods: The study used a difference-in-differences design, based on repeated cross-sectional data from The Gambian Demographic and Health Surveys 2013 (N = 1650) and 2020 (N = 1456). Full vaccination (receipt of one BCG, 3 OPV, 3 DTP, and 1 measles-containing vaccine doses) and rural–urban vaccination inequality were our outcome variables. The intervention, RBF, was implemented in 5 of the 7 health regions. Covariates controlled for included child’s sex, child’s birth order number, socioeconomic status, ethnicity, distance from health facility, maternal education, mother’s age group, mother’s marital status, and mother’s work status. Poisson regression with robust variance was used to estimate whether coverage changed, and difference-in-differences and difference-in-differences-in-differences were used to ‘assess differences in vaccination coverage change and change in inequalities, respectively. Results: Total crude full vaccination coverage in The Gambia was 76% in 2013 and 84.6% in 2020. Overall vaccination significantly increased by 16% (95% CI: 9% to 24%) in 2020 compared to 2013, but with a smaller increase in intervention relative to non-intervention areas [PRR 0.88 (CI: 0.78–0.99)]. Rural-urban inequality in vaccination coverage decreased more – by 13% [0.87 (0.78–0.98)] – in RBF than non-RBF regions. Conclusion: Vaccination coverage improved over the study period though we have no evidence to ascribe the coverage gains to the RBF intervention. However, our study suggests that the RBF project has contributed to reducing rural–urban inequalities in the regions it was implemented.
The study employed a difference-in-differences design, based on secondary repeated cross-sectional data from the children’s datasets of The Gambia DHS conducted in 2013 [11] and 2019/2020 [17]. A binary variable, Results-based financing (RBF) status, was used to differentiate regions where RBF was implemented from non-RBF regions, with RBF regions coded as 1 (k = 5 regions) and non-RBF regions 0 (k = 3 regions). Pre-and post-RBF intervention periods, RBF year variable, were denoted by 0 (year = 2013) and 1 (year = 2019/2020) respectively. This resulted in four comparison groups for the difference-in-differences design, i.e., preintervention-RBF regions, preintervention-non-RBF regions, postintervention-RBF regions, and postintervention-non-RBF regions. For simplicity, the second DHS, 2019/2020, will be referred to as just 2020. Data collection for the 2020 survey was conducted from 21st November 2019 to 30th March 2020, before the first Gambian Covid 19 wave emerged in full (The first Covid-19 case in The Gambia was reported Mach 10th, 2020). A stratified two-stage selection procedure was implemented to select the DHS samples. The 2013 and 2020 DHS waves used enumeration areas from the 2003 and 2013 national censuses as sampling frames. The Gambia is divided into two municipalities and six local government areas (LGA), with the two municipalities considered entirely urban. The health system is divided into seven health regions. The six LGAs were stratified into rural–urban strata, resulting in 14 sampling strata. For the first stage, the areas within each sampling stratum were sorted by lower administrative units (districts and wards) to achieve implicit stratification. The average number of households per enumeration area was 68. Then a predetermined number of areas was then independently selected from each stratum using probability proportional to the estimated area size selection procedure. A total of 281 areas was selected in each survey [11], [17]. Following this, household listing exercises were conducted to update the number of households in the selected areas. Then came the second selection stage. In this stage, 25 households were selected from each area using equal probability systematic sampling [11], [17]. All women aged 15–49 years resident in selected households or who spent the night before the survey in the selected household were eligible survey respondents regardless of their residence status in the area. The number of eligible women interviewed, and response rates for the 2013 DHS and 2020 DHS are 10,233 and 11,865 and 90.7% and 95.1%, respectively. Relevant information, including the vaccination history of each child under five years born to an interviewed woman, was collected. For this study, all children 12–23 months were selected for inclusion because all children in this age cohort are expected to have received all the recommended basic vaccine doses [18]. There are 1660 children aged 12–23 months in the 2013 DHS [11] and 1456 in the 2020 DHS [17]. The proportion of children 12–23 months with vaccination cards was high in both surveys – 90.2% in 2013 and 93.2% in 2020 – with higher proportions in rural than urban areas. The RBF project was scaled up from 1 to 5 regions in 2016. The first DHS was conducted in 2013 [11], a year before the project’s start, and the second one in 2019/2020, three years after the project was extended to four more regions. The project buys predetermined quantity and quality indicators from health facilities. Quantity indicators are purchased per service delivered, whilst quality indicators are paid based on composite percentage scores attained by health facilities following a quality monitoring checklist. Vaccination performance is remunerated under the quality indicators category. Sixty percent of payments made to health facilities is earmarked for service delivery improvement, and the remaining 40% is shared amongst staff. Vaccination aspects monitored in the quality checklist include valid (doses which are age (and interval in case of multidose vaccines) appropriate doses administered; dropout rate; availability of recording, reporting, and monitoring tools; availability of job aids and adherence to standard operating procedures; availability, functionality, and maintenance of vaccine storage equipment; availability and storage of vaccines and all related supplies and; vaccine wastage [16]. These components of the immunization program are undoubtedly essential for the effective delivery of vaccination services. Full vaccination (aim 1 and 2) was defined as children 12–23 months who had received one dose of Bacillus Calmette Guerin vaccine, one dose of a measles-containing vaccine, three doses of the oral polio vaccine, and three doses of a diphtheria, pertussis, and tetanus-containing vaccine. Both vaccination history by card and recall were included. Rural-urban vaccination coverage inequality (aim 3) was operationalized as the disparity in full vaccination coverage between rural and urban areas in RBF regions and rural and urban areas in the non-RBF regions. The Gambia Bureau of Statistics’ designation of census areas as rural or urban was used as this was the rural–urban stratification used by the DHS surveys. Covariates were identified based on priori and their availability in the DHS data sets. Child’s birth order number was recoded into 1, 2&3, 4&5, and 6 or above, while her/his sex was considered as male or female. Mothers’ ages were grouped into 15–24 years, 25–29 years, 30–34 years, and 35–49 years and their marital statuses were defined as currently married or not currently married. Ethnicity was grouped as Wolof, Mandinka, Fula, others, and non-Gambians. Household socioeconomic status was generated from the household wealth quintile variable in the DHS data set. The household wealth quintile is a composite measure of relative household wealth created through principal component analysis using household ownership or access to materials including televisions and bicycles; materials used for housing construction; and types of water access and sanitation facilities [19]. We recategorized quintiles of household socioeconomic status into three groups of rich, middle, or poor by merging the richer and richest quintiles into rich and the poorer and poorest quintiles into to poor while leaving the middle category unchanged. A child’s mother’s work status was considered as either not working or informal work, or formal work, and their education was grouped as no education, primary or secondary and above. Distance to a health facility when in need of healthcare services was classified as a big problem or not a big problem based on the response of the caregiver interviewed. All analyses in this paper were performed using Stata software version 17 [20]. The “svy” command in Stata for survey data analysis was utilized in all analysis to account for the complex design (survey weights, clustering, and stratification) of the surveys. The analytical code is attached as supplemental material 1. In the first set of analysis, bivariate analyses were performed to estimate the frequency of full vaccination across exposure variables. We then estimated rural, urban, and total full vaccination coverage for the RBF intervention group, the control group, and The Gambia pre- (2013) and post-RBF intervention (2020). Following the set of analysis corresponding to the first aim, we utilized Poisson regression with robust variance to evaluate the crude (bivariate) and adjusted (including all covariates) relative change in overall vaccination coverage from 2013 to 2020. To address the second aim, we then evaluated whether there is a difference in changes in full vaccination coverage between RBF and non-RBF intervention sites using difference-in-differences (DiD) analysis in crude and adjusted analyses. Finally, corresponding to the third aim, we assessed the effects of RBF implementation on rural–urban disparities in vaccination coverage between intervention and control areas through a difference-in-difference-in-difference (DiDiD) approach in crude and adjusted analyses. We operationalized the DiD and DiDiD by generating variables that are equal to the product of the respective variables of interest. We used RBF implementation status and RBF year variables for the DiD and added residence to these two variables for the DiDiD analysis. We reported crude and adjusted Prevalence Rate Ratios (PRRs) and their 95% confidence intervals (CIs). Where we report p-values, we take statistical significance to be p ≤ 0.05. Since there is a difference in vaccination card retention rates between urban and rural areas and a difference in vaccination coverage among children with and those without cards, we conducted a sensitivity analysis on the potential effects of recall bias on our main results. We excluded children whose vaccination histories were obtained through caregivers recall and repeated our analysis, then compared the results with those including all eligible children (N = 3116). Please see the supplemental material attached. Demographic and Health Surveys are standard nationally representative household surveys conducted in developing countries to shed light on demographic and health trends across several dimensions. They are ethically cleared by ICF Institutional Review Board (IRB) and usually by IRBs of countries conducting the surveys [21]. Fieldwork for the two surveys in The Gambia was conducted by trained data collectors who interviewed respondents only after obtaining their informed consent. Anonymized DHS datasets are publicly available via the DHS program website https://dhsprogram.com/data/available-datasets.cfm [22].
