In this study, we examined disparities in PHC access in four conflict-affected fragile states. We defined PHC access as the absence of both the geographic and financial barriers to care, based on the PHCPI measurement framework and vital signs profile [10]. For each studied country, disparities in geographic and financial access to care were compared across education and wealth strata in areas with differing levels of conflict intensity; specifically, access disparities reported by women living in neighborhoods (household clusters) with medium or high-intensity conflict were compared to access disparities reported by women living in neighborhoods with no or low-intensity conflict. To define conflict intensity at the household cluster level, organized violence events and their associated fatalities were geographically linked to DHS household clusters' location. Three measures of disparities were computed, and each was compared based on the intensity of conflict the year the DHS data was collected (Cameroon 2018, DRC 2013, Mali 2018, Nigeria 2018).
Selection of the studied contexts
In 2020, the World Bank identified 39 fragile and conflict-affected situations [16]. To systematically select countries for this study, we applied the following inclusion criteria to the specified list: 1) the country had a DHS survey in the past ten years (2010-2020) to ensure the availability of data on access to PHC using standardized indicators; 2) there was an ongoing armed conflict the year the household survey was conducted, with data on violent events and conflict-related deaths available in the UPPSALA Conflict Data Program (UCDP) database. Armed conflict was defined as the presence of at least 25 conflict-related deaths per year [17] ; and 3) the geographic locations of DHS household clusters were publicly available to allow for spatial analysis, including linkage with armed conflict location data. Applying these criteria yielded five countries: Cameroon, the Democratic Republic of Congo (DRC), Mali, Myanmar, and Nigeria. Since this study does not aim for geographic comprehensiveness, but rather the detection of common patterns of disparities across conflict-affected fragile states, we excluded Myanmar from the analysis as it was the only country belonging to a geographically different context.
Data sources
The DHS was used as the source of information on the geographic and financial barriers to care as recommended by the PHCPI methodology [18]. DHS is a nationally representative household survey and a vital source of information on population, health, and nutrition indicators in more than 90 countries [19]. The standard DHS survey is conducted every five years using a large sample size (5000-30,000 households per survey). Based on the availability of census information, most countries apply a two-stage stratified sampling technique. The first stage includes selecting enumeration areas (EA) or clusters with a probability proportional to EA size. An equal probability systematic sampling strategy is then applied in the second stage to draw a fixed number of households per cluster. Each survey can comprise various research tools, including multi-module questionnaires, geographic information collection, and occasional biomarkers collection. The survey duration typically ranges between 18 and 20 months [19]. In the four studied contexts, a two-stage stratified sample was conducted except for in new provinces [20] and some parts of established provinces in DRC, where a three-stage sample was used. Table S.1 provides more details on the characteristics of the demographic health surveys included in the analysis [see additional file 1]
Conflict data were obtained using the UCDP database [17, 21]. UCDP is the primary global source for data on armed conflict and organized violence. UCDP's definition of armed conflict became the international standard allowing for systematic analysis of temporal trends and cross-country comparisons. The unit of the analysis in the UCDP database is an 'event' - an instance of fatal organized violence defined as: "The incidence of the use of armed force by an organized actor against another organized actor, or against civilians, resulting in at least one direct death in either the best, low or high estimate categories at a specific location and for a specific temporal duration." [17]. Each event meeting the former criteria is recorded as one line in the database. Events with uncertain information on the number of fatalities or those with no reported deaths are excluded. [17, 21]. In this study, we extracted and analysed all events satisfying the UCDP definition of an event of organized violence.
Metrics
Access metrics
We analysed both geographic and financial access to PHC services using the DHS question on perceived barriers to care by interviewed women [19]. Perceived barriers due to distance were defined as the percentage of women aged (15-49) years who report specific problems in accessing care when they are sick due to the distance travelled for treatment. Similarly, perceived barriers due to treatment costs were defined as the percentage of women aged (15-49) years who report specific problems in accessing care when they are sick due to issues related to getting money for treatment. A PHC access index score was then computed based on the PHCPI methodology [22] as the average of not perceiving geographic barriers and financial barriers to care as follows:
Disparities metrics
We used the DHS educational status and wealth index variables to define the educational and wealth disparities respectively:
The DHS defines an educational status variable (v106) as the highest level of education attended but not necessarily completed. It is further subdivided into the following categories: no education, primary, secondary, and higher than secondary [23]. Such classification may vary by country, but the standard classification has been consistently reported in the four-studied countries. In this analysis, we define educational disparity as the difference in access to PHC services among women with varying education levels.
The DHS wealth index (v190) is a composite measure that gives a general idea of living standards based on household access to water and sanitation, ownership of certain assets such as TV, bicycles, and household construction material. The index is calculated at the household level using a standardized score for each asset. The individuals are then ranked based on the household's total score and divided into five population wealth quintiles: lowest, second, middle, fourth, and highest [23]. In this analysis, we define economic or wealth disparity as the difference in access to PHC services comparing women with varying wealth quintiles.
Data analysis
We geographically and temporally linked household clusters to organized violence events located within a 50-km distance from the centroid representing the cluster location. The size of the buffer zone was decided based on previous studies examining the effect of armed conflict on maternal and child health outcomes [24, 25]. Conflict intensity was defined as a binomial variable with "medium or high" conflict intensity = 1 and "no or low" conflict intensity = 0. A cut-off of more than two conflict-related deaths per 100,000 population (according to the total number of fatalities best estimate) per household cluster population was used to define medium or high-intensity conflict. This cut-off point was selected based on the World Bank definition of low-intensity conflict [26]. The total population size per cluster, according to 2015 estimates, was used as a reference point for classifying conflict exposure in each studied context. The 2015 estimates were selected as they were the closest estimates of cluster size in the DHS environmental database in the four studied countries. In Nigeria, 13 clusters in Borno state and one cluster in Yobe state were excluded from the analysis due to the lack of information on the total population size. Additionally, we excluded one cluster in the extreme north region in Cameroon with available environmental covariates but no corresponding DHS health variables.
For each health indicator, three measures of disparities were computed: an absolute measure of inequality, a concentration index, and a multivariate logistic regression coefficient. Several studies recommended the use of multiple measures while addressing health disparities [27-33]. For example, Sully et al. [31] highlighted the importance of incorporating relative, absolute, and population impact measures to understand inequalities. Similarly, Alonge et al. [32], in their review of the utility and limitations of disparity measures, concluded that there is no perfect measure of disparity, and each quantifies some aspect of health disparity. They also highlighted the importance of combining measures for a more comprehensive evaluation of health programs. The same conclusion was reached by Houweling et al. [33], who recommended combining both the relative and absolute measures of inequalities while considering the overall level of the outcome.
In this analysis, we also viewed the three measures of disparities as complementary rather than alternatives, each contributing to one aspect of disparity understanding. For instance, the absolutes difference would help estimate and interpret the magnitude of disparity between the highest and the lowest sub-groups; however, it would not consider the indicator's distribution across all other sub-groups. The latter was covered by the concentration index, which assessed relative inequalities and also allowed for the statistical comparison of inequalities between different conflict intensities. Similarly, regression coefficients carried the additional advantage of comparing groups while adjusting for other sociodemographic variables.
In this study, the absolute difference was calculated as a difference in the frequency between the highest and the lowest sub-categories of women's educational status (secondary or more education vs. no education) and the wealth index (q5 vs. q1). The latter has also been reported as part of the PHCPI recommended indicators for measuring economic disparities in financial barriers to PHC services [10].
In contrast to the absolute difference, the concentration index with erreygers correction considered all sub-groups to analyse disparities [34]. Concentration index values range from -1 to +1, with positive values indicating the health metric's concentration among the advantaged groups. Negative values indicate the concentration of the health metric among the disadvantaged groups. The closer the value is to zero, the more likely the indicator is equally distributed across all sub-groups. Previously, concentration index was mainly used to assess economic disparities as an extension of the Lorenz curve and Gini coefficient. However, an adapted version of the concentration curve and index allowed the assessment of inequality in binomial health outcomes over the distribution of other ordered categorical variables as educational groups or wealth quintiles [34]. Z statistics were used to test the difference in concentration index estimates between the two categories of conflict intensity.
Given the hierarchical nature of DHS data, which violates the assumption of observations' independence, we performed a multilevel modeling analysis of disparities to appropriately account for additional sociodemographic variables that may affect the overall wealth or educational disparity levels. A two-level mixed-effects logit (random intercept) regression model was fitted for each studied indicator using STATA 16 "svy: melogit command". The fitted model included the following variables: age of women in years, disparity variable (income quintile of the household or educational status of women), employment status of women (categorical), the number of children per woman (continuous), urban/rural status of the household(categorical). The model was fitted separately for the two categories of conflict intensity; then a combined model was used to test for the interaction between conflict intensity and disparity variable. Table S.2 providers more details on model equations [see additional file 1].