How equitable is health spending on curative services and institutional delivery in Malawi? Evidence from a quasi-longitudinal benefit incidence analysis

Background: Malawi is one of a handful of countries that had resisted the implementation of user fees, showing a commitment to providing free healthcare to its population even before the concept of Universal Health Coverage (UHC) acquired global popularity. Several evaluations have investigated the effects of key policies, such as the essential health package or performance-based financing, in sustaining and expanding access to quality health services in the country. Understanding the distributional impact of health spending over time due to these policies has received limited attention. Our study fills this knowledge gap by assessing the distributional incidence of public and overall health spending between 2004 and 2016. Methods: We relied on a Benefit Incidence Analysis (BIA) to measure the socioeconomic inequality of public and overall health spending on curative services and institutional delivery across different health facility typologies. We used data from household surveys and National Health Accounts. We used a concentration index (CI) to determine the health benefits accrued by each socioeconomic group. Results: Socioeconomic inequality in both public and overall health spending substantially decreased over time, with higher inequality observed in overall spending, non-public health facilities, curative health services, and at higher levels of care. Between 2004 and 2016, the inequality in public spending on curative services decreased from a CI of 0.037 (SE 0.013) to a CI of 0.004 (SE 0.011). Whiles, it decreased from a CI of 0.084 (SE 0.014) to a CI of 0.068 (SE 0.015) for overall spending in the same period. For institutional delivery, inequality in public and overall spending decreased between 2004 and 2016 from a CI of 0.032 (SE 0.028) to a CI of -0.057 (SE 0.014) and from a CI of 0.036 (SE 0.022) to a CI of 0.028 (SE 0.018), respectively. Conclusions: Through its free healthcare policy, Malawi has reduced socioeconomic inequality in health spending over time, but some challenges still need to be addressed to achieve a truly egalitarian health system. Our findings indicate a need to increase public funding for the health sector to ensure access to care and financial protection. We applied a Benefit Incidence Analysis (BIA) to assess the distributional incidence of both public and overall health spending on curative services and institutional delivery at three time points. Public health spending refers to public subsidies allocated to health facilities for the provisions of care. Overall spending refers to all sources of health spending allocated to health facilities in terms of public spending, external support and out-of-pocket expenditure (OOPE). BIA measures whether the financial benefits of health services reach individuals across socioeconomic groups equally at a specific point in time [34, 35]. BIA has traditionally been employed to assess the distributional incidence of public spending. Given the growing global emphasis on fostering UHC by combining multiple financing mechanisms, McIntyre and Ataguba recently argues in favor of expanding the scope of the BIA methodological approach to assess the distribution incidence of overall health system spending. Our work builds on their theoretical postulations and their specific methodological guidance [34]. Due to the nature of BIA methodology and the available data, it was impossible to conduct strictly-speaking longitudinal analysis. Therefore, we describe our study as quasi-longitudinal. The computation of BIA relies on two datasets: data on health service utilization stratified by socio-economic status and data on the unit costs of different types of health services. In other words, BIA expresses in monetary terms the distribution of health benefits. As such, BIA aims to capture the extent to which investments in the health sector reach equally all strata of the population. To perform this analysis, we used data from available nationally representative repeated cross-sectional household surveys and National Health Accounts (NHA) for health service utilization and health spending, respectively. Before deciding on the time points of our analysis, we depicted and attempted to match, to the extent possible, for three time points: (1) available health policies and interventions (Fig. ​(Fig.1)1) that were implemented in Malawi to foster progress towards universal coverage of curative and maternal services; (2) the household survey data on utilization of curative services and institutional delivery; and (3) available data on health spending on curative services and institutional delivery available. We repeated the BIA at the selected three time points to explore changes in the distributional incidence of health spending in relation to the different UHC reforms implemented in the country. Table S1 in the Additional file 1 illustrates which variables were used from each household survey, for each year, and briefly describes the sampling strategy of each survey. The repositories of the used datasets are provided under availability of data and materials. We derived our data on health care utilization from the Integrated Household Living Condition surveys (IHLCS) for curative services and the Demographic and Health Surveys (DHS) for institutional delivery. These nationally representative household surveys, normally conducted every five years, contain data on the utilization of curative services and institutional deliveries differentiated by provider typology and a measure of socioeconomic status (SES), allowing us to categorize individuals by weighted SES quintiles. Table ​Table11 indicates the health variables we extracted from each household survey. Variables and data sources 2004 2010 2015 2004 2010 2015 As a ranking variable to build socioeconomic strata, we used per capita consumption expenditure based on the total household food and non-food expenditure for IHLCS data sets and the household-wealth-index factor scores generated through the principal components analysis based on household material asset ownership for DHS data sets. We estimated the annual visits to curative services and institutional deliveries in the study year for individuals across different socioeconomic groups. We used a binary variable for curative services indicating whether the respondent had used curative services in the previous 14 days and a binary variable for institutional delivery indicating whether a woman had delivered in a facility in the prior twelve months. Counts of curatives services were annualized to obtain yearly counts by multiplying the visits recorded for the 14-day recall period by 26. We categorized curative services and institutional delivery by different health facilities types depending on data availability in each survey and NHA. Due to seasonal variation of disease incidences and use of health services, the literature indicates that the annualized utilization of health services may be underestimated or overestimated based on the period of data collection [36]. To account for these seasonal variations, we conducted a sensitivity analysis by adjusting the utilization of curative services and institutional delivery from the household surveys by the monthly seasonal variations in the use of these services. We built a monthly seasonality index using data from the 2014-2018 Health Management Information System (HMIS). We estimated averages of monthly health care utilization reported in the HMIS between 2014 and 2018 and used these averages to calculate the monthly seasonality indices. We then accounted for seasonal variations in the utilization of curative services and institutional delivery using corresponding seasonality indices depending on the months for which health service utilization was reported in the household surveys. We derived data on health spending from the National Health Accounts (see Fig. ​Fig.11 and table ​table1).1). We estimated the unit cost using recurrent public spending, donor spending and household OOPE from the National Health Accounts. We applied the constant unit subsidy assumption for the public and donor spending to estimate the unit subsidy at different types of health facilities. We determined the unity subsidy of each type of health service at each type of health facility by dividing the total health spending for one type of service by the total utilization of that service at this health facility. For OOPE, we relied on the constant unit cost assumption for each quintile based on the percentage of OOPE incurred by each quintile at different types of health facilities. We adjusted the OOPE across quintiles based on the works by Nakovics et al. [27] for curative services and by Chinkhumba et al. [37] for institutional delivery. The OOPE adjustment was based on the fact that the individuals belonging to different SES quintiles generally display different OOPE at different types of facilities. Hence, using a constant unit OOPE at each type of facility would overestimate the OOPE incurred by the bottom SES quintiles. The studies by Nakovics et al. [27] and Chinkhumba et al. [37] indicated that the least poor incurred approximately twice as much OOPE for curative services and one third more OOPE for institutional delivery than the poorest segment of the population. We estimated the unit cost for each quintile and each type of health service at each health facility by dividing the total OOPE incurred by that quintile for that service at the same health facility by the total utilization accrued to that quintile for that service at that health facility. We computed traditional BIA by measuring only the distributional incidence of public spending and comprehensive BIA by looking at the distributional incidence of overall health spending, including public and donor subsidy allocation to facilities as well as OOPE incurred by individuals. Our choice was motivated by a wish to provide a comprehensive health system assessment, in line with the policy intention of the reforms implemented to promote and sustain increased service coverage in the country. Based on the data availability (table ​(table1),1), we decomposed our analysis by different health facility typologies for both curative services (public facilities vs faith-based facilities vs private facilities) and institutional delivery (public health centers vs public hospitals vs faith-based health centers vs faith-based hospitals vs private facilities) and each year. To determine the total financial health benefits at each type of health facility, we multiplied the unit subsidy or unit cost by the total utilization of health services at each type of health facility. We used concentration indices to measure the degree of inequality in the distribution of public and overall health spending on curative services and institutional delivery across socioeconomic groups. The concentration index (CI) quantifies the degree of wealth-related inequality and ranges from -1.0 to +1.0. The CI takes a negative (positive) value when the financial health benefit is concentrated disproportionately among the poor (least-poor). If the CI is close to zero, a lower degree of inequality is present, and if it is zero, there is an absence of wealth-related inequality [35]. We adjusted the concentration indices by the sampling weights of the IHLCS and DHS household surveys to scale up sample-specific estimates to reflect the national population. The standardized concentration index (Ch) is estimated as follows [35]: Where hi is the health variable (e.g. health care utilization) for individual ί, μ is the mean of health variable, Ri is individual i’s fraction socioeconomic rank, and Cov (hi, Ri) is the covariance. We used convenient regression [38] to calculate the standard errors of the concentration index. The formula is: Where 2σ2R is the variance of the fractional rank variable, β is the estimator of the concentration index. We performed a dominance test at a significance level of 5%. The dominance test is used to statistically verify if a determined pro-poor or pro-least-poor distribution holds across the entire distribution of the socioeconomic variables [34, 35], especially when it is not clear if a distribution is pro-poor or pro-least-poor [39].