Street-level workers’ inadequate knowledge and application of exemption policies in Burkina Faso jeopardize the achievement of universal health coverage: evidence from a cross-sectional survey

Given the very low use of health services by indigents in Burkina Faso [24], we hypothesized that street-level workers’ knowledge and application of exemption measures were also very low. In addition, given the pyramidal structure of the health system, we hypothesized that street-level workers’ knowledge and application of these measures would increase in proportion to their levels of education and job responsibility. However, we also expected that personnel working in obstetric services would be more knowledgeable about the user fee exemption for indigent women’s deliveries and would apply those measures more often than would others. Given the conflict of interest that arises because health workers receive, from the facility management, part of the fees paid locally by users for curative consultations [25], we hypothesized that the application of the general exemption would be lower than that of the delivery exemption, which is reimbursed by the central level.

Burkina Faso is a French-speaking landlocked country in West Africa. It has a population of approximately 18 million people, mostly engaged in the informal sector, primarily in subsistence agriculture [26]. With an annual per capita Gross Domestic Product (GDP) of USD 650 (2016), Burkina Faso is one of the poorest countries in the world, where nearly 44% (2014) of the population live below the national poverty line of USD 1.90 a day [26].

In 2014, total health expenditure in Burkina Faso amounted to nearly 5% of GDP, with private out-of-pocket expenditure accounting for nearly half (48%) of this amount. Health service provision is organized in a three-tier system, with first-line facilities (Centre de Santé et Promotion Sociale – CSPS) located in the rural areas, second-line facilities in the district capitals, and tertiary referral facilities located in the regional capitals and in Ouagadougou. Health service provision continues to be characterized by large inequalities between rural and urban areas, mostly due to problems linked to inadequately equipped and staffed health facilities [27].

In the early 1990s, the Bamako Initiative, combined with the introduction of structural adjustment policies, resulted in user fees being introduced at all levels of the health sector [28]. Since then, similarly to what has been observed elsewhere, user fees in Burkina Faso have been proven to impede access to care and to expose households to catastrophic health expenditure [29].

In an attempt to enhance equity in access and foster financial protection in a system largely based on user fees, the government of Burkina Faso has enacted two measures in the years since the 1993 Bamako Initiative. First, in January 2007, when passing legislation to reduce user fees for obstetric services, the government decided to exempt the poorest quintile of women from paying any fee, including the reduced co-payment for delivery services foreseen by the new policy. This 2007 measure was funded by the State’s national budget from 2007 to 2015 [30]. Second, starting in 2009, the government prescribed that the worst-off (les indigents) should be fully exempted from paying user fees for all preventive and curative services provided by public facilities. Local health management committees were mandated to fund these exemptions by redistributing resources acquired via cost-recovery schemes. Throughout their implementation, neither measure has systematically included any specific effort to identify potential beneficiaries, leaving it to individual providers to implement the exemption.

These two exemption measures are the focus of our analysis.

To assess knowledge and application of the two exemption measures described above, we used cross-sectional data from the baseline round (carried out between October 2013 and March 2014) of the health worker survey conducted as part of the impact evaluation of a performance-based financing (PBF) intervention [31].

In line with the World Bank toolkit for impact evaluation [32], the structured survey collected information on individual health workers, specifically: age, sex, and years on duty; income and its sources; training and clinical knowledge; working conditions; and motivation and satisfaction with multiple aspects of daily work. Given our specific research interest in exemption policies, we added to the standard survey tool a section specifically exploring their knowledge and implementation of existing policies and their attitudes towards forthcoming ones.

The survey targeted 1521 health workers in all 498 first-line (CSPS), 23 second- and third-line (district and regional hospitals) facilities, and 11 private or other facilities included in the sample and distributed across 24 districts. The sample included a census of all public and private facilities in 12 intervention districts and a random sample of one-third of all public and private facilities in control districts. Research assistants were instructed to interview as many health workers as possible during their one-day stay at each facility. In practice, this resulted in approximately three health workers surveyed per facility, which is equivalent to 80% of all health workers present on facility premises on the day of the survey (the remaining 20% not being interviewed due to unavailability or lack of time, not due systematic screening by the interviewers or to an explicit refusal). Our sample included an ample variety of health workers: doctors, registered nurses, assistant nurses, midwives and assistant midwives, as well as mobile health workers (agents itinérants de santé, AIS). We purposely administered the survey to this wide range of health workers and kept them all in our analysis because they all interacted with patients to provide care and, as such, all made decisions on the application of exemption policies [27].

We did not restrict our analysis to a specific cadre, as in other analyses based on the same survey [33]. However, due to missing data, the effective sample varied across outcome variables. Specifically, we could obtain complete information for: a. between 1345 and 1314 health workers on the primary outcomes (for indigent exemption and delivery exemption respectively); b. between 1503 and 1472 health workers for the secondary outcome. Descriptive statistics on 1311 health workers, due to missing values in the explanatory variables.

Table 1 provides an overview of all outcome and explanatory variables included in the analysis and their measurement. Specifically, we defined four primary outcome variables: two to assess knowledge of the two above-mentioned exemption measures and two to assess the application of the two measures. All outcome variables were measured using a dichotomous scale.

Knowledge and application for exemption measures for preventive and curative services were initially assessed with the questions outlined below, whereby we later categorized yes as 1 (one) and no as 0 (zero):

Knowledge and application for exemption measures for for delivery services were initially assessed with the questions outlined below, whereby we later categorized yes as 1 (one) and no as 0 (zero):

We purposely assessed knowledge and application at the level of the individual health worker and not at the facility level because, as mentioned, the government left the responsibility for identifying potentially eligible individuals up to individual health workers, with no specific role in this process assigned to the facility administrative manager. Thus, based on our experience in the country, we expected to observe variation in both knowledge and application, even within individual facilities.

Most explanatory variables are self-evident and reflect the health workers’ sociodemographic profile. Education is listed as an explanatory variable because, as mentioned, our sample included lower-level workers, such as the AISs. Thus, we expected variation in education levels (Additional file 1). To counteract problems due to small samples within single occupational types, we grouped health workers into four major categories: medical doctor (very few observations) or nurse; midwife; assistant midwife; and AIS [27]. The variable “dual practice” assessed whether a health worker held an additional permanent post. The variable “performance evaluation in the last year” measured whether the health worker had undergone an evaluation within the previous year. Six dichotomous variables assessed whether the health worker had received training in any of five specific fields over the past 12 months (emergency obstetric care and neonatal care; pro-poor targeting; user fee exemption; health insurance; facility financial management). Two variables measured health workers’ perceptions regarding their capacity to influence decision-making processes and control work at their facility. Both variables elicited answers on a Likert scale with values ranging from 0 to 10.

Five variables measured facility-level characteristics, specifically: rural vs. urban location; facility type (1 = primary facility; 2 = secondary facility – including the few regional hospitals included in the sample; 3 = private); district; number of clinical staff assigned as personnel; and proportion of very poor people residing in a specific district. This last variable was computed using data from the parallel household survey conducted in the catchment areas of all facilities included in our sample. We computed this variable by looking at how many people were listed in the lowest wealth quintile in each of the districts. To generate this variable, we first computed an asset-based wealth index for the overall sample (i.e., pooling data from all 24 districts) and then checked what proportion of very poor resided in each district. We treated this as a continuous district-level variable, with district-specific values ranging from 0 to 1.

In addition to exploring what factors caused variation in knowledge and application of exemption policies, we included a secondary outcome, asking health workers to identify what proportion of the total population should be considered ultra-poor and therefore be exempted from paying user fees. Five categorical options were provided (1%, 5%, 10%, 20%, 25%).

We used univariate descriptive statistics to explore the distribution of the variables of interest in the sample. We ran an ANOVA model with a Sidak multiple-comparison test to identify statistically significant differences across districts for each of the four outcome variables. Given the large number of districts in our sample, we grouped districts that displayed values close to average (i.e., districts whose values on the four outcome variables on average deviated less than ten percentage points in either direction) into one category and compared them against the 10 remaining districts, whose values on average diverged by more than +0.1 or −0.1 from the mean of the four outcome variables.

We ran four random effects models (one for each outcome variable) to identify associations between outcomes of interest and selected sociodemographic and health facility characteristics of health workers. We used random effects models because we wanted to account for potential clustering at the district level, based on the large observed differences in knowledge and application of the exemption mechanisms that emerged from our initial descriptive analysis. All four models accounted for clustering at the district rather than the facility level for two reasons. First, descriptive statistics suggested the presence of district-level effects rather than of facility-level effects. Conceptually, this might be due to the influence of the district health management team as well as to the mobility between facilities within a given district, a phenomenon that is well-known in Burkina Faso [34, 35] and that provides a conceptual grounding for our empirical analytical choice. Second, the number of health workers interviewed in each facility was very small (fewer than three health workers in two-thirds of all facilities), which did not allow for meaningful clustering at the facility level.

Specifically, we ran multilevel models with random coefficients in Stata version 13.1, using the xtlogit command. We expressed our effect estimates as beta coefficients and displayed the standard errors. In the regression equation, Y represented outcome variables, while X represented explanatory variables. In a multilevel model, we assumed that each class (subscript j), here districts, had a different intercept coefficient (β_{0 j}) and different slope coefficients ((β_{1 j}, β_{2 j},…). Subscript i represented the individual units, here health workers; e _ij represented the residual errors. The equation for each of the four multi-level regression models was as follows:

The results are relatively disturbing, as they show that very little has changed since the early 2000s. Health workers have little concern for equity issues in West Africa [36] and may not be sufficiently empowered to promote equity in health. This study confirmed our earlier qualitative and quantitative studies, which showed that these user fee exemption policies were not being sufficiently applied [22, 37]. The State decided on several measures to support the worst-off and did so without external funding, which is remarkable and speaks to the high level of political will for equity, especially in the delivery exemption for the indigents [30]. However, political will cannot stop at declarations and the formulation of public policies; measures must be put in place to support their application. This gap was again noted recently in Burkina Faso in the context of a national program to cover indigents using mutual health insurance programs. The program’s content was formulated and funded, but it was never properly implemented, such that potential indigent beneficiaries were once again left out [38]. Implementation science has a long history, yet the lessons it has offered for decades seem to have not yet come to the attention of decision-makers [8, 39]. Even though a causal link between knowledge and its application is not guaranteed, as long as street-level workers do not know the content of policies, they will certainly not be able to apply them. We have shown in previous work that issues of equity and questions relating to exemption policies are not on the health-related curricula in Burkina Faso [40], and they are no longer a topic of discussion in national policy dialogues.

Our study suggests that only a small proportion of health workers appear to know about issues pertaining to targeting and exemption. That suggestion appears to be confirmed by parallel data indicating that less than 2% of all health workers interviewed reported ever having received explicit training on either exemption policy. As such, there is definitely a pressing need for both basic and continuing education on this issue if any action is to be taken to close the obvious implementation gap. Furthermore, the fact that health workers would like to keep the indigent selection coverage to a low level (Table 2 shows that about 30% of health workers indicated less than 5% as an appropriate targeting goal in a country where 9% of the population lives in situations of extreme poverty) confirms prior evidence [41] and may not be due only to lack of training, but also to a lack of concern for equity, as indicated in earlier research [22, 40]. Current payment systems, however, are also likely to be largely responsible for the low application of the exemption policy, given the net gain that health providers make on consultation fees and drug sales. In addition, health workers may be also sincerely worried about the impact of user fees exemptions on health centres’ financial sustainability in a context of cost-recovery schemes and long reimbursement delays. A comparison of the implementation processes for three indigent targeting models in one district of the country showed, in fact, that health workers had little financial incentive to exempt indigents from paying user fees [42]. A pilot project begun in 2015 to exempt indigents from user fees in 12 districts of the country, in which consultations and health worker bonuses are being paid by the World Bank [31], is expected to produce valuable lessons on the effects of strategic purchasing on equity and on the use of services by the worst-off.

This quantitative study provides new and very useful knowledge for decision-makers, while also opening up new and important research needs. For example, the amount of variance in the outcome variable that is due to districts cannot be explained by our model. Here is where it is essential to study the role of context, but this is still rarely done in this type of interventional research [43]. Even though the concept of context is still at a stage that is not very “mature” [44], it has been used to understand the heterogeneity of impacts of the delivery subsidy policy in a district in the north of the country [12], as well as why some women in a western district continued to give birth at home [45] despite this policy. It is thus important to undertake qualitative research to understand the observed differences in the knowledge and application of these policies between districts.

At the same time, however, this study was able to show the preponderant role of certain contextual elements in explaining the results. For example, the distribution of respondents in our samples suggests that AISs were often present as healthcare providers in the centres at the time of the survey. Our experience and evidence [12, 27, 46] suggests that it is also very likely that this is the case every day, and therefore that AISs are the ones most in contact with populations and health services users. Workers who are more qualified or have more responsibilities are more often absent and so do not attend training sessions or go to meetings far from their health centre. It is therefore logical that AISs are the ones who are best informed and most concerned about this issue, and that they would report applying the measures to a greater extent. It would certainly be interesting to consider relying more heavily on these health workers, who need to be better trained, to ensure more diligent application of the indigent policies.

Likewise, the leadership role of health centre heads or district chief medical officers, well understood in the country [34] was again highlighted by the results of this study. The directive to exempt indigents was kept very confidential; only those in charge received it. They were supposed to distribute the message to the health centres, but this was not really done, as is often the case in West Africa [47]. Top-down transmission of information in a pyramidal structure is often not effective, as we showed recently in the case of PBF in Burkina Faso [42]. It would be interesting to discuss this hypothesis with the district chief medical officers to understand how information about these directives is distributed, knowing there must still be large differences among districts.

The environment in which health workers live and work may also influence their ideas and views regarding poverty and the needs of surrounding populations. This could explain why those living in districts with higher proportions of indigents were more likely to know and apply the exemption measures for deliveries but not for indigent healthcare. This difference might be due to the fact that the national policy on deliveries was more recent and better organized [30] than the indigent care policy, which remained more confidential [48]. Moreover, we saw that those working in districts where efforts to support indigents were actually implemented (e.g. Nouna [49]) knew more about the exemption measures than did others. The role of ideas in public policy implementation has been widely debated in the scientific literature, both with respect to public policies in general [4] and to policies on user fee exemptions [50]. This study could show that, in fact, there is a link.

The strength of this study lies in the large sample size (1345 health workers on the primary outcomes) and geographic spread (24 districts), as well as in the application of multilevel regression modelling techniques to assess associations between street-level workers’ knowledge and application of exemption policy and socio-demographic and health-facility factors. The large sample size contributes to the external validity of the study, providing reassurance that the results represent the variety of experiences across the entire country. The application of multilevel modelling techniques made it possible to produce a reliable analysis to identify relevant associations with the outcomes of interest.

Nevertheless, we cannot exclude the presence of selection bias, since, as explained in the methods section, the combination of the time available for the survey and the caseload at the facility, allowed us to interview only 80% of all health workers present on the day of the interview. We have no means of assessing if the non-interviewed health workers differed significantly from the interviewed ones. In addition, we have to acknowledge possible social desirability reporting bias, especially in relation to questions on the application of the exemption mechanisms, and that actual behaviour in applying the exemptions may differ from what was declared during the survey. In addition, the lack of district-level explanatory variables (such as district leadership and management, district wealth, existing health interventions, etc.) made it impossible for us to expand the model and explain the district-level variance that persisted across all models. Further studies looking into district-level characteristics are urgently needed to explain the observed geographical variance.