Impact of the free healthcare policies in Burkina Faso: underscoring important nuances Open Access

Abstract

The article recently published by Aye et al. (2024) in Health Policy and Planning is a major contribution to understanding the medium-term (5 years) effects of the free healthcare policy introduced in 2016 in Burkina Faso. The study makes rigorous use of interrupted time series with a nonequivalent control group and presents a wealth of information on the methodology used. Remarkably, numerous sensitivity analyses were conducted to strengthen the credibility of the results and limit the risk of bias.

Three salient conclusions are presented: (I) free healthcare had no effect on the proportion of pregnant women who gave birth in a health center, either immediately or after 5 years; (II) free healthcare led to an immediate and significant increase in the rate of consultations for children <5 years of age; and (III) after this immediate increase, free healthcare led to a gradual decrease in the rate of consultations for children <5 years in the medium term. We believe it is essential to highlight some important nuances regarding these conclusions and highlight some methodological issues.

First, while the intervention under study is uniformly referred to as “Gratuité,” this policy comprised two different modalities (Gouvernement du Burkina Faso 2016). On the one hand, this policy made healthcare services free of charge for children under 5 years old, for whom direct payment was reduced from 100% to 0%. The cost of the consultation was not subsidized at all before the introduction of the policy, and it was totally subsidized after its introduction. The situation is different for maternal care and the other outcome under study, i.e. deliveries. Indeed, deliveries had already been 80% subsidized since 2007, well before the introduction of the Gratuité (Ministère de la Santé 2006). Therefore, in 2016, the new policy only covered the residual amount, i.e. 20% of the total cost of deliveries that still had to be paid by patients. The term Gratuité in the text therefore covers two completely distinct changes, and it seems misleading to combine both interventions.

This confusion is reflected in the conclusion that “no significant effects on the use of maternal care services [are] attributable to the gratuité” (abstract). If the article claims to measure the impact of free healthcare policies (title), the analysis should not focus only on the stage that covered the residual 20% of the cost but must also include the prior stage that covered 80% of the cost. Rather, it would have been necessary to consider the two stages during which maternal care gradually became free. Admittedly, the authors mention at the end of the article (p. 897) that a prior intervention existed since 2007, and they hypothesize that this could explain the absence of statistically significant effects observed in 2016. However, it is misleading not to mention from the start, including in the abstract, that the analysis excludes the prior intervention, for which numerous studies have demonstrated the immediate and medium-term benefits on the utilization of maternal care services (Johri et al. 2014, Ganaba et al. 2016, Langlois et al. 2016).

Moreover, it is essential to consider the heterogeneity of the context in impact evaluation, especially when using quasi-experimental evaluation designs or natural experiments (Shadish et al. 2001). In Burkina Faso, healthcare did not become free at the same time in all of the 71 health districts, for both maternal and child health services. This heterogeneous context of implementation is mentioned by the authors (pp. 892–893), who cite their own work examining free healthcare pilot projects introduced before 2016. However, the modeling approach adopted in the Aye et al. study does not allow to take this heterogeneity into account, since the data are analyzed in a single time series for the entire country. Therefore, a unique precise cut-off point was used: “we set June 2016 as the interruption point and divided the overall observation period into two segments: a pre-policy period until May 2016 (41 months) and a post-policy period from June 2016 onwards (67 months)” (p. 893).

This discrepancy between modeling and reality is not insignificant: it concerns at least 5 health districts (out of 71) in addition to 3 regions (out of 13 in the country) where healthcare was already free before June 2016. This misclassification of exposure leads to a dilution of impact estimators (bias toward a null effect) (Szklo and Nieto 2007). This bias could be even more pronounced due to the likely presence of another type of misclassification, i.e. the contamination between the target group (children <5 years) and the control group chosen (children ≥5 years). Indeed, previous studies have shown spillover effects of the free healthcare policy in Burkina Faso, enabling ineligible children ≥5 years to benefit from free care (Druetz et al. 2021). This situation is likely to have created an increase in consultation rates in the control group, thus reducing the effect attributed to the Gratuité by the difference-in-differences estimate (or relative risk ratio).

The threat posed by such biases appears all the more serious, given that some key results show significant instability in sensitivity analyses. One of the main findings is that the immediate gains attributable to free healthcare for children <5 years wane over time. The analyses reveal that the trends in consultation rates are negatively affected after June 2016 in both groups, but significantly more in the exposed group of children (<5 years) than in the control group (≥5 years) (coefficient −0.93, P < .005). However, this coefficient is no longer statistically significant in the sensitivity analyses when (I) a quadratic term is introduced for the trend in the post-policy period or (II) when the cut-off point for the intervention period is advanced by 2 months (see Appendices C2 and F). Sensitivity analyses suggest a lack of robustness for this result.

Beyond the problem of the internal validity of this waning effect, its interpretation leaves us skeptical. The main hypothesis lies in the increasing difficulties that the policy may have encountered: “gratuité implementation may have faced more difficulties due to increasing instability caused by conflicts and COVID-19 in recent years” (p. 897). We share these concerns to a certain extent: COVID-19 does not appear to have caused the healthcare disruptions that many expected (Druetz et al. 2020, 2022, Cooper et al. 2023). But more importantly, many other health interventions specifically targeting children <5 years have been progressively introduced or scaled up in the last few years. This could explain why consultation rates for children <5 years have fallen more rapidly than for children ≥5 years.

In particular, national authorities have gradually scaled up two interventions targeting children <5 years since 2015: the Integrated Community Case Management strategy and the seasonal malaria chemoprophylaxis program, whose effects on malaria-attributable morbidity have been demonstrated (Druetz et al. 2015a, 2015b, 2018). In addition, nongovernmental organizations and international institutions have implemented numerous interventions or pilot projects to reduce morbidity and mortality of children <5 years, such as management of severe acute malnutrition, deworming, and vaccination campaigns (Vanderkooy et al. 2019, Ouédraogo et al. 2022, UNICEF 2024). The proliferation of health interventions targeting children <5 years of age is a well-documented phenomenon in the wake of the Sustainable Development Goals (Bhutta et al. 2019). Arguably, this context cannot be overlooked to understand the recent gradual reduction in consultations among children <5 years in Burkina Faso.

Challenges in maintaining the implementation of free healthcare may have adversely affected service provision, while the success of other interventions concomitantly reduces the demand for services. These are just two among many hypotheses that remain to be explored to better understand the recent reduction in consultation among children <5 years. To this end, it would have been constructive to give access to the codes used in the present study in a supplementary file, or at least to make them available upon reasonable request. Not only would this have allowed the analyses to be replicated, but it would also have helped to overcome methodological challenges and test new hypotheses. Much remains to be done to promote open science, a priority highlighted in the Transparency and Openness Promotion Guidelines to which Health Policy & Planning, like thousands of other scientific journals, is a signatory (Belliard et al. 2023, Kabanda et al. 2023).

Acknowledgements

The authors used DeepL.com and the Translate tool in Microsoft Word to translate this article, originally written in French. The text was reviewed and edited after translation.

Conflict of interest

None declared.

Funding

None declared.

References