Implementing performance-based financing in peripheral health centres in Mali: What can we learn from it?

Introduction: Numerous sub-Saharan African countries have experimented with performance-based financing (PBF) with the goal of improving health system performance. To date, few articles have examined the implementation of this type of complex intervention in Francophone West Africa. This qualitative research aims to understand the process of implementing a PBF pilot project in Mali’s Koulikoro region. Method: We conducted a contrasted multiple case study of performance in 12 community health centres in three districts. We collected 161 semi-structured interviews, 69 informal interviews and 96 non-participant observation sessions. Data collection and analysis were guided by the Consolidated Framework for Implementation Research adapted to the research topic and local context. Results: Our analysis revealed that the internal context of the PBF implementation played a key role in the process. High-performing centres exercised leadership and commitment more strongly than low-performing ones. These two characteristics were associated with taking initiatives to promote PBF implementation and strengthening team spirit. Information regarding the intervention was best appropriated by qualified health professionals. However, the limited duration of the implementation did not allow for the emergence of networks or champions. The enthusiasm initially generated by PBF quickly dissipated, mainly due to delays in the implementation schedule and the payment modalities. Conclusion: PBF is a complex intervention in which many actors intervene in diverse contexts. The initial level of performance and the internal and external contexts of primary healthcare facilities influence the implementation of PBF. Future work in this area would benefit from an interdisciplinary approach combining public health and anthropology to better understand such an intervention. The deductive-inductive approach must be the stepping-stone of such a methodological approach. The term ‘implementation’ refers to one or more processes organised in a particular context to help achieve the changes intended by an intervention through the means being deployed [31]. Many theories and conceptual frameworks exist to understand the implementation of interventions. For this study, we chose to use the CFIR, which consolidates key constructs of implementation theories. It was proposed by Damschroder et al. [32] to help assess how effective the implementation of an intervention is in a specific context. The CFIR has proven useful in a wide range of scenarios, including low-income contexts [33]. We chose the CFIR for two reasons. First, it is easy to apply because of its adaptability to the context and research question. Second, it is one of the few tools that can provide a comprehensive view of the intervention within a logically coherent framework. According to the CFIR, to understand an intervention’s implementation, five ‘domains’ must be studied, namely (1) the characteristics of the intervention; (2) the external context of health facilities; (3) the internal context of health facilities; (4) the characteristics of individuals; and (5) the implementation process. The CFIR consists of 39 constructs and sub-constructs divided among these five domains. The research design we adopted was that of a contrasted multiple case study with several embedded levels of analysis [34]. The cases were community health centres (Centre de santé communautaire; CSCOMs), i.e. primary care centres. Data were collected between December 2016 and January 2017 in three of the 10 HDs in the Koulikoro region – HD1, HD2 and HD3. The three HDs were selected on the basis of specific criteria, namely an agricultural site that had experienced being involved in a cash transfer programme for the poorest; a site where it was planned to test a communal mutual insurance program; and a site with an urban character. Of the three HDs selected, only one (HD1) had taken part in the first PBF pilot project in 2012–2013. Since we were not able to conduct our study in all CSCOMs due to budget and time constraints, we selected four CSCOMs per HD, two from among the highest-performing (CSCOMs ’++’ and CSCOMs ‘+’ according to performance level) and two from the lowest-performing (CSCOMs ‘- -‘ and CSCOMs ‘-‘ according to performance level), for a total of 12 CSCOMs within the three HDs [35] (Table ​(Table1).1). The CSCOMs’ performance level was defined on the basis of qualitative and quantitative criteria that emerged from a participatory and consensual process involving the reference health centre (CSREF) teams and a research team composed of the principal investigator (AC) and a doctoral student to support study preparation and to participate in the selection of study sites in view of collecting data for her thesis [35]. For researchers, this offered a timely opportunity to test a model of participatory case selection [36]. This is an innovative approach that makes it possible to legitimise the criteria for site selection beforehand, in particular, the notion of performance by taking into account the health workers’ perspective. The highest-performing CSCOMs were often associated with better involvement of community leaders in activities, greater community mobilisation, greater demographic density, better involvement of the Community Health Association (Association de santé Communautaire; ASACO) in activities, dynamic health care personnel, etc. Compared to other CSCOMs, the HD1 CSCOMs had the advantage of benefiting from some of the infrastructure put in place during the first PBF project. Number and level of performance of Centres de santé communautaire (community health centres; CSCOMs) by health district (HD) The four CSCOMs selected in each health district were composed of one urban CSCOM (CSCOM of the district capital) and three rural CSCOMs, with the exception of the Koulikoro HD where there are two urban CSCOMs. Some common characteristics were noted. They relate to the type of the infrastructure (generally including a consultation room, a delivery room, a nursing and a hospitalisation room), the profile of the personnel (most often composed of the technical director of the centre (TDC), nurses, midwives, birth attendants, nurses’ aides, vaccinators, drug depot manager and a hygienist). The criteria for the inclusion of CSCOMs in the study were defined as follows: have a community health centre status, located in one of the three HDs selected for this study, and be among the CSCOMs classified either as most efficient or less efficient by the end of the selection process. The exclusion criteria were as follows: any CSCOM not located in one of the three HDs selected for this study and any CSCOM not selected by the end of the selection process. The second PBF pilot project involved a certain number of institutional actors and functions (Table ​(Table22). Functions and tasks of institutional actors involved in performance-based financing implementation ASACO Association de santé communautaire (community health association), CSCOM Community health centre, CSREF reference health centre, HDMT Health district management team, RHD Regional Health Department At the local level, PBF was implemented in CSCOMs and CSREFs. The quantitative and qualitative results of these providers are evaluated by the HD management team for the CSCOM level and the Regional Health Department for the CSREF level. Once evaluated, results are purchased by the local authorities, which are involved in signing performance contracts with the providers (town hall for the CSCOM and circle council for CSREF). The regulatory function (i.e. checking whether norms and standards are respected) is carried out by the Regional Health Department for the CSREFs and the HD management team for the CSCOM. The funds used to purchase the results are mobilised by the Project Coordination Unit/Strengthening Reproductive Health Project (SRHP). A counter-check is carried out by an independent agency to find out whether patients actually received the health services and their level of satisfaction. As in any PBF intervention, health facilities are funded based on the purchase of quantity indicators (Table ​(Table3)3) and quality indicators (Table ​(Table4).4). Ten quantitative indicators reflecting major maternal and child health issues were selected for the pilot project in accordance with the priority topics of the World Bank’s SRHP. In addition, three quality domains corresponding to specific scores were covered, namely resources and processes (30%), clinical quality (50%) and user satisfaction (20%). Quantity indicators selected for the performance-based financing pilot scheme in Koulikoro Quality indicators by category selected for the PBF pilot scheme in Koulikoro ASACO Association de santé communautaire (community health association) Typically, quantity indicators are purchased at a fixed price, whereas the payment for quality indicators depends on achieving a minimum target. In the case at hand, after an audit identified any discrepancy between the figures reported by the CSCOMs and the actual provision of services as well as by assessing users’ satisfaction level and verifying the health workers’ quantitative results, an invoice was drawn up and sent via a web portal to a payment agency, which then made the transfer to each CSCOM’s account. Semi-structured interview guides were developed for each category of individual actors interviewed. These various actors included CSCOM personnel, ASACO members (i.e. members of community health associations responsible for managing the CSCOM on behalf of the community), community health workers (i.e. community members who assist qualified health workers mainly through disease prevention and treatment activities within the community), and community leaders who have the power to influence public opinion and members from the commune (local elected officials). The guides were translated from French into the Bambara language and were then pre-tested. An observation grid was also used. These various guides were developed considering the five domains of the CFIR. Discussions were held beforehand among the co-authors of the article (VR, LG, TZ and VR) to reach a common understanding of the different constructs and sub-constructs. Applying a purposive selection sampling strategy, data were collected from different individual actors involved in PBF implementation in the CSCOMs. We hired research assistants (n = 3) to collect the data. The first author (AC) trained them to use the interview guides. At the start of data collection, AC also provided ‘formative supervision’ by monitoring some interviews conducted by the assistants or by conducting some interviews in their presence. In total, we conducted 161 formal interviews (Table ​(Table5)5) and 69 informal interviews (Table ​(Table6)6) in the three HDs based on the respondent profiles. This informal approach was adopted so the interviewers could collect “confidences and gossip” that would have been difficult to access otherwise. Distribution of respondents by category and health district, semi-structured interviews ASACO Association de santé communautaire (community health association), CSCOM Community health centre, HD health district Distribution of respondents by category and health district, informal interviews ASACO Association de santé communautaire (community health association), CSCOM Community health centre, HD health district Data was collected from these respondents based on their availability [37] and presumed ability to shed light on the situation under study. CSCOM personnel are often limited in number (five to six staff members). In most cases (10/12), all CSCOM personnel were interviewed. In large CSCOMs and other organisations involved in PBF implementation, we chose to interview the people in charge of or overseeing health activities. By including all these actors, we made sure to consider the diversity of the different actors involved in the implementation of PBF. As the study population was approximately the same size in all districts, we chose a similar sample in each of the three HDs included in the study (69 respondents for the three HDs). We reached saturation with the total number of interviews conducted in the three HDs. Our deductive approach is based on the application of predetermined codes to the data and, in such an approach, saturation refers to the extent to which predetermined codes or themes are adequately represented in the data [38]. Among CSCOM personnel, the illiterate or least educated are often healthcare assistants or hygienists. We have taken into account the influence of this reality on certain facts, particularly the level of information on PBF, i.e. by comparing this category of staff’s perceptions to those of respondents with higher levels of education, such as nurses, TDCs, depot managers. At the level of each CSCOM, the first interview was typically conducted with the TDC, who then introduced us to the other actors (health workers, ASACO members, etc.). Research assistants used a recorder to record the interviews as well as personal notebooks to record their own reflections along with data collection. Informal interviews were most often conducted outside the CSCOM, either at the respondent’s house or in another place chosen by the respondent. Non-participant observations were conducted in the CSCOMs to study interactions between service providers and patients during medical consultations as well as changes made at the managerial level since the launch of the PBF activities (hygiene of premises, data recording procedures). The observation results were recorded in a logbook. The investigators conducted 96 observation sessions for the 12 CSCOMs (HD1 = 32, HD2 = 32, HD 3 = 32), i.e. an average of 8 observations per CSCOM. We considered that this amount of observation sessions were sufficient to reach saturation and address key topics of interest, namely hygiene, the way in which the tools used by the personnel were filled, etc. The researchers classified all interviews conducted and observation notes according to HDs and CSCOMs. All formal interviews, conducted in Bambara, were audio-recorded and then fully transcribed directly into French. Translations from Bambara to French and from French to English were performed by the first and second authors and were further polished by professional translators. The entire dataset (transcripts and notes) was subsequently coded using QDA Miner Lite software. The researchers familiarised themselves with this software beforehand in a 3-day training workshop led by LG. For coding and data analysis, we adopted a deductive–inductive thematic analysis using the CFIR domains, constructs and sub-constructs. These CFIR dimensions guided the definition of the initial codes, subcategories and themes. Coding was performed by two research assistants trained and guided by the first author (AC). Of these two coders, only one took part in the data collection. A coding tree was developed by AC and validated by all the co-authors of the article. The code tree presented the five CFIR domains as general themes while allowing for the inclusion of other themes suggested by the data. An additional file shows this in more detail (see Additional file 1). An interpretive thematic analysis was then applied. Reporting of the findings was entirely guided by the CFIR five domains. We also provided a summary of key findings highlighting critical inter-case comparisons. At the time of data collection, most of the CSCOMs had started the PBF implementation just 3 or 4 months earlier. As such, it was difficult for the workers to perform certain activities that required more time to become apparent. The following constructs and sub-constructs were not informed by our data: ‘reflection and evaluation’, ‘opinion leaders’, ‘internal leaders formally appointed for implementation’, ‘champions’, ‘evolution’. All these constructs and sub-constructs belong to domain 5 (‘process’) of the CFIR. Although these excluded constructs and sub-constructs are certainly important to describe the implementation process, we were unable to use them due to insufficient empirical data. These sub-constructs were especially related to the theme of stakeholders’ involvement. In making sure that we would still report findings related to this theme, we focused our attention on other forms of actors’ involvement that could be informed by the data, in particular the strategies used to encourage the involvement of various actors or to promote their commitment. Further rationale for excluding these constructs and subconstructs is provided in Appendix in Table 9. Measures were taken to protect respondents from the potential risks associated with their participation in the study. Every precaution shall be taken to ensure that information concerning them is not divulged, including anonymising of audio recordings and transcripts. Respondents were informed that they could decline to answer certain questions and that they had the right to withdraw from the study when they wished.