Applying the Consolidated Framework for Implementation Research to Optimize Implementation Strategies for the Friendship Bench Psychological Intervention in Zimbabwe

Scaling up an evidence-based mental health program such as the Friendship Bench (FB) Zimbabwe bares many difficulties. Guided by the Consolidated Framework for Implementation Research (CFIR), this study sought to understand the contextual determinants of FB performance and to identify and match novel implementation strategies to optimally overcome barriers. The CFIR was applied to guide twenty-five focus group discussions with n = 152 stakeholders in nine high- and four low-performing clinics across three cities in Zimbabwe to identify the constructs distinguishing clinic performance. Identified implementation barriers were included in the CFIR-Expert Recommendation for Implementation Change (ERIC) matching tool to inform strategies to optimize future FB implementation outcomes. Constructs peer pressure, goals and feedback, learning climate, planning, formally appointed leaders, and external change agents strongly distinguished between high- and low-performing clinics. Barriers to implementation were intervention complexity and organizational incentives and rewards; intervention-related constructs were identified as enablers. Strategies such as identifying and preparing champions, altering incentive structures, developing a formal implementation blueprint, and organizing regular clinician meetings were chosen to address barriers and optimize FB implementation. There was variability in the implementation performance of FB three years post-scale-up. High-performing clinics appear to be influenced positively by peer pressure, goal setting and feedback, better staff communication and integration of the program, overall learning climate, and clinic-level planning. A set of enhanced FB implementation strategies should be applied to improve implementation performance.


3
Few of these proven effective psychological interventions have been scaled up in LMICs (Eaton et al., 2011;Wagenaar et al., 2020) and there is little evidence on how best to deliver these at scale (Patel et al., 2017). There are a paucity of studies across Sub-Saharan Africa examining barriers and facilitators to scaling up task-shared psychological interventions in routine primary care (Awenva et al., 2010). Task shifting involves the redistribution of tasks among health care workers, where appropriate tasks are moved from high-qualified health workers to health workers with less qualifications to ensure efficient use of available human resources in health (World Health Organization, 2008). The FB program is based on a task-sharing approach in which trained and supervised Community Health Workers offer a course of problem-solving therapy to clients on primary health care or at community level, the intervention has been described elsewhere (Chibanda et al., 2016). In the FB program, trained community health workers are supervised by FB staff trained in supervision. Primary health care clinic nurses are first referral points for those clients who need a higher level of care detected by their screening tool score and questionnaire items that indicate the possibility of more severe symptoms, such as suicidality and psychosis. Nurses can refer to psychiatric units.
The question remains on how mental health treatments shown to be effective in RCTs will translate to public health impact in routine, non-research settings (Murry et al., 2014). Poor governance is repeatedly found as a barrier to integration of mental health care into Primary Health Care (PHC) (Thornicraft et al., 2010). Primary health care is a national approach consisting of basic clinics distributed according to population density and staffed by minimum nurse levels. Primary health care was an introduction of health care by the government to disadvantaged areas ensuring active participation of the communities in transforming their health (Ray & Masuka, 2017).
Lack of overall leadership, inadequate financing, workable mechanisms for intersectoral collaboration, and community and service user engagement have an impact on implementation of mental health interventions (Murry et al., 2014;Petersen et al., 2017). Few studies have tested or examined strategies for optimal implementation outcomes and sustainability of psychological interventions during and post-scale-up in LMICs (Wagenaar et al., 2020).
One of these psychological interventions is the Friendship Bench (FB) program, a brief psychological intervention delivered by trained lay health workers through delivering problem-solving therapy on a wooden bench at a primary health care clinic. According to Chibanda and colleagues (2016) Problem-Solving Therapy (PST) has shown to be effective at managing kufungisisa, a local Shona concept of 'thinking too much" (Fernando et al., 2021). Friendship Bench was scaled up in Zimbabwe to 36 primary health care (PHC) clinics in 2016 (Chibanda et al., 2016). To analyze how the evidence-based FB has been performing in a real-world setting, we chose an implementation research approach and applied two widely used frameworks, the Consolidated Framework for Implementation Research (CFIR) (Damschroder & Lowery, 2013) and the Reach, Effectiveness, Adoption, Implementation, and Maintenance, RE-AIM framework (Glasgow et al., 1999).
The RE-AIM framework has been used successfully to evaluate and plan various public health care interventions within various settings (Glasgow et al., 1999). Glasgow and Estabrooks, (2018) describe the 5 RE-AIM domains as follows: reach, the number and proportion of the intended audience that receives the service of the intervention; effectiveness, the degree to which the intervention changes the health outcomes and quality of life of those who receive the service; adoption, the number and proportion of intervention settings that agree to initiate the intervention; implementation, the degree to which the intervention is delivered as intended; and maintenance, the sustained effectiveness, and delivery of the intervention (Glasgow & Estabrooks, 2018). The CFIR framework was created as a comprehensive guide based on defined constructs across 5 domains that potentially influence implementation within different intervention settings (Damschroder & Lowery, 2013).
The CFIR provides a basis to which interventions can identify key barriers to effective implementation and then match key implementation strategies to overcome known implementation barriers. The 5 CFIR Domains include (1) characteristics of the intervention; (2) outer (external) setting; (3) inner (organizational) setting; (4) characteristics of the individuals involved; and (5) the process used to implement the intervention. The CFIR framework is well suited to guide implementation of health care delivery as it identifies factors that may emerge in various contexts of the intervention (Gimbel et al., 2016). The additional tool to support researchers and implementers, the CFIR Expert Recommendations for Implementing Change (ERIC) (Waltz et al., 2019), is a list of implementation strategies that are endorsed by implementation science experts and help to match potential barriers.
The development of the CFIR-ERIC involved a Delphi process with experts to create consensus on the listed 73 strategies and their definitions. The overall objective of this study was to utilize the CFIR framework to identify key determinants of successful FB implementation and develop an optimized package of strategies to overcome key barriers to implementation. The findings from high-and lowperforming clinics, as determined through our classification of performance using the RE-AIM framework, were used to explore these barriers and to improve continued scale-up of the FB program to primary care clinics in Zimbabwe (Verhey et al., 2021). We anticipate this paper to be of interest to researchers, implementers, and policymakers leading scaleup of primary care models for treating common mental disorders in diverse LMICs.
Implementation research aims to provide a systematic way to describe and evaluate how an evidence-based intervention is integrated into a real-world setting. This study analyzes the implementation performance of the Friendship Bench program after having been scaled up to 36 clinics in 2016. This is the first time, to our knowledge, that the Consolidated Framework of Implementation Research (CFIR) is used to analyze a mental health intervention in Zimbabwe with regards to the Friendship Bench program barriers and enablers.

Study Setting
This study focused on the implementation of the FB program in 36 primary health care clinics in Harare, Chitungwiza, and Gweru. Harare is the capital city of Zimbabwe. Gweru is located 300 km southwest of Harare while Chitungwiza is 27 km south of Harare. In all three cities, the health system is run by an independent City Health Department that oversees primary health care, including hiring of clinic staff such as doctors, nurses, midwives, HIV counselors, pharmacy technicians, community health workers, community health worker supervisors, and district health promoting officers. District health promotion officers are assigned to supervise districts which contain various clinics depending on the size of the catchment area. District health promotion officers are responsible for overseeing the community health workers' activities, both at clinic and outreach levels.
The primary delivering agents for the FB program are community health workers who are non-specialized health workers employed by health authorities to carry out health promotion duties at both the clinic and community levels. Community health worker supervisors are community health workers selected as peer supervisors. Besides having received training on health promotion themes, community health worker supervisors have been trained to offer the FB intervention and are supervised by district health promotion officers, who hold managerial positions at district level within the health department. District health promotion officers work closely with other clinical staff at the clinics. Most clinics in the three cities are in areas with highpopulation density (20,000-80,000 people per catchment population). Clinics are classified into three types, large, medium, and small, according to the size and the range of services offered. The size of the clinics is a combination of the number of services offered at the facilities, the largest offering 3 (maternal health services, opportunistic infections services, and family health services) and the size of catchment population served by the clinic.
While at the clinic, the community health workers offer psychoeducation to all visiting patients at waiting areas about mental health and the FB program. Those who are interested in the program or in receiving a service are referred to the FB where they are screened with the Shona Symptom Questionnaire (SSQ-14) (Patel et al., 1997). Community health workers explain the results to the client (psychoeducation) and offer a first (and/or follow-up) counseling session. Clients are also invited to join a voluntary peer-led support group. These groups are an additional support for FB clients focusing on behavioral activation and income generation. More details on the clinical protocol of FB are described elsewhere (Chibanda et al., 2016).
Our overall study methodology utilized a sequential mixed-methods approach with two steps (Verhey et al., 2021). In the first step, FB clinics were classified according to implementation performance utilizing the RE-AIM framework . We focused pragmatically on Reach, Implementation, and Maintenance and designed specific questionnaires to seek the information from the key stakeholders. The 36 clinics were scored across the 3 RE-AIM domains and were ranked from clinic number 1 to number 36.
In the second step, the CFIR framework was utilized to analyze determinants of implementation success comparing high-performing clinics to low-performing clinics. Thirdly, we utilized the CFIR-ERIC matching tool to generate a package of optimized implementation strategies specifically targeting identified barriers to effective FB implementation.
Step One-Classification of Clinics Using the RE-AIM Framework

Procedure
The 36 clinics were classified into high and low performing at implementation as found in our study using the RE-AIM evaluation framework . Utilizing the support RE-AIM website as a guide, the research team designed indicators for the domains Reach, Adoption, and Implementation. The indicators were designed based on how the intervention was being delivered and how the selected RE-AIM domains are defined.
Given this, we focused more on how the intervention was being implemented and how far it had reached out to the targeted communities. As part of the first step of our implementation research study, we focused on which data had been collected since scale up to analyze reach of the program.
For this part of the study, performance data were collected by trained and supervised research assistants (RAs) in clinic visits (n = 36) through observations and individual interviews with the following key informants (n = 152): Community health workers, community health worker supervisors, nurses in charge, and district health promotions officers . On purpose we did not reveal the names of each group in high-and low-performing clinics to decrease social desirability and the chances of bias. The data collection exercise was carried out from June to August 2019 (3 months) by visiting each clinic for two consecutive days.

Analysis
Clinics were ranked according to their results per domain. Supplementary Table 1 shows all domain indicators that were used to classify the clinics into high-and low-performing groups. All indicators were weighed the same. All indicator values in each domain were added, giving each clinic a unique sum per domain on which they were ranked. Final ranking was calculated based on the mean of all domain ranks for an individual clinic (Farris et al., 2007). Based on the final ranking results, we selected the ten strongest and ten weakest clinics in terms of implementation performance as described in our prior publication  as focus group discussion sites to understand factors influencing performance.
Step 2-Applying CFIR to Analyze Barriers to Implementation

Procedure
As part of a participatory process and in preparation for the upcoming qualitative interviews, a stakeholder meeting was held in October 2019 in Harare with a selection of key stakeholders (n = 50) from the three cities to discuss step 1 findings and to determine key factors to consider in improving program performance. The stakeholders were from the 3 cities to allow a full representation of the communities and clinics under the study, including the various local authorities. Participants included ministry of health officials, city health officials, clinical nurses, community health workers, FB clients, and the full FB research team. In this meeting, participants were grouped by profession to discuss the initial findings of clinic rankings and potential reasons for heterogeneity in implementation success.
Following this stakeholder meeting and based on the input we received, the research team met and selected CFIR constructs with high potential contextual relevance based on stakeholder feedback. Constructs selected and excluded are shown in supplementary table 2. The research team, led by the international researchers, identified constructs that were identified by at least 50% of the stakeholder groups. We utilized the models available on the CFIR website (Consolidated Framework for Implementation Research, n.d.) as a starting point and made modifications for local applicability and relevance to design final focus group discussion guides (supplementary file 1; supplementary file 2).
Focus group discussions were carried out in nine highperforming and in four low-performing clinics in two districts (Harare and Gweru) (Supplementary Table 3). We had hoped to conduct more focus group discussions; however, the SARS-CoV-2 pandemic hindered data collection efforts. We conducted separate focus group discussions (n = 25) in which we met with 152 participants, 99 implementers (these included community health workers, nurses, and district health promotion officers), and 53 clients. Clients who had attended at least one follow-up session were invited for the focus group discussion.
Focus group discussions for implementers and clients, respectively, were carried out on the same day at the health facility. Data collection occurred from February to March 2020. Each focus group discussion took an average of two hours to complete. Focus group discussions were conducted in Shona by the lead author (CC) who has 4-year experience in conducting qualitative research in various community projects. The FGDs were audio recorded and then transcribed and translated to English by research assistants who were trained in qualitative data collection by a local qualitative researcher and were supervised by the lead author (CC). Transcripts were reviewed and compared with recordings and, if needed, translations were rectified by the bilingual translation team. The exercise was completed by April 2020.

Analysis
We developed a contextually relevant code frame using the Consolidated Framework for Implementation Research (CFIR) code book and CFIR memo template from the CFIR website as guiding frameworks. Individual focus group discussion transcripts (separated by implementers and clients) were summarized into facility-level CFIR memos. CFIR Memos with a heterogeneous score were reviewed by a third person who supported consensus finding. For each facility and focus group discussion group pairing, each CFIR construct was rated on whether it enhanced or hindered the implementation of the FB intervention.
Each construct included a rationale for the rating and key focus group discussion quotes. Valence ratings followed suggestions outlined by Gimbel and colleagues (2016), ranging from − 2 (strongly hindering implementation), − 1 (weakly hindering implementation), 0 (neutral influence), + 1 (weak positive influence on implementation) to + 2 (strong positive influence on implementation). A total of 4 pairs of RAs reviewed the 25 focus group discussions in subsets and gave valence scores for each CFIR construct. In total, we conducted 50 analyses. Valence scores were averaged across the two analysts for each focus group discussion, and this was the final valence score that each CFIR construct across each clinic-group combination (implementers vs. clients) received (Table 1). This process was done for each individual focus group discussion transcript: all received a final summary score. Development of memos was finalized over a period of 3 weeks in April 2020.
Following this, groups were sorted into (1) Constructs where both high-and low-performing clinics scored well (high and low performing with an average valence > = 1); (2) Constructs where both high-and low-performing clinics scored low (both with average valence < 0); (3) Constructs where high-performing clinics scored well and low-performing scored low (constructs with distinguishing (difference > 0)); and (4) Both high -and low-performing clinics scored in a moderate range (combined valence between 1 and 0).

Matching Identified Barriers to Implementation Strategies
We input the seven barriers to implementation identified through step 2 into the Consolidated Framework for Implementation Research-Expert Recommendations for Implementation Change (CFIR-ERIC) matching tool to identify implementation strategies to overcome known barriers (Consolidated Framework for Implementation Research, n.d.). The CFIR-ERIC matching tool helps to identify implementation strategies to overcome identified barriers (Weir et al., 2021). The research team utilized the cumulative percent to identify a "menu" of potential strategies. From this, 4

Results
This study focused on the stakeholders' perception of the program implementation. We chose the relevant CFIR constructs for this study based on the stakeholder's input following our presentation and discussion of the study's step one results. These results from the classification of clinics using the RE-AIM Framework showed, for example, that there was substantial variability in performance across domains; for example, some clinics that had performed well in the implementation domain, scored low in reach . Based on these results, stakeholders were engaged to discuss their thoughts on the differences in clinic performance across RE-AIM domains. After stakeholder and research team input, a total of 25 CFIR constructs were included as relevant from the implementers' perspective, while 9 were named by clients. The CFIR constructs for the clients' discussions were derived from the group of constructs that were included for the implementers (see Supplementary table 2). A few constructs were determined by stakeholders to not be relevant for FB scale-up or inappropriate for the setting. These were excluded, namely trialability; cosmopolitanism; tension for change; and opinion leaders (Supplementary Table 2). Table 2 shows all constructs and their influence on implementation. Of the 25 CFIR theoretical themes transpiring from both the implementer and client focus group discussions, six were identified as barriers for low-performing clinics and enablers for high-performing ones ("group 1: distinguishing constructs"). Two constructs were identified as "universal barriers" (group 2) in both high-and lowperforming clinics (Table 2). Nine constructs were identified as "universal enablers" (group 3) for implementation in both high-and low-performing sites. The remaining themes (n = 8) were not seen as relevant for the implementation success by the key stakeholders. For constructs such as learning climate, only high-performing clinics were discussed because there was no evidence of a learning climate in the low-performing clinics, and we were focusing on universal barriers as opposed to individual barriers.

Distinguishing CFIR Constructs
These constructs were of primary interest to our team as they facilitated implementation in high and acted as barriers in low-performing sites. The constructs were peer pressure, goals and feedback, learning climate, planning, formally appointed internal implementation leaders, and external change agents.

Outer Setting
Peer Pressure The theme of peer pressure was consistently reported as a key barrier to implementation strength in low-performing clinics. When delivering agents felt that they did not have information on performance of other FB sites, they felt less motivation to improve. One of the community health workers in a low-performing clinic explained: "We want to hear that information, so that if others are performing well, we also copy them…we don't know anything because we are working in the dark, maybe we are doing the wrong things." In high-performing clinics, knowledge about other FB sites' activities was found to be motivating and encouraged "We then look down upon ourselves and consider ourselves as people of a low status because no one will be coming to give us feedback saying: "Ah, this month you worked hard. That month, why has your performance gone down?" So, because of that we then look down upon ourselves." In high-performing clinics, although also lacking clear FB work-related goals, community health workers reported that they received consistent feedback on their FB work from nurses, city health representatives, and the FB implementation team. Community health workers described feeling acknowledged for their contribution to the overall primary health care clinic work. Community health workers highlighted that receiving feedback helped them self-reflect, evaluate, and improve their work.
Learning Climate In high-performing clinics, the community health workers engaged with each other and their supervisors. The community health workers felt seen by the clinic staff. The clinic staff recognized that they benefit from referring patients to the FB program. However, clinic staff felt that the community health workers needed further training which they could not provide. The nursing staff of clinics were rather removed from the (external) FB training and expressed that it should be FB's responsibility to offer ongoing training for the community health workers to improve their performance. This nurse explained: "…. for you to conduct a refresher course for them [community health workers], I think it's very important you saw when you came, they had forgotten some of the steps (Problem-Solving Therapy) or some of the things, so I think it's ideal to conduct a refresher course..." A delivering agent in a high-performing clinic described a skill-based and more helpful scenario about her FB supervisor trying out ways to supervise her colleagues: "Alright, the supervisor also helps when we are writing. We help each other to check if we are doing what we were taught and doing corrections here and there."

Process
Planning In low-performing clinics, stakeholders highlighted how the FB is perceived as a program that is run by the city health authorities, and stakeholders are feeling less involved in decision-making. Clinic-level delivering agents and beneficiaries did not feel included in the process of the intervention. One of the community health workers from a low-performing clinic highlighted this by saying: "We were not there when it was planned, we were just told." A FB beneficiary from a low-performing clinic shared her opinion about the importance of engaging communities to increase visibility and acceptability of the program as part of the planning process: "People just see them [the FB community health workers] but they don't know what they do. There were women with green branded zambias [a piece of traditional cloth material used to wrap around the body], I asked someone "who are those people with green zambias and what do they do?" A community health worker from a high-performing clinic highlighted how the clinic-level FB delivering agents felt included and involved in the planning process of the FB intervention. They explained how the program was clearly described to them in the beginning and this led to a higher level of acceptance, saying "They first explained about FB when we were at a conference, do you still remember? [all agreed]. They then wrote down our names saying they will want to work with us. That is when we knew that there was FB. They then came and put us in small groups and asked us to go and learn how to deliver the FB together until the program spread everywhere." Formerly Appointed Internal Implementation Leaders An example of internal implementation leaders are the supervisors who are also doing the FB work (community health worker supervisors). Community health workers in lowperforming clinics indicated that there was no clarity on how peer supervisors were selected and that their roles and responsibilities were not clearly defined. This created confusion about roles and responsibilities for leadership and man-1 3 agement of the FB program. A community health worker from a low-performing clinic described her disappointment at how her supervisor was selected: "This exam was not given to us all, only one member from [clinic name] was selected and went for an interview…'we were told that so and so had passed and they were now our supervisors." These experiences in low-performing clinics stood in stark contrast to the process of selecting implementation leaders in a high-performing clinic as one community health worker explained: "We wrote a short exam then they called...how many were we…we were three. Then they called us and started asking us questions like what you are doing right now. So, l think on those questions that is when some excelled more than others." External Change Agents Community health workers in lowperforming clinics highlighted that they felt left out without support from their external superiors. For example, a community health worker described the situation in her clinic: "So when he [the nurse in charge] came here, he found FB already operating and did not join the training." In contrast, a community health worker of high-performing clinics described how other external change agents from the health department were involved in the FB implementation: "From Rowan Martin (name of office building of Harare City Health Department) they come to see how we work, and we gave them information on how we work. They know that these people are doing their Friendship Bench work and so on, they in a way supervise our work, they follow up on every work that we do, how it is going."

Intervention Characteristics
Complexity of the Intervention While most elements of the FB intervention and the program itself were viewed favorably, there were elements of personal involvement required to offer the intervention that made it highly complex in the eyes of community health workers and which were seen as barriers. One community health worker described the stress dealing with clients presenting with complicated mental health problems: "It is not easy to work on the Bench, sometimes we are overwhelmed by some of the complicated cases we must handle. So much so that sometimes we even find ourselves in need of counseling. Sometimes the client will be of the same age as your own child. So, it is painful. What they are disclosing to you will be almost the same as having a client with a red flag [suicidal behavior], so it will be really touching." Community health workers described how screening, guiding the clients through the Problem-Solving Therapy, and documenting the process takes more time than anticipated. One community health worker described this by saying: "You can spend one and half hours or even two hours with a client, some need time to cry. So, we must be calm and patient enough to set aside everything else and give them time." Speaking to clients who presented with high levels of emotional disturbance was seen as difficult, and community health workers expressed feeling stressed and sometimes unable to handle these cases. The FB intervention is based on community health workers applying counseling skills. They have been trained to provide a therapeutically valuable response to a person who is in mild to moderate distress. So, when clients stated they needed their session to be completed in a short space of time, instead of spending a long time in a discussion over their problem with the community health workers, community health workers reported feeling additional stress. Specifically, shorter sessions impeded the counseling process as, to be successfully applied, the community health workers need sufficient time to go through the FB intervention.

Inner Setting
Organizational Incentives and Rewards Key informants talked about how essential incentives and rewards are to a strong program implementation and to keep work morale high. Community health workers are based in the same community as their clients, they face similar pervasive life stressors such as poverty which led to a community health worker expressing her frustration about the stipends they receive from their employer. She nevertheless points out that they are dedicated to their work: "It's too little [money], we cannot even afford to buy mealie meal to cook, even cooking oil or soap... [Despite this], we tell [the FB visitors] that we are always motivated and we [try to] come to work stress free not thinking about the situation we left home." Non-financial incentives such as T-shirts with the FB logo were seen as attractive and important to motivate the community health workers as they associate it with being formalized through wearing work-related attire. Over and above their City Health uniforms, one community health worker highlighted that wearing a T-shirt with a mental health message gave them authority and recognition which is a non-financial reward: "So, if we have a t-shirt that shows that we are from Friendship Bench, clients will listen to us." Overall, the described barriers reflect the self-perceived ability of delivering agents to offer the program and the recognition of their work, financial, and non-financial. Both aspects need to be addressed to optimize the implementation.

Characteristics of the Intervention
Constructs related to the FB intervention were almost universally rated as facilitators (4 of 5 constructs) showing that the intervention itself had a positive influence on the implementation in our setting. Overall, the positive views of the program were highlighted by all interview partners as contributing to success across both high-and low-performing clinics. The program's perceived evidence strength and quality, relative advantage compared to other potential interventions, its adaptability to the local context, and its design quality and packaging are described below.
Evidence Strength and Quality FB stakeholders, including clients, universally believed that the FB intervention is effective in helping individuals deal with common mental disorders. This positive attitude toward the program was found in all sites, independently of their performance level.
As an example, seeing clients having a positive view of FB, one of the community health workers stated: "According to our own observations many people like the Friendship Bench because you hear them saying: If you have any problems, go and see those grandmothers. They will encourage you to sort out your life… those grandmothers are doing a great job trying to sort out what will be troubling you in your life." Relative Advantage One of the nurses pointed out how important the FB was in addressing health and psychosocial problems that clinic users were facing. This showed how the program serves as a great adjunct integrated into the other programs at the primary health center. One of the nurses had this to say: "They [community health workers] are aware that when patients or clients come to the clinic, they have physical problems and the medical staff on the ground may not necessarily be paying attention to the mental health of the patients. Therefore, they are really a good hand in the clinic for the psychological support of the patients. They are also there in the community offering psychological support to clients who come forward on their own." Adaptability B stakeholders reported that the FB program, which is a locally developed intervention, was easily adapted to fit into microlevel local contexts, such as overall level of clinic activities determined by clinic size. Large clinics were able to connect their service delivery to FB by referring clients for mental health care. Community health workers were also able to adjust their service delivery to their clients' needs and referred to the medical care provided at the clinic.

Design Quality and Packaging
The design quality and packaging of the FB program were seen very positively by the key stakeholders. Community health workers reported that the training material was user-friendly and well designed. The cultural adaptation of Problem-Solving Therapy into Shona terms was successful and made it more feasible to explain to clients. They explained being comfortable with the model to guide their clients through the step-by-step process of Problem-Solving Therapy. It was highlighted that the FB tools are helpful in tracking the session progress.

Inner Setting
Networks and Communication Open and clear communication among community health workers with the clinic staff was found to be necessary. From the perspective of community health workers, clinic leadership showed their interest by taking part in clinic meetings when the program was presented as part of the primary health care clinic services. When the nurses are involved in the service delivery procgood becausess by referring clients, community health workers see it as approval of their work, and they engage in collaborative and holistic care: "Our relationship is very e the nurses come across some problems in the clinic and say: "No, let me send these people to the FB and ask the grandmothers to talk to the client." "Because of their workload, they don't get time to attend to the client. Then she sees that the grandmothers are there outside who do that work. And she brings the client. And we then give her feedback [about the client]." Internal networking is highlighted by this community health worker mentioning mutual practical skills support among colleagues: "We can help each other if, let us say, l don't know how to write something we can help each other as a group. The meetings also help us remember things we would have forgotten; we assist each other." If the medical clinic staff was not seen as contributing to the FB program nor communicating well with the community health workers, the community health workers focused on peer support and saw the value of working as a team.
Relative Priority Community health workers saw the importance of mental health care being offered in the primary health care clinics, especially as FB is the only program focusing on mental health. They perceive medical staff who refer clients for psychological support to them as sharing this view.
Community health workers also see how they contribute to strengthening the overall primary health care service by offering mental health care and by working with the bidirectional referral pathway to ensure continuous care as highlighted below: "They [nurses] help us by bringing clients to the bench if they see he/she has got a problem." One community health worker explains how she feels assured that she can refer clients to the nurses: "Yes, they make it easy, especially the nurses, when l come across a red flag l always refer to them and they help the client."

Leadership Engagement
Besides referring clients to FB, clinic leadership supported logistical needs. One community health worker experienced a nurse making clinic furniture available: "I have noticed that they value Friendship Bench, from the sister in charge to the other nurses here. If we have run short of benches since one of our benches is not yet fixed, they just say take some chairs inside. We take those chairs, sit and counsel our clients." Clinic staff can emphasize that FB benches are neatly kept outside as one community health worker points out: "Even those that will be doing community service, they [the nurses] tell them to slash the grass, so it means that they value Friendship Bench." These examples describe an integration of the FB program driven by individual clinic leads, without them having been formally appointed.

Individual Characteristics
Individual Identification with Their Organization This theme was seen as an enabler across all clinics. Besides seeing the intervention positively, statements like the following describe how a community health worker felt connected with the program. Being given tokens of appreciation by the FB team and to belong to a recognizable team is highlighted in this statement: "Giving us those wrapping cloths (Zambia) and umbrellas, those things made us happy. I went for a funeral, and I was advertising my umbrella which was written Friendship Bench. You can advertise that umbrella because it has a message to people." Delivering agents want to be recognized for their FB counseling work. Not only do they feel appreciated, but they also understand this to be helping the program visibility that they are proudly supporting.

Process
Engaging Community health workers found engaging clients easier when they were clearly recognizable as FB counselors. Wearing program-specific t-shirts is common in Zimbabwean community health programs. Community health workers expressed that they felt more engaged with the program when they were seen by their clinic's lead and their work was integrated in the clinic's program as well as when they received recognition from stakeholder groups, including "beneficiaries."

Matching CFIR Determinants to Implementation Strategies
Six CFIR Constructs (n = 6) that were identified as barriers across both low-and high-performing clinics were matched with the 73 strategies. Five of the CFIR constructs were found to be barriers in low-performing clinics (see Table 1 above) and the sixth was highlighted in both high-and lowperforming clinics.
Sixteen strategies were suggested by the Expert Recommendations for Implementation Change (ERIC) matching tool to potentially address the 6 CFIR barriers. For the specific FB settings, time frame of our study, and availability of resources, 4 (top 3 and strategy number 9, listed below) were selected by the research team to develop for implementation. All other strategies (assessing readiness and identifying barriers and facilitators; conduct local consensus discussions; facilitating, recruiting, designating, and training for leadership; informing local opinion leaders; conducting local needs assessments; auditing and providing feedback) had been addressed by the FB team prior to the initial implementation and were therefore not applied.
A stakeholder meeting from three cities was used to present the selected strategies: (1) identifying and preparing champions; (2) developing a formal implementation blueprint; (3) altering incentive/allowance structures; and (4) organizing regular clinical implementation team meetings. Stakeholders confirmed the selection. They were engaged in the participatory development of the formal FB implementation blueprint. Due to the SARS COV pandemic, a modification to our protocol (Verhey et al., 2021) was required as the planned trial to test the strategies was not feasible. In contrast to the original plan, we abandoned the trail due to the time delay and continuing inaccessibility of clinics due to the COVID-19 pandemic. With the remaining time being insufficient for the trial requirements, we decided to conduct ethnographic observations from each performance group. The observations focused on the detailed service delivery in the community based on the ethnographic approach and will be reported elsewhere.

Discussion
To evaluate the Friendship Bench program post-scale-up, we used the CFIR framework, applied for the first time in sub-Saharan Africa in Zimbabwe. Additionally, we applied the CFIR-ERIC matching tool to thoroughly understand and learn about strategies to improve the implementation which was one goal of our implementation research study. Furthermore, the RE-AIM model was applied to understand the performance of FB implementation sites; results guided the present study which aimed to describe barriers and enablers affecting the previously identified differences in performance strengths . Our results show the importance of examining the implementation of an evidence-based intervention into the real world through involving the stakeholders and learning from their perspectives.
We included a large number of stakeholders as our scaleup had been carried out in three cities, and we needed a full representation of the communities and clinics as well as local authorities. As we found diverse program activity levels in the 36 clinics, barriers to the implementation were of great interest to us. CFIR constructs organizational incentives and rewards, and complexity were reported as universal barriers to implementation in high-and low-performing clinics. A lack of adequate remuneration for the work was seen as a major barrier, independently of how well sites performed. Various studies show how important it is to give appropriate incentives and rewards for community health workers (Maes & Kalofanos, 2013). Task shifting activities in health care are attractive for resource constrained health care setting, as they offer increased access to care and are found to be cost-effective (Siedman & Atun, 2017).
The main advantage of task-shifting is its ability to increase access to health care. Long-term success of taskshifting depends on financial and political commitment as well as planned integration and creation of regulatory framework (Lehmann et al., 2009). This is in line with the literature emphasizing the need for incentives that can be complementary, directly financial, or non-financial or indirect from the health system or community (Colvin et al., 2021). Furthermore, success of task shifted programs depends on community health workers motivation that is influenced by being valued, having clear roles and responsibilities, and opportunities for growth (Colvin et al., 2021).
When working on reducing barriers to improve a program implementation, it is important to leverage the determinants that influence the outcome positively. The universal enablers evidence strength and quality, relative advantage, adaptability, and design quality and packaging that were found in our study point to FB being recognized as a strong and attractive program for all stakeholders. Furthermore, we detected a group of CFIR constructs that reached opposing scores in the analysis depending on whether stakeholders were attached to high-or low-performing FB sites. We called these distinguishing constructs. They were seen as enablers in strong FB implementation sites and hindered the program in lowperforming ones, clearly showing the evolved differences four years after scaling up of the FB program. These differences were a key finding and the CFIR aided in explaining them for both high-and low-performing clinics. These factors were goals and feedback, learning climate, peer pressure, planning, formerly appointed internal implementation leaders, and external change agents. Where these determinants were seen positively, the FB program was active and well frequented and vice versa.
Community health workers are an important group of health workers and contribute to improved health care; however, multiple implementation factors can influence their performance in health care delivery (Kok et al., 2017). Our findings show that a strong implementation of the FB program is more likely when stakeholders feel integrated in the planning and implementation process. Furthermore, good intra-clinic communication and clear work expectations for community health workers are enablers, as all are expected to perform behavioral change at some level to integrate the program (Milat et al., 2015;World Health Organization, 2002). What made a difference between high-and low-performing Friendship Bench program sites was highly influenced by the people who were delivering the service, how involved and recognized they felt, and what levels of support they were given from the overall health care system.
Besides remuneration for their work, capacity and confidence building through appropriate trainings of staff and organizations, strengthening of interdisciplinary collaboration, and community empowerment are factors supporting program roll-out (Bennett-Levy et al., 2017;Dadi et al., 2021;Lopez-Carmen et al., 2019;World Health Organization, 2008). For a successful integration into health care and systems strengthening, all staff levels must be involved to increase leadership and support of an intervention (Colombini et al., 2017). Independently of whether other work aspects are addressed, these determinants are decisive for implementation strength and quality as they directly influence motivation and, in turn, recruitment, retention, and performance (Daniels et al., 2014). Implementation strength depends on critical mass of mental health specialists, inclusion of mental health indicators in the health information system, and strengthened governance (Petersen et al., 2019).
Our study has shown both enablers and barriers to implementation success, many of which have been found in other studies. The strategies for improvement of the program implementation suggested by the CFIR-ERIC matching tool 1 3 were found to be applicable to our setting and addressed the described barriers.
The research was not without its limitations. Data were collected between February and March 2022 in all high-performing clinics which were all located in Harare. However, we were not able to collect data in three of the low-performing clinics that were in Chitungwiza. Specifically, due to the initiation of the COVID-19 lockdown in Zimbabwe, we received data from four implementation sites instead of seven. Obtaining most of the information from providers attached to high-performing clinics possibly introduced a bias in our perception of the service delivery context. Future research should address equity in representation. We recommend that Friendship Bench pay ongoing attention to the work context of low-performing clinics as this affects the effectiveness and acceptability of the program. Furthermore, social desirability might have influenced how stakeholders responded.

Conclusion
This framework-driven analysis of the contextual determinants of the FB program revealed the barriers and enablers influencing FB since scale-up efforts in 2016. The combination of CFIR and CFIR-ERIC matching tool, within the RE-AIM framework, were valuable to select the most applicable strategies to improve implementation in future. The results of this study can guide efforts to improve mental health interventions, such as the FB. Careful evaluation of implementation of evidence-based interventions can allow them to be replicated in different contexts, including primary health care settings. To ensure high implementation quality, interventions such as the FB should apply a rigorous and data-driven approach in collaboration with key stakeholders, such as the national public health agencies.