How do patient feedback systems work in low- and middle-income countries? Insights from a realist evaluation in Bangladesh.

Well-functioning patient feedback systems can contribute to improved quality of healthcare and ultimately make health systems more accountable. We used realist evaluation to understand the functioning of patient feedback systems at frontline health facilities in Bangladesh. We collected and analysed data in two stages using: document review; secondary analysis of data from publicly-available web-portals; in-depth interviews with patients, health workers and managers; non-participant observations of feedback environments; and stakeholder workshops. Stage 1 focused on identifying and articulating the initial program theory of patient feedback systems. In Stage 2, we iteratively tested and rened this initial theory, through analysing data and grounding emerging ndings within substantive theories and empirical literature, to arrive at a rened program theory.

managers can be both principals and agents, and how information asymmetry and agency loss can be addressed.
Practically, we demonstrate the importance of awareness raising and non-threatening environment to provide feedback, adequate support to staff to document and analyse feedback and timely actions on the information.

Background
Effective interaction between patients and their systems is critical to well-performing health systems (1)(2)(3)(4), and functioning patient feedback systems can contribute to improved quality and responsiveness of health services and ultimately good governance and accountability of national health systems (2,3,(5)(6)(7)(8)(9).
Patient feedback systems can be framed within two theoretical perspectives. First, providing feedback is an act of citizenship involving social identity and practices by people with different capacities and resources within political and social structures and institutions (10)(11)(12)(13). Second, multiple principal-agent relationships between healthcare providers, patients and managers, occur within contexts of entrenched bureaucratic and professional hierarchical roles and relational dynamics (14)(15)(16)(17).
Effective feedback systems involve two key features. First, the supportive environment for patients to provide feedback on their experiences (1,5,18). Second is the health system's ability to adequately respond to, and act upon feedback (1,6,8). Substantial research covers patient feedback systems, their typologies, assessments and contributions to service quality improvement (1,6,9,(19)(20)(21)(22)(23). Most of this evidence comes from high-income countries and hospital settings with less research from lower-income countries and frontline health facilities.
We report results of a realist evaluation of patient feedback systems at local-level health facilities in Bangladesh, addressing overarching question: what about the patient feedback systems has worked, for whom, in which circumstances, and why? Our ndings should be of interest to academics and practitioners engaged in advancing the understanding and improving, patient feedback systems.

The Context
Bangladesh comprises 8 Divisions, 64 Districts, 481 sub-districts (Upazilas) and 4,403 unions. Most health budget is earmarked for Upazila Health Complexes (UHC), making them a backbone of the country's public health system. UHCs serve a population of 200-400 thousand, offer both in-patient (31-50 beds) and out-patient services, and act as rst level referral for community clinics and village (union) health centres. UHC has between 93 and 128 staff, including 9-20 doctors, 13-16 nurses, 2 pharmacists and 2-5 laboratory technicians (24). UHCs provide preventive and basic curative services, have an ambulance and a pharmacy. A health management committee comprising local politicians, facility managers, civil society representatives and local leaders, monitors UHC work.
Improving health system's accountability to the population is high on the policy agenda (25). An overarching framework, known as a Citizens Charter, summarises patient rights within public health facilities (Fig. 1). Figure 1 hereMultiple centrally and locally-managed patient feedback systems operate at UHCs (Table 1). Their common strength is the underlying political commitment to enabling citizens voice and accountability, whereas common weaknesses include unclear processes and limited promotion of available channels. Unlike the locally-managed, the centrally-managed systems tend to have more functional record-keeping.

Methods
We conducted a mixed-methods realist evaluation of patient feedback systems at two UHCs. A realist approach starts and ends with a Program Theory (PT) which articulates the logic of the program (26).
PTs represent hypotheses that can be tested, re ned and consolidated. Researchers interrogate their initial PTs through identifying causal pathways of how speci c mechanisms (reasoning and resources) are triggered in different contexts, to produce (un)intended outcomes. These pathways are articulated as Context-Mechanism-Outcome (CMO) con gurations (27)(28)(29). The re ned PT is based on the evaluation about what aspects of the intervention worked, for whom, in which conditions and why. RAMESES standards guide reporting of realist evaluations (30), which also guided this paper (see Additional File).
The study was conducted in two UHCs of Comilla district which neighbours the capital Dhaka, and has one of the highest feedback rates. This selection was based on: (a) analysis of publicly-available web portal containing data on patient feedback and (b) following non-participant observations of patient feedback environments in outpatient and emergency units and female and male patient wards (31). These observations lasted between 30 minutes and 2 hours and involved re ecting on the degree of visibility, user-friendliness, utilisation and maintenance of key feedback channels in the UHCs such as suggestion boxes or telephone hotlines. A semi-structured observation checklist highlighting presence, usage and maintenance of feedback channels at UHCs was used.
We collected data in two stages using combination of different methods ( Table 2). A retroductive approach to data analysis was used, combining inductive and deductive logics to identify hidden causal mechanisms through iterative engagements with the data, literature and programme theory (32, 33).  Table 2 here Stage 1 was exploratory and focused on identifying the Initial Programme Theory (IPT), drawing on different methods, to elicit the perspectives of policymakers, facility managers and patients, to understand the logics behind patient feedback systems. We began with a review of design-related documents; practice guidelines, which articulated processes of the feedback system and the different roles involved; job descriptions of involved personnel; internal reports; and news items about patient grievances. In total, 18 documents were reviewed; these were obtained from Directorate General of Health Services (DGHS), Ministry of Health and Family Welfare (MoHFW), and through a web search.
We then conducted in-depth interviews (n = 5) with key actors (public representatives of the Upazila council, policymakers, facility managers) to understand their perspectives and further developed the IPT. All interviews were transcribed verbatim and uploaded into NVIVO for coding by the contexts, mechanisms and outcomes. With this version of the IPT, we conducted a stakeholder workshop to glean their understanding of how the feedback system was supposed to work, and elaborate speci c CMOs. Throughout Stage 1, we explored established theories which could help frame the logics underpinning patient feedback systems and the speci c CMOs.
In Stage 2, we tested and re ned the IPT. This drew upon the observations and in-depth interviews (n = 20) with health staff and patients. The IPT provided the basis for the interview guide and for the checklist for observations. The interviews were also oriented towards interrogating the veracity of aspects of the IPT. Each interview lasted 25-60 minutes, was audio-recorded and transcribed verbatim, translated into English where required and uploaded into NVIVO where coding was now more driven by elements of causality within the CMOs. Two researchers conducted the observations. The checklist focused on the presence of feedback materials and processes, and on the functioning of the feedback system.
At this stage, we conducted two stakeholder workshops. These lasted 4-5 hours and each involved 20-25 representatives from government, non-government and international organisations. Workshops included presentations of emerging ndings, followed by a plenary discussion, and then more in-depth work in smaller groups to validate results. While the primary aim of these workshops was to share and validate emerging results with key stakeholders, proceedings were audio-recorded following informed consent and were treated as further data for analysis.

Results
Our IPT ( Figure 2) was gleaned from the iterative review of documents, perspectives of key stakeholders and the literature.

Figure 2 here
It was framed within two theoretical perspectives. First, providing feedback and expecting improvements are acts of citizenship. As Lister explains(34 p41) "To be a citizen in the legal and sociological sense means to enjoy the rights of citizenship necessary for agency and social and political participation. To act as a citizen involves ful lling the potential of that status". Thus, citizenship is both an identity and a practice (10,11,35). Understanding of identity shapes one's exercise of citizenship within social spaces (home, community, institutions, national politics, the global arena). People have different capacities and resources to express their citizenship identity within contexts of socio-political opportunities through available places and spaces (10,12,13). Expressions of citizenship include political forms such as voting in elections (10,(35)(36)(37) and patient feedback systems provide platforms for people's engagements in decision-making within health facilities.
Second, the Principal-Agent (PA) theory helps understand multiple relationships. It postulates that all organisations require employer-employee cooperation (14). Using the metaphor of a contract, the PA theory highlights the agency in the relationship where the principal delegates work to the agents (15,16).
Central to this are the goal con icts and differing preferences within hierarchical relationships. Two assumptions contribute to potential agency loss, a common metric for determining whether agents act in the principal's interests (16): diverging and independent interests of each party and an information asymmetry with agents being generally more knowledgeable of the local circumstances including their efforts and capacities. The PA theory seeks the most e cient contract with assumptions about the individuals (e.g. self-interest, bounded rationality, risk preferences), the organisations (e.g. goal con icts), and information (i.e. acquired commodity leading to information). Patient feedback systems involve relationships between three groups (patients, healthcare providers and managers) within contexts of information asymmetry and potentially diverging expectations.
Three speci c CMO con gurations underpinned our IPT, corresponding to three steps in feedback management (23): 1. Collecting feedback: In contexts where people are aware of their rights to provide feedback and perceive that they will be heard, and not penalised, if the means to provide feedback are accessible, it will trigger amongst them a sense of trust and con dence in the system, translating into exercise of agency for greater use of feedback channels.
2. Processing feedback: In contexts where there are well-functioning health facility management committees, if clear policies and guidelines for service quality improvement exist, feelings of being supported and not threatened by patient feedback will be triggered amongst health workers, leading them to value the feedback, to discuss it openly, and to share it transparently within the UHCs.
3. Acting on feedback: In contexts of available guidelines and regulations that promote integration of feedback into service quality improvement and internal accountability, if adequate supportive processes are given to health workers, it will trigger amongst them a sense of being respected, motivating them to value the processes of feedback management and to act upon patient feedback.
These CMOs were iteratively tested and validated through against the data and the literature.

Collecting feedback
The interviews revealed that patients were generally aware of their rights to express opinions. Most managers concurred and shared the view that while feedback systems existed at UHCs, patient awareness of both their rights to and of available feedback channels, was limited. Service providers re ected that lack of awareness was due to a mix of low literacy and insu cient system's efforts: "Basic reason for low awareness is poor education. Another reason is information gap... they do not know about the systems." [Provider: 003] The feedback channels that existed, were not accessible. Observations revealed that the information about the SMS system was on a whiteboard on a wall, at a not easily visible height (  Social access to feedback channels was often more important than physical. In Bangladesh, the socioeconomically disadvantaged often perceive themselves as not deserving of raising voice against the unfairness they encounter, as illustrated in the following quotes.
"A patient like me is not likely to discuss such issue... If I say something, they wouldn't listen. It would prove helpful, but I do not think I can have the capacity …" [Patient: 006] "People like you and me ... we know that we can protest against anything wrong... However, those who come from villages … they simply accept the mistreatments in silence. They do not even know how to complain or whom to complain to. If someone asks for a bribe, they simply bribe the person to get help." [Patient: 007] In the second excerpt a socio-economically well-off patient explains these class differences. This poverty of agency, and the lack of access amongst those at the bottom of the socioeconomic hierarchy is well recognised in Bangladesh. However, the multiple feedback channels do not su ciently recognise this reality and fall short in supporting feedback from most vulnerable.
Our analysis also revealed low levels of trust in the feedback processes at UHCs and the wider health system. Patients' distrust and their resultant hesitation to provide feedback, were rooted in their doubts about the benevolence of the health system and in fear of consequences: "Because when we try to say something, we are afraid of not getting treated properly or be harassed. There is always a fear and we do not say anything. Fears of retribution and distrust of the system's ability to act fairly, consistently featured as logics underpinning people's decisions to not provide feedback. While some expressed this openly, others were more reticent. Many providers recognised this, but proffered rather simplistic explanations effectively dismissing these fears as unfounded.
"They have no reason to fear. It is due to ignorance and lack of education. Sometimes, when the patients make verbal complaints, we advise them to place written complaints. However, they do not agree." [Provider:003] Such explanations re ect a deep disconnect between providers and patients. They spotlight entrenched prejudices and the class divide whereby the privileged inhabit public systems and view the underprivileged as being ignorant. Such social relational dynamics are fundamental to people's distrust of the health system, manifested in multiple calls from patients for anonymization. The non-anonymised feedback systems are therefore under-utilised. Consequently, people provide feedback through their acquaintances.
"Most of the time, they express their opinion to those persons who are very familiar to them. Suppose a sweeper or cleaner…" [Provider: 004] As the quote illustrates, personal connections were preferred feedback routes. While this re ects low levels of trust in the system, it could also re ect preferences for relational ways of interacting in the society, rather than processual bureaucratic ways that current feedback channels offer. When queried about a possibility of low trust being the reason for the limited uptake of feedback channels, providers pointed to the high use of services as a counterargument. While this was plausible, it is more likely that the high service use merely re ects lack of alternative healthcare options.
Our ndings are consistent with the literature which highlights awareness about rights as a prerequisite to exercising one's agency (38)(39)(40)(41)(42). Our ndings suggest that while being aware is necessary, mere awareness is not enough for exercising agency and rights, echoing the literature on lack of complaining by the socio-economically disadvantaged in Nepal, Russia and Israel (19,41,42). Fears of retribution and distrust of the system consistently underpin people's decisions to not complain (19,(41)(42)(43)(44)(45). Patients distrust available feedback systems because they doubt the benevolence of the health system and are afraid of retribution. Conversely, if people were to perceive that they will be heard and not penalised for their views, they are likely to trust the system and use available channels. This suggests that our initial CMO which posits that awareness of rights combined with accessible channels, will trigger a sense of translating into exercise of agency, is generally valid.
Preferences of informal feedback have been widely reported across public services (7,(46)(47)(48). This suggests that giving feedback involves culturally-mediated processes governed by local social norms about acceptable conduct. This is particularly so where formal processes are weak or trust in the system is low, or where there is general preference for relational ways of interacting.

Processing feedback
The centrality of clear policies, guidelines, processes and roles and fostering overall system's accountability and transparency, emerged as a key nding. One provider emphasised that "Of course, a guideline is needed. Without a guideline, the process cannot be maintained in an organised way". This limited clarity was a critical missing link that led to different interpretations of and practices around feedback management. It also seemed to underpin the apparent lack of shared goals between policymakers, local managers and providers: "I think it is also important to know who is designated for which work. It is necessary to know who has what authority. There needs to be a guideline. For instance, detailing where to begin and whom to go to; what is the process to arrive at a resolution" [Provider: 010] These views were echoed by patients, particularly those with good knowledge of the healthcare system: "…I would want to know: who is in charge, who will work on it, how will they, then who is going to solve it? If it is solved, how will they inform the patients? Every step should have speci c guidelines" [Patient: 010] Clarity about roles and responsibilities could help bridge the information asymmetry gap between the patients and the healthcare providers, ultimately enabling people to exercise their agency to provide feedback. Multiple providers consistently viewed this dysfunction as a management failure. They pointed to an ad-hoc nature of feedback management, and how "Complaints have never been taken by us positively"; expressing displeasure "If a process existed, then our good o cers would not suffer like this."; and articulating the desire for fair and non-punitive feedback systems "If there were guidelines… regarding complaints, and if it were clearly written, I think that would bring some transparency to the process…".
These quotes also highlight the cooperation challenges between the UHC leadership (the Upazila Health and Family Planning O cers, UHFPOs) and health workers. Given the nature of healthcare delivery, UHFPOs do not always know the details of strengths and limitations of efforts by providers. UHFPOs, perhaps understandably, use patient feedback as a lever to extract accountability from staff. The absence of transparent processes, particularly documentation of feedback, means that such tactics by UHFPOs attract resentment from staff. The earlier quotes suggest that in the context of weak feedback management processes, the UHFPO's tactics undermine the learning potential from patient feedback for service quality improvement. Instead, the disconnects between providers, patients and managers fuelled resentment amongst staff towards feedback systems.
A key constraint to feedback processing was the unavailability of dedicated staff with relevant expertise.
Observations revealed that UHCs were generally understaffed and feedback management was an additional responsibility for clinical staff. Managers and health workers recognised these constraints.
Providers argued that "An extra person be recruited and given the responsibility to maintain Many providers explained this disengagement to the absence of mechanisms for members to exercise their authority in the committee. However, this explanation does not hold up given that most members outrank the UHFPO and have no need to favour him/her. The most plausible explanation is that local leaders do not see su cient political value in these committees. This is consistent with the low political priority of health in Bangladesh. This overarching contextual issue upends a key context-related assumption of our IPT.
Literature shows that clear quality improvement guidelines and supportive policies enable wellfunctioning patient feedback systems (44). Similar to our results, studies have shown that frontline workers need support to manage feedback effectively and that adequate supervision can help staff value patient feedback (6,43,49,50). However, the mere existence of guidelines is not enough; frontline workers need to be aware of feedback management processes. They require skills to deal with often di cult interactions (49). Together with our ndings, this supports our initial CMO that highlighted clear policies and guidelines as triggering health workers' feelings of being supported, not threatened and valuing patient feedback.
However, we found that a critical contextual aspect of our initial CMO was missing. UHC management committees were non-functional with non-involvement of local leaders. Substantial literature examined the conditions under which local social accountability structures can ful l their mandates (51)(52)(53)(54). It spotlights that constructive and sustained local political involvement is a key to active local accountability structures, and to improvements in healthcare quality, responsiveness, and equity. This counterfactual analytical rendering of what was amiss in our study context thus supports the logic of our initial CMO.

Acting on feedback
Robust regulatory framework and institutional support were seen by all interviewees as being crucial to enable staff to act upon feedback. This view was echoed in the stakeholder workshops. Participants agreed that training of providers, clear Terms of References or guidelines, explicit roles, and resources were critical to effective feedback management. While there were plans to introduce such frameworks by the MOHFW, none were yet in place.
During the interviews, most providers revisited their initial resentment towards feedback systems. Many re ected that "Through the feedback, at least our work would get (some) appreciation", recognised its value in being able to "praise the good performance and to punish misdeeds", and appreciated the learning opportunities from feedback, saying that "[in response to feedback] exemplary action should be taken so that with one example, others become cautious.". Many added that any effective regulatory framework should include an appropriate balance between incentives and sanctions, arguing for links with staff appraisal, rewards and recognition.
Patients accorded high importance to the user-friendliness of feedback channels and that actions on feedback ought to be transparently communicated to enhance the credibility of the feedback.
"It is better to inform patients because in this way they will get to understand that through this system they solved my problem. This patient will spread it to others." [Patient: 005] This centrality of open, transparent communication between service providers and users to creating trust and improving staff-patient relationships was also consistently recognised by providers.
"…if we can inform the patients about the solution, they will be pleased thinking that their complaints led to some solutions. By being happy, they will encourage their neighbours, thinking that problems are being solved and communicated well. It will improve the relationship between hospital and patients." [Provider: 001] The literature echoes our ndings and adds that while policies and guidelines can catalyse action on feedback, unsupportive institutional cultures and ineffective communication skills of service providers may hinder the desired effects (8,55). Furthermore, a receptive and learning institutional culture can help staff recognise the value of transparent and fair feedback management (56). Consistent with our results, scholars have found that transparency in feedback management can enhance the health system's credibility and foster patient trust (57). This con rms our initial CMO which linked clear guidelines and processes along with support to facility staff, with a sense of being respected and motivation to value and act upon feedback.

Discussion
Our overall PT (Fig. 5) was iteratively tested and re ned throughout data collection and analysis, at stakeholder workshops and consolidated against the literature. Figure 5 here The testing of CMOs revealed the contingent nature of triggering of mechanisms, and the variable achievement of outcomes. Our revised PT highlights this contingency, serving three objectives: rst, it helps show the many ways in which feedback systems (not) operate to (not) achieve their intended outcomes (26). Second, it speaks to a central tenet of critical realism -of causality (58). Thirdly, the 'ifthen' propositions can serve as a practical heuristic for informing future interventions. The contingent nature of contexts, interventions, mechanisms, and outcomes, further highlights the interconnectedness of the three steps of the feedback process.
Awareness of feedback channels and people's trust in the health systems are important determinants of people's citizenship identity and willingness to exercise their agency. Awareness and trust are, however, insu cient and need to be bolstered by recognition of rights to provide feedback, accessibility of feedback channels, clear policies and guidelines, and appropriate incentives and sanctions to ensure staff compliance (1,5,23). In line with published evidence (1,5,8,59), a key determinant of decisions to provide feedback is people's con dence not to be disadvantaged after providing feedback. Societal preferences for relational ways of interacting and social norms about appropriate ways of expressing dissatisfaction also shape the use of feedback channels (57).
The primacy of vertical accountability in feedback management at UHCs raises many questions. The UHFPO's leadership has its advantages. The UHFPOs are the principals vis-à-vis the health workers but are the agents vis-à-vis the DG and the patients. Effective performance of both the principal and agent role by UHFPOs is likely to be di cult and untenable. Further, the UHFPOs' current role reinforces hierarchies and concentrates power. This may demotivate staff from learning from feedback and may prevent some patients from providing feedback. Weak horizontal accountability, evidenced by nonfunctional management committees and lack of engagement from local leaders, may re ect political realities, and that people have to turn to own social networks to redress grievances, is problematic. From a social equity perspective, reliance on social networks to express grievances systematically disadvantages the exercise of citizenship by those with the least social and relational capital. Weak horizontal accountability represents a missed opportunity in UHCs, given that literature attests to potential quality, equity and responsiveness gains through local accountability processes (51)(52)(53)(54). While in the immediate future, improvements to patient feedback systems could leverage the currentlydominant vertical accountability, it would be critical to recognise its limitations in providing equal opportunities and spaces for those most disadvantaged (12,13). This literature notwithstanding, our ndings caution against a universalist normative understanding that healthcare can be held to account through local political structures in all contexts. Our ndings suggest where health is not a political priority and where local leaders are not answerable to people, horizontal arrangements like the management committees at UHCs, are unlikely to be equitably effective.
We explore the application of PA theory in healthcare settings. Our ndings suggest that the two fundamental tenets of PA theory (information asymmetry and divergent goals) are less clear-cut within patient feedback systems. Three groups of actors (managers, providers and patients) can be both principals and agents. Such blurred identity and relational boundaries highlight the multiple, dynamic and often con icting, roles and responsibilities within principal-agent relationships (17) at the frontline of healthcare provision. We extend the understanding of PA theory in two inter-related ways. Patient feedback systems can loosen information asymmetry between the agents and principals, for example through patients communicating information about the health workers' conduct to the managers. Consequently, feedback systems can therefore contribute to alleviating agency losses, for example through health workers empowering patients by sharing actions taken in response to their feedback.
Combining the citizenship and PA theories has allowed us to gain insight into the logics underpinning the three steps of patient feedback management processes (23). People's use of available feedback channels entails people expressing their citizenship and agency, within the context of interpretations of one's identities and power relations and information asymmetries between patients and healthcare providers (10,12,13,16). Adequate processing and analysis of patient feedback is contingent on health workers' willingness to engage with feedback within the context of in-situ organisational dynamics, target-setting, and staff performance management (6,43,49,50). Actions on patient feedback, including reporting back to patients, entail bridging of information asymmetries across various principal-agent relationships, and enabling the expression of citizenship and exercise of agency of patients, health workers and facility managers alike (16,34,56).
From the aforementioned, we propose three implications for future policy and practice. First, health systems should ensure and maintain people's awareness of their rights to provide feedback, and of available and easily-accessible feedback channels, within a non-threatening environment in which patients can express their views without fears of subsequent retribution. Second, clear policies and operating guidelines with adequate support and dedicated resources, will enable health facility workers to value, document and analyse information from patient feedback. Last but not least, communicating timely actions taken in response to the feedback will help maintain people's satisfaction with, and trust in, their health systems and will help maintain the rapport between the people, health workers and managers.

Study limitations
We recognise two study limitations. First, our inquiry was framed in two substantive theories, and while we feel using citizenship and PA theories allowed us to understand the logics of patient feedback systems and advance the understanding of these theories, future studies can examine patient feedback systems using other theories such as on relational trust or motivation. Second, we examined patient feedback systems at limited grassroots-level health facilities and in one country only, and future research can test our re ned theory in hospital settings and different countries.

Conclusions
Appropriate policy frameworks and clear implementation processes and explicit consideration of historical, social and institutional relational arrangements, are key to the design and effective implementation of complex programmes such as patient feedback systems. Further, in contexts where there is a preference for relational ways of interaction, people will exercise their citizenship and agency to provide feedback only if they can trust the health system.

Declarations
Ethics approval and consent to participate

Consent for publication
All study participants consented for their anonymised data to be published

Availability of data and materials
The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request

Competing interests
The authors declare that they have no competing interests

Authors' contributions
TM and RH conceived the study; ZA and RH collected the data with guidance from TM and BE; ZA analysed the data with guidance from TM, RH, BE; TM and SK wrote the manuscript with inputs from RH, ZA and BE; all authors read and approved the nal submitted version  RESPONDreportingguidelinesform.docx