Initiating a participatory action research process in the Agincourt health and socio- demographic surveillance site

Background Despite progressive health policy, disease burdens in South Africa remain patterned by deeply entrenched social inequalities. Accounting for the relationships between context, health and risk can provide important information for equitable service delivery. The aims of the research were to initiate a participatory research process with communities in a low income setting and produce evidence of practical relevance. Methods We initiated a participatory action research (PAR) process in the Agincourt health and socio-demographic surveillance site (HDSS) in rural north-east South Africa. Three village-based discussion groups were convened and consulted about conditions to examine, one of which was under-5 mortality. A series of discussions followed in which routine HDSS data were presented and participants’ subjective perspectives were elicited and systematized into collective forms of knowledge using ranking, diagramming and participatory photography. The process concluded with a priority setting exercise. Visual and narrative data were thematically analyzed to complement the participants’ analysis. Results A range of social and structural root causes of under-5 mortality were identified: poverty, unemployment, inadequate housing, unsafe environments and shortages of clean water. Despite these constraints, single mothers were often viewed as negligent. A series of mid- level contributory factors in clinics were also identified: overcrowding, poor staffing, delays in treatment and shortages of medications. In a similar sense, pronounced blame and negativity were directed toward clinic nurses in spite of the systems constraints identified. Actions to address these issues were prioritized as: expanding clinics, improving accountability and responsiveness of health workers, improving employment, providing clean water, and expanding community engagement for health promotion. Conclusions We initiated a PAR process to gain local knowledge and prioritise actions. The process was acceptable to those involved, and there was willingness and commitment to continue. The study provided a basis from which to gain support to develop fuller forms of participatory research in this setting. The next steps are to build deeper involvement of participants in the process, expand to include the perspectives of those most marginalized and engage in the health system at different levels to move toward an ongoing process of action and learning from action.

The study was conducted at the MRC/Wits Rural Public Health and Health Transitions Research Unit, which oversees the Agincourt HDSS located in rural Mpumalanga, a province of 4 million people in rural northeast South Africa. Established in 1992, the HDSS covers a population of approximately 115 000 people, over 450km2, 31 villages, and 20 000 households [26,27] (Figure 2). A dedicated Public Engagement Office works to enhance community and health systems engagement at different levels. The office regularly provides data and discusses research findings with the community and health system at different levels. Map of Agincourt HDSS in rural northeast South Africa. Serving the Agincourt study area within a radius of 20–60 km, is a network of ten government run PHC clinics that provide free basic outpatient health services during regular working hours. Services include routine maternal and child health interventions (including integrated management of childhood illnesses, well child visits, growth monitoring, routine immunizations), sexual and reproductive health services, testing and treatment for sexually transmitted infections, including HIV, minor trauma and routine care for chronic illnesses [28]. In 2015, attendance at antenatal care (ANC) clinics before 20 weeks of pregnancy in Mpumalanga was 56%, 80% of children one year and below had complete immunization, and the facility–based under–5 mortality rate was 8.3% against a target of 5% [29]. There are also limited private health care services in the area. We developed a process based on PAR. PAR is a non–linear, context specific process, with cycles of observing, reflecting, acting and learning from action. The repeated cycles build a sustained process that enables community ownership [30]. Within the time and resources available, it was possible to initiate the process and so the following description offered by Loewenson et al was adopted: “start by obtaining an insight into the communities and their conditions. This provides the information to support inclusion in the work, to systematize experience and to draw out priorities for attention” [30] (Figure 3). PAR process, with the initial elements highlighted [30]. We defined communities geographically, as residents of a specific area with shared social and health conditions. To prioritize and maintain prior linkages, we attempted to re–engage participants involved in a previous community–based participatory research (CBPR) pilot study in the Agincourt HDSS [31]. In the previous study, three village–based discussion groups had been convened. Villages had been selected on the basis of demographic variation and feasibility (Table 1) and in each village, discussion groups comprised women of reproductive age, family members, traditional healers, religious leaders, community health volunteers, health workers and community leaders. To mitigate any potential biases due to power differentials, in one village, the group consisted of women only (Table 2). Characteristics of selected villages Source: [32]. Composition of village based discussion groups *All participants recruited were 18 years of age or older. Although participants are likely to be categorized by more than one role in the community, one role per individual was considered for the purposes of convening the focus groups. We agreed roles with participants to identify what they feel to be their primary role in the community. †Close relative: parents, grandparents, siblings, children, in–laws, nieces, nephews and cousins. ‡We acknowledged that people with working arrangements, particularly village health workers and village officials may not be available for a series of six weekly meetings. We also acknowledged the ethical imperative of engaging participants who would otherwise be involved in earning income and or the provision of public services. The groups were therefore based on these compositions, with careful consideration of minimizing disruption to local services. Public Engagement Office staff approached individuals involved in the earlier CBPR study in villages, and described the current study, activities and intended outputs. Written consent forms and information sheets were provided, and participants were invited to ask questions at the time, or afterwards by telephone. For those willing to be involved, a convenient time was arranged for the first meeting at which participants were asked to sign and return the consent forms. Through this process, all participants from the prior study agreed to be involved. In the first meeting, and to encourage participant control over how the topics for discussion were framed, we asked people’s opinions about conditions to examine. We also consulted the Directorate for Maternal Child, Women and Youth Health and Nutrition (MCWYH&N) in the provincial Department of Health (including co–authors BS and MVDM) and considered conditions with high prevalence rates identified in Agincourt HDSS. Through this approach, under–5 mortality and HIV–related mortality were selected. The discussion groups then embarked on a series of six weekly meetings to consider the conditions in terms of causes, contributory factors, and actions to address the identified issues (Table 3). This paper reports on the process as it related to under–5 mortality, the results on HIV–related mortality are reported elsewhere [33]. Schedule of village–based meetings In the second meetings, VA data on under–5 deaths were presented. 110 such deaths had been recorded by Agincourt HDSS in 2012 and 2013. The leading causes of death were acute respiratory infection (including pneumonia), HIV/AIDs–related death and malaria accounting for 18%, 15% and 13% of the total burden respectively. Overall, 61% of deaths were due to infectious causes. Furthermore, 49% of deaths occurred among children 1–4 years of age, 30% to infants and 21% to neonates (Table 4). The VA data also contained indicators on the circumstances of mortality, developed in the same project [35]. These data indicated multiple problems with access to care at and around the time of death. Specific issues identified were: families not calling for help (34% of all problems reported), not going to a facility at the time of death (29% of all problems reported), and that the overall costs of care were unaffordable (14% of all problems reported) (Table 5). Cause–specific mortality fraction (CSMF): all under–5 deaths, age/sex sub–groups *Deaths due to congenital malformations include conditions that have their origin in the perinatal period even though death or morbidity occurs later [34]. Frequencies of responses to new Verbal Autopsy indicators on circumstances of mortality *The denominator is the number of respondents who reported traveling to hospital/health facility. Respondents were able to indicate more than one ‘circumstance of mortality’ indicator for each death reported. After presenting the VA data, we invited participants to share their knowledge and experiences in an open discussion. Participants were prompted to share views on symptoms, modern and traditional therapies, health service responses, and what happens in the village in acute situations. Issues that arose were recorded on a flip chart that was visible to all participants. When a sufficient amount of discussion had occurred in the time that was available, and no new issues were identified, the facilitator (co–author SN) summarized the discussion and checked the list with participants for completeness. We then undertook a process to systematize individual views and experiences into shared accounts using ranking and diagramming. For the ranking, the flip chart with the initial long list was put on a table in the center of the group, and participants were invited to interrogate it. Participants were given adhesive markers to nominate issues they considered to have the highest priority (Figure 4). Two rounds of ranking were conducted to ensure the issues were re–visited and re–checked and to validate the list before recording the flip chart in a photograph and closing the meeting. In the subsequent meetings (meeting three) we used diagramming to revisit the ordered list. We adopted a ‘problem tree’ diagram to organize issues identified into proximate determinants, mid–level systems factors, and social and structural causes of under–5 mortality [30] (Figure 5). Systematising subjective perspectives – ranking. Validating by consensus – diagramming. One discussion group used a visual participatory technique called Photovoice to explore the use of contemporary methods employing mobile and digital technologies. We selected the remotest, all–female discussion group (Group C) for the Photovoice method. Participants in Group C were provided with digital cameras to take photographs of their physical environments as a further input to the discussions [36,37]. We provided basic training on photography, explained why and how to secure release permissions from subjects of photographs, and provided consent forms for permission releases. In the subsequent weekly meetings, we projected the photographs taken by group members during the discussions. Photographers were invited to describe and explain their images and the group considered the issues they represented as additional inputs to the discussion (Figure 6). Participatory photography. In the final meetings on under–5 mortality, a summary of the process was fed back to each group to verify content and meaning, upon which discussions were held about actions to address the issues identified. This involved re–visiting and re–checking the outputs of the prior process, and moving from causes and contributory factors toward remedial actions. We used ranking to identify priorities for action through which only issues that were nominated by the group were registered. We also indicated that the outputs of the process would be provided to the local health authority. All discussions were facilitated by a senior qualitative researcher (co–author SN) with knowledge of the local area, assisted by a qualitative field research assistant and co–researchers (co–author LD). The focus group discussion (FGD) method was employed [38] using topic guides to structure the discussions in meetings lasting 90–120 minutes. With separate permissions, the discussions were audio recorded and transcribed verbatim. Transcripts were translated from the local language, xi–Tsonga, into English. SN oversaw and performed transcription and translation with the field assistant. Observational notes were also taken (SN, LD and field assistant) and analyzed. The visual and narrative data were thematically analyzed to complement the collective analyses. Thematic analysis was conducted in parallel to, and following completion of, data collection. NVivo Version 10 was used for data entry and management [39]. Transcripts were analyzed based on combined inductive/deductive framework analysis (co–authors OW and LD). This involved a sequence of steps of increasing abstraction from data to findings [40,41]. The transcripts were read several times to familiarize researchers with the main ideas, paying attention to recurring patterns and themes. Initial themes and sub–themes were noted as codes. Transcripts were re–read, re–checking for themes, how themes supported the data and vice versa, identifying relationships within and between themes. This was done iteratively until thematic saturation. The visual data were cataloged and a ‘word cloud’ was generated though which frequencies of terms were graphically represented. Informed consent was sought from all participants. Participants were provided with information in the local language and contact details for the research team, and given time to consider this before agreeing to be involved. Separate consents were gained for audio recordings. All participants were assured anonymity, and that taking part would have no influence on care available to themselves or their families. Participants were also assured that they were free to leave the process at any time and for any reason. Participants were reimbursed with travel expenses, provided with refreshments in meetings, and given a voucher of ZAR300 (approx. US$ 23) at the end to reimburse for time spent participating and as a token of appreciation. All identifiable data were anonymized. Institutional review boards at the Universities of Aberdeen, Scotland, and Witwatersrand, South Africa, and the provincial health authority in Mpumalanga, reviewed and approved the study protocol.

N/A

• Dima Health