Participatory learning and action to address type 2 diabetes in rural Bangladesh: a qualitative process evaluation

The intervention was implemented in four upazillas in Faridpur in central Bangladesh from July 2016 to December 2017. An upazilla is a geo-political sub-unit of a district. We collected process evaluation data in all intervention upazillas in January 2018, before the trial analysis. Faridpur is around 2000 km² and has a population of over 1.7 million, with a predominantly agricultural economy of jute and rice farming. The population is mainly Bengali, as is the case in most of Bangladesh. Almost 90% of the population in Faridpur are Muslim, with Hinduism being the second largest religion [19]. Bangladesh is a patriarchal society, with patrilocal marital practices where women come under the guardianship of their husband’s family after marriage [20]. Moreover, in Bangladesh, purdah (female seclusion), restricts women’s freedom of movement and their access to public spaces. Women are expected to behave demurely, avoid attracting attention from men, and uphold family honour [21]. Norms about the division of labour within and outside the household, as well as norms about movement outside the household are highly gendered [22].

A baseline survey found that over 70% of men and women reported at least three risk factors for type 2 diabetes. Women were more likely to be overweight or obese and spend less time doing physical activity at a younger age than men. Less than 30% of women aged 70 years or more had adequate levels of physical activity. Men in all wealth groups had higher levels of tobacco use than women, with 63% of men currently using tobacco compared to 41% of women. Women were more likely to chew tobacco or use snuff than men. Consumption of white rice, fried food and sweets was common among both men and women. Only one in three respondents could report any causes of type 2 diabetes, and only 14% of respondents had had a blood glucose test, despite high prevalence of intermediate hyperglycaemia and type 2 diabetes (men: 17.2 and 8.9%; women: 23.4 and 11.5%) [23, 24].

In the study area, diagnosis and management of diabetes was only possible at two diabetic hospitals run by the Diabetic Association of Bangladesh in urban areas of Madhukahli and Faridpur, and three upazilla health complexes (government hospitals). Community Clinics and Family Welfare Centres at village level should be able to screen for diabetes and refer but, in practice, blood glucose testing facilities were not routinely available. Informal health workers and drug vendors were often used to monitor diabetes.

We conducted process evaluation (PE) research in 15 intervention villages. We requested supervisors of PLA groups to identify groups which had been most active in implementing communication action to prevent and control diabetes. We sampled in this way because we were limited in time and resources and sought to explore how the intervention could be effective in an optimal context. We asked PLA group facilitators to locate four different types of participant across the 15 villages who were then invited to discussions in these villages. We purposively sampled 1) men and women who were over 30 years old, had regularly attended PLA group meetings, and were interested to participate, 2) men and women with diabetes who were over 30 years old, regularly attended PLA group meetings, and were interested to participate and 3) PLA group facilitators. We conveniently sampled 4) men and women who did not attend PLA groups. The PE manager (KAk) facilitated discussions to explore how participants and communities had experienced the intervention.

KAk conducted four group interviews (GI) with male attenders, four GIs with female attenders, one GI with male diabetic attenders, and one GI with female diabetic attenders. She conducted one GI with a male non-attender, and one GI with a female non-attender (Table 1). GIs had a similar format to a semi-structured interview, and conversation occurred primarily between the researcher and the participant, as opposed to between participants [25, 26]. GIs had 3–5 participants and we used participatory photography, or ‘photovoice’ [27], and discussion to explore personal and group experience of the intervention. Photovoice typically entails giving participants cameras and inviting them to take photos which represent issues of importance to them. The photographs and the narrative about the issues they identify can facilitate communication about an issue within and between groups of people. Typically, photovoice has been used in participatory action research with marginalised groups to identify and communicate about their experience of social problems in order to stimulate action to address problems [28, 29]. To a lesser extent, it has also been used in evaluation research [30‐32]. Photovoice enables greater participant control over the research process [28, 29, 33], and enables critical reflection prior to discussion with researchers [34]. This can increase participant confidence and facilitate better communication between researchers and participants [35]. KAk took four mobile phones with cameras to each group and asked them to take photos that illustrated how the intervention had affected them, their family and community. If they could not take photos of the subject they wanted to, they should take photos of a representation of that subject. Participants discussed if they wanted to take photos individually, or together, and how to share cameras. After they had taken photos, they informed KAk who arranged a convenient time and place to conduct the GIs. Photos were printed and given to participants at discussions.

KAk also conducted two focus group discussions (FGDs) with 6–8 male attenders, and four FGDs with 6–8 female attenders to explore how the intervention had affected communities and explore triangulation with the data collected through other methods. KAk took observation notes during and after field visits. She used topic guides in GIs and FGDs to explore the barriers and enablers of behaviour change and how the intervention was able or unable to address these. The topic guide and photovoice process were piloted in one group discussion with non-diabetic men and, after minor adjustments, we felt that further piloting was not necessary. Discussions ranged from 35 min to 2 h and 5 min. KAk took informed consent from all participants and informed consent to show the photos in public forums. Participants also took informed consent from those who were the subject of the photograph.

FGD data were digitally recorded and KAk made notes about the findings in English. She translated field observation notes to English. Other data were digitally recorded and directly transcribed and translated into English. A sample of translated transcripts were checked against the recordings by KAk. Photos were inserted in transcripts and analysed with verbatim data. Transcripts were imported into Nvivo for analysis. We used inductive content analysis to generate theory from the data [36]. Data were analysed by KAk who is Bangladeshi and bilingual, HMJ who is fluent in Bangla and has lived in Bangladesh for several years, and JM who has lived in and conducted research in South Asia for over 10 years. JM and HMJ familiarised themselves with the data and listed the key themes independently before discussing these together with KAk and conceptualising a draft analytical framework. This enabled us to be reflexive and check our possible bias. This framework was applied to four transcripts by JM and adapted before all data were coded in Nvivo. JM compared data from the different types of respondents, looking for patterns, and developed a description of the main findings which she discussed with KAk, HMJ and other members of the trial team. Finally, we compared our findings with mixed methods longitudinal PE data (reported elsewhere) to explore triangulation and enhance the validity of our study [11].

The ‘active education’ component of PLA was important in increasing health knowledge and confidence, which stimulated other change mechanisms. It is important to differentiate health literacy from health knowledge, as this inadequately conveys the cognitive skills and embedded understanding which develops in becoming health literate [39]. Health literacy is defined as the personal, cognitive and social skills which determine a person’s ability to gain access to, understand, and use information to promote and maintain good health [40]. Low levels of health literacy are associated with worse health outcomes for persons with diabetes [41‐43], and therefore it is an important intervention target area. Group members felt enabled to change their behaviour because they had learned and become confident. Some attenders had low levels of functional literacy, and yet our data show that they demonstrated communicative, interactive and critical literacy in their ability to extract information and use it to exert control over life events and situations [40]. Participants attributed this change to the intervention and to the participatory methods which engaged group members in critical thinking through problem-posing and dialogue. Our intervention was also in congruence with published strategies to improve health literacy, such as limiting the amount of information to key messages, avoiding jargon, and using pictures to explain concepts [44]. Future research could determine the suitability of health literacy measurement tools (such as [45]), before adapting and applying them to evaluate how health literacy mediates changes in health outcomes.

Social relationships impact on physical and mental health [46]. Research on family and social support and network interventions has shown positive effects on self-management of diabetes in high-income countries [47‐49] and risk factors [50]. Our study differs from these interventions, as it engaged the general population and sought to address several diabetes risk factors. The social networks which were established or activated through the PLA intervention were organic and enabled the provision of emotional support (warmth and commitment), appraisal support (helping a person understand an event and the resources that could help deal with the situation) informational support (advice and information), and tangible support (material or practical help) [51]. Participation in a community-based organisation has been associated with better health in persons with diabetes [52]. Our study shows that group participation and group action can increase social support networks beyond the group into the family and social context. It will be important to collect and analyse data on social networks and social support in future PLA interventions, particularly because they may play an important role in the sustainability of the intervention and its effects [53]. Future research could explore in more depth how social support interacts with other pathways of change [54, 55], and how the type of network and the way that it works affects pathways of behaviour change [51, 56].

We found that there was some engagement with structural constraints to healthy behaviour. For example, groups pressured men and communities to support women who walked to prevent or control diabetes and pledged not to tease or criticise physical activity. Other actions acknowledged structural constraints and took a pragmatic approach to addressing them. For example, gender norms limited movement of women preventing their access to vegetables, and groups addressed this by having renewed energy and commitment for kitchen gardening. After a few failed attempts to address the lack of blood glucose testing facilities in government health facilities, groups organised and paid for community level private testing. Our intervention was implemented over a relatively short period of time, and group members may have sought to exhaust local solutions before engaging with state systems. Alternatively, they might have felt unwilling or unable to demand systems change, perhaps lacking in confidence that they could achieve change [57]. Community participatory approaches have been criticised for devolving responsibility to address health problems from the state to local communities [58]. While we do not fully agree with this criticism, our findings suggest the need for public health policy responses to non-communicable diseases to complement community based efforts [59, 60]. It will be important to evaluate the effectiveness of group-initiated strategies to maintain behaviour change over time, and the possible amplification effect of complementary systems interventions.

We used group facilitators to identify participants for the study, because the group intervention had ended, which may have biased sampling towards recruiting active attenders. Attenders may have been motivated to respond positively because they sought to secure continued support for the intervention, but we also found that data from non-attenders was overwhelmingly positive. The PE manager (KAk) spent substantial time in Faridpur observing the intervention during the trial and collecting longitudinal PE data. We compared our findings with longitudinal data and found there was a high degree of triangulation, indicating the validity of our findings. Data collection and analysis were undertaken before the trial analysis, to generate hypotheses [61]. While writing this article, the trial results were made public, which may have biased our interpretation of the data.