Conceptual model for dietary behaviour change at household level: a ‘best-fit’ qualitative study using primary data

We chose a qualitative approach to achieve the objectives and the study has been described here based on RATS guidelines for reporting qualitative studies [12]. The study was a part of the exploratory phase for a community-based dietary behaviour change intervention for prevention of chronic non-communicable diseases in rural Kerala [11]. It was conducted in Thiruvananthapuram district with a population of about 3.3 million. Kerala follows a de-centralised system of administration at the state, district and block panchayat-levels. The latter are further de-centralised to rural (grama panchayat) and urban (municipality) administrative units [13]. Chirayinkeezhu taluk, with four block panchayats, in turn consisting of 22 grama panchayats and 2 municipality areas, is the setting for the study. It is one of four revenue divisions of Thiruvananthapuram district and has a population of 550 thousand (about 130 thousand households). The study was conducted in the rural areas in two of the grama panchayats. One coastal and one non-coastal area were selected, as fishing was the predominant occupation along the coast and seafood formed a major part of their diet in contrast to the non-coastal area. Kerala has a well-developed and functioning women’s self help group network called the Kudumbasree, which is organised in the form of neighbourhood groups or ayalkootams. Different socio-economic areas were identified through the Kudumbasree registers based on their household income, and the individual households were sampled purposively from these localities. With the help of community volunteers, the participants were invited for the focus group discussions and individual interviews.

We invited men and women of different religions and socio-economic strata (SES), between 24 and 75 years of age to participate in three focus group discussions (FGDs) and 17 individual interviews. The participants of FGD 1 and 3 were from low and middle SES respectively; while FGD 2 included participants from both low and middle SES. The individual interview participants however, belonged to different SES (3 individuals from low SES; 8 from middle; 6 from high). As the study focused on dietary decision-makers, the participants were mostly female head of the households and any other member identified by her as being involved in dietary decision-making in their household. The voices of the men in the community were also found to be important, and the third FGD was primarily organized among men.

The research team included the principal investigator (MD), an assistant, one Swedish (RW) and three Indian (MR, TKSR and KRT) public health scientists. The first author (MD), who was also the interviewer and moderator is a medical practitioner with public health training. MR is a qualitative researcher with significant experience in framework analysis. TKSR and RW are also well experienced in qualitative research and looked at the data with insider and outsider perspectives respectively. KRT being a medical and public health practitioner in the state also provided an insider perspective.

The community volunteers who arranged the FGDs and interviews, resided in these areas and interacted with the people on a regular basis. They approached the residents and requested time for the respective events. None of those who were approached with a request to participate declined the invitation. All individual interviews were conducted in the residences of the participants, except one, which was conducted in a neighbour’s house. FGDs 1 and 2 were conducted in a local school, while FGD 3 was conducted in the residence of one of the participants in the coastal village. Both the interviews and FGDs were conducted based on an interview guide to understand: 1) feasibility and acceptability of a proposed dietary behaviour change intervention; 2) beliefs about foods, particularly fruits and vegetables; 3) decision-making in households with reference to food choices and access; and 4) to gain insights into the kind of intervention strategies that may be practical at community and household level (Additional file 1). The individual interviews took between 45 minutes to 1.5 hours, while the FGDs went on longer, between 1.5 to 2 hours. Data collection was stopped once we reached saturation.

All of the collected data was used for the analysis. Parts of the same data (section I-IV, Additional file 1) have been analyzed using qualitative content analysis in another study by the authors to understand the process of food decision-making in rural households [Daivadanam et al: personal communication 2014]. The objective of the present study was to develop a conceptual model that was grounded in the reality of everyday living, in order to facilitate the development of dietary intervention strategies for a household-level behaviour change process. To this end, the data was analysed with a view to identify how this change process can be facilitated by understanding current practices and their reasons, how and why changes are triggered in households and how these changes were dealt with.

The data were analysed using a modified framework analysis [14] similar to the innovative ‘best-fit’ framework synthesis methodology used for systematic reviews of qualitative literature [15, 16]. Here, we worked on primary qualitative data using a combination of inductive and deductive reasoning. Unlike the framework analysis, which is primarily a deductive approach [14, 17], we did not try to ‘fit’ all the data into pre-identified or a priori themes.

The four steps involved in the data analysis process are outlined below:

Step 1: Identification of relevant behaviour change theories and ‘a priori’ themes: Four behaviour-change theories were identified from literature as relevant to the context [1, 2]: one change theory - the Trans-theoretical Model (TTM) [4]; and three explanatory theories - Health Belief Model (HBM) [18], Social Cognitive Model [5], and Theory of Planned Behaviour (TPB) [19]. All core and peripheral constructs of the four theories were listed and defined using, as far as possible, definitions of the original authors (Table 1).

Step 2: Deductive analysis process: finalizing ‘a priori’ themes: The deductive part of the analysis was conducted in two stages. In the first stage, the data was fragmented into meaning units and the identified a priori themes were initially applied to five transcripts (2 FGDs and 3 individual interviews). Those a priori themes that did not fit the data at this stage were excluded and the remaining were finalised as such or with modifications. In the second stage, the finalized a priori themes were then re-applied across all the transcripts and used to index the meaning units identified from the data.

Step 3: Inductive analysis process: identifying and finalizing ‘new’ themes: Data not ‘fitting’ into these a priori themes were not ‘squeezed’ into them; rather we identified newly emerging themes that were relevant to the setting [16], but were not part of the core or peripheral constructs identified from the original behaviour change theories.

This study was conducted according to the guidelines laid down by the Indian Council of Medical Research. The Institutional Ethics Committee of Sree Chitra Tirunal Institute for Medical Sciences and Technology, Thiruvananthapuram, Kerala, India approved all the procedures involving the study participants as well as the interview and FGD guidelines. Participants were recruited only after recording verbal informed consent.

Twenty-two constructs were identified as a priori themes from four HBTs (Table 1). Three of these HBTs (TTM, HBM and TPB) were assessed as relevant for the conceptual model after step 2. All core constructs of SCM, excluding goal setting, were common with the other three theories. Goal setting was also excluded in step 2, as it was not found to fit, when the a priori themes were applied across the five initial transcripts. Hence, the SCM was excluded after step 2 of the analysis.

The status of the twenty-two a priori themes after the first stage of deductive analysis process is described in Table 2. Seven of the themes were not found to be relevant and excluded and six were retained as such, while nine were retained with modifications. Most of the modifications were minor involving change of name to better fit the context under study. ‘Perceived pros and cons’, ‘perceived benefits’ and ‘perceived barriers’, which had subtle differences were combined to form one theme called ‘decisional balance’.

The stage construct was also modified to ‘identifying stages-of-change’ to look for specific beliefs and concerns that would differentiate households on the basis of the stages-of-change. Since this study was conducted as a preparation for a community-based dietary behavioural intervention, we decided to modify the original stages-of-change [4] from five to three for practical purposes. Thus, the modified stages were as follows: 1) pre-contemplation, which was the same as the original pre-contemplation; 2) intention, which combined contemplation and preparation stages; and 3) action, which combined action and maintenance stages [4]. We were also concurrently developing a simple algorithm-based questionnaire as a tool to identify the household-level stage-of-change that could be administered by community volunteers. So, during this study, our focus regarding staging was to identify beliefs, concerns or thoughts that could be used to identify the different stages.

Four new themes: accessibility, perceived needs and preferences, societal norms, and perceived household response were identified using inductive reasoning from data that did not fit into any of the a priori themes (Table 2). Accessibility, in particular its affordability aspect was found to be important in the lower SES. Similarly, the importance given to perceived needs and preferences in households were found to differ based on the position or expected role of the individual in the household hierarchy.

The final themes that were mapped to construct the conceptual model are defined in Additional file 2.

The final step of the modified framework analysis was the mapping of the final themes to identify the main factors, component factors and sub-factors of the conceptual model based on the research objectives (Table 3). During the analysis, the sub-factors were further divided into categories, but these are not shown in Table 3. In this section, we have mainly described each of the main and component factors; and outlined any associations that were identified. All quotations are in italics. Any text within the quotes that are enclosed by square brackets have been inserted by the authors.

The findings of the ‘best-fit’ framework analysis were assimilated and merged into a schematic representation of the conceptual model (Figure 1). Thus, the model is composed of five parts: the central three-stage spiral representing the three stages-of-change; the ‘impact factors’ along the stem of the spiral; the ‘change processes’ on the right that enable progress from one stage to the other; the background factors on the left that act at all stages; and the overarching factors represented by a transverse plane between the first two stages.

Households could therefore enter the spiral at any stage and move forward aided by the behaviour-change strategies through the change processes; or relapse if the barriers prove stronger than the facilitating factors. Risk perception, attitude and decisional balance were the three key impact factors identified through the analysis, of which the latter was the most consistent. While they can influence outcome at any level, it is more important to address these through various strategies during the first two stages when resistance seems to be maximum. Similarly, specific change processes also work best at specific stages except the stage-identifying factors, which are relevant to all stages. Hence, the change processes are represented in a hierarchical manner based primarily on how the TTM explains the interaction of various change processes at various stages [4, 20]. This is supported by our analysis, which found the cognitive processes to be more relevant towards the lower stages where the resistance to change is often a psychological phenomenon. On the other hand, the action-oriented processes were more relevant towards the later stages, when the household’s response to the proposed behaviour change was more clear or ‘committed’.

In addition to the impact factors and change processes, we have identified background factors or modifiers, namely the societal and gender norms that act at all stages. The three overarching elements: affordability; availability; and perceived needs and preferences, were further identified during this analysis. These elements may act as barriers, preventing households from progressing across the stages, and as resistance is likely to be at a maximum before entry into the intention stage, they have been placed as a transverse plane between the pre-contemplation and intention stages. Basic or absolute affordability could prevent households from even entering the spiral, unless this issue is specifically addressed.

Many studies have modified, combined or built on existing theories, as a single theory may not address the issues related to the specific study setting [2]. For example, the COM-B system proposed by Michie et al [23], explains the interaction of capability (C), opportunity (O) and motivation (M) to generate behaviour (B) [23]. While this is highly relevant for behaviour at the individual level, it seems to be insufficient to address the same at the household level, as is the case in the present study. Here, opportunity would rest on the decisions made by someone else and an individual’s motivation and capability, however strong, may not stand against the overall collective nature of the decisions made in the household. Although the Behaviour Change Wheel model [23] provides a broad framework to develop behavioural interventions, its contextualization possibilities need further exploration. For example, factors specifically applicable to settings such as ours, where the decision-making process incorporates a collective component, may need to be added. Modifying or combining theories or models alone may therefore not address all aspects related to a specific setting, simply because these theories or models were developed in a different context with different baseline assumptions [3]. We addressed this in our study by integrating the understanding derived from theory with the findings from our primary data.

Previous work had identified that dietary decision-making processes in rural Kerala essentially took place at the household rather than an individual level [3, Daivadanam et al: personal communication 2014]. As a result some of the constructs were not found to be applicable to this process in this context and these were excluded. Four of the excluded themes: emotional reaction, identifying temptations, reinforcements or rewards and commitment are highly relevant in the individual context, but not so in the household context. They seem to disappear or get lost during the collective decision-making process. It is possible, that further studies in this area could lead to identifying proxy themes for each of these at the household level, like ‘commitment’ for example, which could not be identified at the household level. However, the category ‘perceived household cooperation’ under the sub-factor ‘perceived household response’ was identified as a potential proxy for ‘commitment’ at the household level.

Based on the contextual realities of the study setting, this model is currently focused on the household level. However, it could potentially be modified and applied to other settings and situations where a collective decision-making component is pre-dominant, such as in many other low- and middle-income settings, or in studies involving children, immigrants or other vulnerable groups in high-income countries. The discarded themes may then become important, when the model is expanded to include individual and community level factors relevant for dietary behaviour change.

This study has two major strengths: 1) It is an integration of existing theories with findings from primary data analysis, which has enabled us to overcome the gaps in existing theories pertaining to their focus on the individual. This has further ensured that the conceptual model is theory-based but grounded in the reality of the study setting, making it a strong base for developing intervention strategies. 2) The combined use of inductive and deductive reasoning allowed new themes to emerge, which are critical to a dietary decision-making process at the household level. This would not have been possible if we had used framework analysis in its original form [17]. Moreover, the FGDs and interviews complemented each other, with the former contributing to our understanding of general practices and common beliefs, and the latter providing deeper insights into the food decision-making process and specific procurement or consumption practices within households.

While the combination of deductive and inductive reasoning has been used before with both primary data as well as in systematic reviews of qualitative papers [14, 16], it is still relatively new. This could be considered a limitation, as it is not an approach that has been ‘tried and tested’ repeatedly to be accepted without question. Moreover, we opted to conduct a qualitative study to develop this conceptual model rather than a mixed methods study that would have included a quantitative component testing this model. This could not be carried out at this stage due to time and financial constraints but is planned as a next step to further improve and validate the model. At that stage, we may also need to revisit the themes that were discarded during this modified framework analysis process.