Qualitative Data Analysis in Health Psychology: Testing Theoretical Models Using Qualitative Data

Project Overview and Context

This case study is based on an article published in 2016 reporting the results of a qualitative investigation into the management of type 2 (non-insulin-dependent) diabetes mellitus (T2DM; Newton-John, Venturi, Mosely, & Brown, 2016). This case study demonstrates the application of qualitative methodology for the purpose of explicit hypothesis testing. The case study highlights the various issues involved in sampling and interviewing participants with chronic illness and how to interpret their data in light of existing theoretical models.

More than 220 million people worldwide are known to have T2DM, and the prevalence is growing at an alarming rate (Shaw, Sicree, & Zimmet, 2010). As the developed world ages, becomes more sedentary, and increasingly obese, T2DM rates are expected to continue to rise.

As there is no curative treatment for T2DM, patients with the condition must carry out a complex series of behaviors on a daily basis to remain healthy. These behaviors include checking blood glucose levels, taking medication, engaging in regular physical activity (but not excessively), and modifying their food choices. Not carrying out these monitoring and self-regulatory behaviors risks major health complications, with cardiovascular disease, nerve damage, and blindness being prominent among them (Gregg et al., 2014). Hence, the person diagnosed with T2DM “walks the tightrope” between daily, effortful behavioral requirements—which may be painful, tiring, and accompanied by side effects—and the constant threat of serious health complications and eventually even death if the behavioral requirements are not adhered to. Adding further to the complexity of managing the condition is the fact that in addition to a genetic predisposition, T2DM often develops as a result of a premorbidly unhealthy lifestyle. Neglect of a healthy diet and regular exercise greatly increases the chances of developing T2DM, and yet these are critical components of treatment adherence to avoid the long-term disease complications.

Because of the complexity and unrelenting nature of disease management in T2DM, we were interested in learning more about how individuals cope with their disease on a daily basis. In particular, we were interested in the social aspects of their condition. Research has previously found that close relationships can have a significant impact on other forms of chronic illness, such as chronic pain (Newton-John, 2013), and diabetes is no different. In many cases, the data show close relationships having a positive effect on diabetes management—for example, marital satisfaction is significantly positively related to prescribed exercise behavior, and couple problem-solving ability is associated with healthier eating choices by the person with T2DM (Johnson et al., 2014). However, the T2DM literature is not unequivocal when it comes to the effect of close relationships on coping. Whether the motivation is intentional or not, there are data to show that close relationships can have a detrimental effect on coping with T2DM. For examples, spouses can undermine dietary self-management by encouraging unhealthy eating or can place undue pressure on their partner by nagging or badgering about health behaviors (Rook, August, Stephens, & Franks, 2011). Thus, it is not a given that the presence of a spouse will have a beneficial effect on T2DM self-management.

There have been two theoretical models by which the social network of an individual with chronic illness is thought to influence health behaviors. Health-related social support refers to the partner and close family providing encouragement and positive feedback to an individual already perceived to be coping successfully with their chronic illness. In practice, this kind of social support behavior might include expressing admiration for maintaining exercise regimes or preparing separate diet-appropriate foods for the person with T2DM. By contrast, health-related social control behaviors are efforts by the family to correct what they perceive to be suboptimal health behaviors by the person with T2DM. Here, there might be covert attempts by family members to prevent engagement in unhealthy behaviors (e.g., hiding cigarettes or sugary foods) or outright confrontation about the frequency of blood testing or the results of the latest hemoglobin check.

To date, research in this area has examined social support and social control effects in spouses and immediate family members only. However, daily social interaction involves a much wider social circle than just close family members. In fact, many people spend more hours per day interacting with work colleagues or friends than their immediate family! Hence, we expanded the investigation of social experiences of people with T2DM to include not just partners and family but also friends, neighbors, extended family, and work colleagues. We were interested in understanding whether relationships in this wider social circle were also perceived to be affected by living with T2DM, as well as whether these relationships were thought to have an impact on T2DM management.

Our second research question related to whether the health-related social support and health-related social control models also applied to interactions with the wider social circle. These models had been tested only in primary relationships to date; however, our contention was that these kinds of health behaviors could potentially arise from anyone in the social network with whom the person with T2DM had regular contact. Ultimately, the goal of the present research was to expand on knowledge of what promotes and also what hinders the effective self-management of T2DM, as treatment adherence in this condition is of crucial psychological, social, and economic importance.

Research Practicalities

We determined that for the purposes of addressing our research questions, a qualitative methodology was the most appropriate approach to take. Qualitative research allows the researcher to gather rich information about participant experiences, with minimal influence of researcher preconceptions or biases (Creswell, 2014). Using a semi-structured interview with broad, open questions, we were able to tap into participant health experiences in an unconstrained fashion.

Our first research practicality was to recruit participants with a confirmed diagnosis of T2DM. The most direct way of achieving this would be to recruit participants via medical clinics—endocrinologists see the majority of diabetes patients, but ophthalmologists and bariatric surgeons (treating morbid obesity) would also see significant numbers of patients with the disease. However, we chose not to recruit via this method as the likelihood of recruiting patients with poor T2DM management would be elevated if we sampled those attending specialist appointments. This in turn would bias our findings—we would be gathering information on the social experiences of those with suboptimal treatment adherence, which the evidence suggests would be very different experiences from those who manage their diabetes well.

We therefore decided to recruit our participants through a diabetes society membership list. There are benefits and limitations with this approach also. The advantage of this approach is that you are more likely to access participants with a wide variety of disease severities, ages, and social backgrounds. The disadvantages are that you may be accessing a patient group who are more health literate than usual, as they will be receiving information about their disease through their group membership. Hence, this may introduce the opposite problem to medical center recruitment—this health-literate sample could be more treatment adherent than the typical patient with T2DM. However, on the basis that we also needed to recruit according to a number of different demographic factors (see below), we selected this recruitment method as the best way to access our participants. In addressing the issue of the validity of the T2DM diagnosis, we asked a series of questions regarding where and when the diagnosis was made and by whom. If there was any uncertainty about the information provided, it was followed up by the researcher during the subsequent interview.

Our next research practicality issue was in relation to the variance within the sample itself. We required access to 20 to 30 participants who would represent a cross section of the T2DM population to give our results a reasonable degree of generalizability. Therefore, we carried out purposive sampling—the deliberate recruitment of participants who met demographic and clinical criteria that we believed to be important when it comes to the variable of interest. In this case, the “outcome variable” is adherence to T2DM treatment, and we sought a range of experiences here. Thus, we were seeking to recruit participants who use a variety of T2DM treatment methods, from urban and rural settings, different ages, and a gender mix. We also needed participants to speak English, have a degree of social connectedness (although this was not explicitly mentioned as part of the study information for recruitment), and be available for an interview lasting up to 1 hr.

We placed an advert on the website of a membership group (Diabetes Victoria, the peak body and leading charity in the State) and also on local social media webpages relating to diabetes, requesting that interested individuals contact us via post or email. We sent out a total of 108 letters or emails with information about the study in response to our advertisement. This also allowed us to collect some anonymous information about the demographic variables we were interested in and on that basis send invitations to a subset of those who met our stratification criteria. The final sample of 25 participants ensured adequate representation of women (48% of the sample), a wide age range (22-79 years), a wide range of T2DM treatment methods (insulin injections, oral medications, lifestyle modification alone), and a good range of disease durations (less than 12 months since diagnosis to 29 years since diagnosis, mean of 5 years of T2DM). This spread of participants was achieved with a relatively small sample size (N = 25), which would be highly unlikely using the typical random sampling methods used in quantitative research. In return, participants received an AUD$20 gift voucher as a token of appreciation for taking part.

A final practical issue for recruitment emerged during the interview process itself.

One participant, who was a member of the diabetes association and who had responded appropriately to the screening questions for inclusion, was found at interview to demonstrate significant cognitive impairment. During the interview, it also transpired that he had not been formally diagnosed with T2DM (he thought that it was “likely” that he had the disease!). On both counts, he was excluded from participating in the study, and with his consent, contact was made with his general practitioner (GP) to advise of the situation.

As for the interviews themselves, the researchers made a number of planning decisions to ensure they were conducted as smoothly as possible. First, appointments were scheduled by telephone, but a reminder text or email was sent out approximately 5 days prior to the agreed time to minimize non-availability. Researchers made a clear statement at the beginning of the interview that if the participant needed to interrupt the interview at any stage to attend to their health needs (such as blood testing and injecting), they should feel free to do so. Finally, researchers ensured that the conversation remained on task for the duration of the interview and did not divert into non-research-related areas. Of course, some participants occasionally wanted to discuss other topics during the course of the interview, but the risk here is that the interview extends out well beyond the allotted hour and the participant becomes unnecessarily fatigued. The research team varied in their ability to contain the more garrulous participants, so we arranged for some mid-data collection communication skills training to take place. Here, the senior researcher carried out some role-plays of appropriately assertive ways of containing talkative participants. For example,

As a result, the interviews were efficiently run without excessive burden on participants.

Research Design

Participants were interviewed on an individual basis, in their homes for convenience, for approximately 1 hr at a time. They were not randomized to interview.

This research was a qualitative study using thematic analysis to explore the social experiences of individuals living with T2DM. Thematic analysis (Braun & Clarke, 2006) was chosen as the qualitative method because it allows for the explicit testing of hypotheses, even though the data are non-numerical. This is because the thematic analytic approach adopts a phenomenological position, such that the semantic content of the data (the explicit meaning of the data) is analyzed, rather than its latent content (the underlying or deduced meaning of the data). As stated earlier, one of our research questions was to test whether the social control and social support models of health behavior applied to this data set. Thematic analysis allowed us to test those models because the methodology allows for predetermined themes to be identified in the data that are being analyzed.

There are five stages in a thematic analysis according to Braun and Clarke (2006). These are as follows:

• Stage 1—Familiarization. The researcher(s) simply read the transcripts to gain a sense of what theme ideas might be generated. No coding or data elimination takes place at this stage; it is just an overview of the data to assess the content in its broadest form.
• Stage 2—Generating themes. Here, the data are systematically investigated for its thematic content. The researcher(s) cluster responses according to conceptually similar ideas, and these can be refined as the coding process continues.
• Stage 3—Defining themes. Once broad themes have been generated, they need to be given specific operational definitions. This should be a joint exercise between researchers if there is a team involved, and it ensures accurate, replicable coding is being carried out.
• Stage 4—Reviewing. The researcher(s) now return to the data and systematically work through it based on the themes that have been defined. Any discrepancies between researchers on how data should be coded should now be resolved, and no new themes should emerge after this stage.
• Stage 5—Reporting findings. Quotes and compelling extract examples are drawn from the data to illustrate the theme being represented. Information regarding how many references to each theme were made in the data, and the number of participants referring to that theme is also given alongside the theme description.

“Method” in Action

Participants who were selected to take part were interviewed face-to-face by one member of the research team. To minimize data loss, participants were interviewed by the researcher in their homes. This of course adds to the cost of the study and to some extent adds to the time required for the study, as the researcher’s travel time added to the overall duration of data collection for the study.

All the interviews were audio-recorded (with the participant’s consent) and later transcribed verbatim and uploaded into NVivo V.10 qualitative data analysis software. The interviews lasted an average of 55 min, with a range of 25to 103 min.

The researcher presented the participants with three broad, open-ended questions designed to elicit realistic and representative examples of social interactions around T2DM. The primary question was “What do your family/friends/work colleagues think or feel about your diabetes?” Following saturation of that topic, there were two follow-up questions asked of all participants: “What do they (family/friends/work colleagues) say or do that is encouraging or supportive?” and “What do they (family/friends/work colleagues) say or do that is discouraging or unsupportive?” For those participants who were no longer employed, the work colleague prompt was not used.

Participants were thanked for their time, and the gift voucher was given on completion of the interview.

Practical Lessons Learned

Conclusion

The research methods used in this study successfully addressed a number of issues. Purposive sampling gave us a much wider coverage of relevant demographic and clinical factors than we would have expected had we used the more typical random sampling method. The qualitative data analysis method also allowed the true experiences of the participants to emerge, without the preconceptions of the researchers influencing the data in any way. Thus, we were able to test a theoretical model using qualitative research methods, which demonstrates the flexibility of this approach. However, no research study is without limitations, and we found that our participant sampling was not as evenly spread across the demographic criteria as we had hoped. This may reflect one of the ongoing issues in psychological research—how truly representative of the target population are the participants in the research study.