Healthcare research: Using NVivo in a commercial setting

18 June 2014 - BY Alexander Howson IN analysis, coding, comparative, constant, for, health, mac, methods, mixed, nvivo, research

Healthcare research: Using NVivo in a commercial setting

Way back when (actually, in 1991), I learned how to use NUD*IST 6 as part of my doctoral training at the University of Edinburgh.

This was also the version I used when I did funded research as a faculty member at the universities of Aberdeen and Abertay Dundee, in Scotland.

A decade after leaving academia to pursue a freelance research and writing career, I find myself again using QSR software, except this time, it’s NVivo 10 (both Windows and Mac Beta versions). At least no need to explain to clients what the acronym NUD*IST denotes…

Working with physicians and clinicians

As an independent writer and researcher, I develop content for continuing education and quality improvement programs in healthcare.

Part of that work entails conducting and analyzing qualitative interviews with physicians and other clinicians to identify clinical practice patterns and challenges associated with particular conditions.

For instance, in the last 12 months I’ve worked on projects in type 2 diabetes, sleep/shift work disorders, low testosterone, and multiple myeloma, and I’m about to start working on a project focused on overactive bladder.

Typically, the clients I do this work for are medical education companies and quality improvement organizations, and their clients are usually pharmaceutical companies and medical societies.

Poster presented at Alliance for Continuing Education in the Healthcare Professions Annual Conference 2014, based on results of survey data, qualitative interviews, and chart review.

Dealing with fast turnaround times and teamwork

These qualitative projects turn around quickly – typically 6-8 months from kick off to reporting – and so they differ markedly from the slower (more leisurely?) pace of data collection and analysis I was familiar with when I did academic research. Another departure from academic research is the context: I’m a work for hire, and clients need actionable information, so there’s much less opportunity for mulling over findings than in an academic context.

I need to be able to analyze the data quickly, show the client what the findings are, and report in a way that highlights the implications for creating education programs or quality improvement interventions.

Usually, projects also involve working as a team member, most of whom do not necessarily have a research background, and if they do, it’s most commonly in quantitative analysis. In a typical project, I’ll work with client project managers and strategists, interface with my client’s client, and often collaborate with other researchers in conducting interviews and generating analysis.

NVIvo is a great resource for collaboration that allows me to share entire projects with all nodes, coding, memos and charts intact.

Getting started with constant comparative analysis

A typical project for me includes anywhere from 10-30 interviews, which are usually 60 minutes in length -- quite a substantial volume of data (and remember, analysis needs to be turned around rapidly, usually in a 4-8 week time slot).

Despite the rapidity of turn around, I usually use constant comparative method for most of the projects I work on.

Once I’ve uploaded all the transcripts and created a classification grid, I start reading through each transcript to get a feel for content and to add notes or memos if I remember something particular about an interview (i.e. the nonverbal stuff that can sometimes be important in analysis).

Moving from description to interpretation

A common next step that I increasingly find helpful is to code by question.

For instance, in a recent project on managing patients with shift work disorder, the client provided interview transcripts with 20 physicians that included only 4 questions. These questions mapped directly to areas of change in clinical competence for which the client already had quantitative data. So I created nodes for each of these 4 questions to get a feel for the descriptive results, then coded within each question node where data seemed to demand further interpretation and pointed to emergent themes (as a preface to, as Howard Becker would put it, getting out of the data and moving to conceptualization).

This 2-step process of beginning with descriptive nodes then moving to thematic nodes helps me to:

  • Quickly analyze the data
  • Report preliminary descriptive findings to the client
  • Develop a more nuanced analysis that potentially adds value for the client based on emergent insights, rather than descriptive findings

Exploring key differences

I worked with a colleague,(Dr. Wendy Turell – also a researcher/analyst with an academic background who now works for PlatformQ, and with whom I’ve written about using qualitative analysis for generating education needs assessment data), on a project exploring how physicians manage patients with multiple myeloma in both academic and community settings.

Both of us interviewed clinicians in person and by telephone. Wendy created the project in NVivo and did the initial descriptive coding, analysis, and reporting. These descriptive findings pointed to some key differences across academic and community settings (e.g. in how physicians selected therapies for their patients and collected information from patients about side effects and toxicities).

I wanted to test out the Mac version of NVivo, which supports project exchange with NVivo for Windows. So I imported the Windows version of the project to Mac, then went back into the data to re-analyze the descriptive nodes.

Uncovering assumptions

As I worked through key nodes that were of interest to the rest of the research team, (which included a commercial client and academic physicians), I began to see differences in how clinicians talked about their decision-making processes that seemed to map onto their descriptive reports of what they said they did in practice.

As I analyzed further, it seemed to me that clinicians were also drawing on background assumptions about experience and expertise as they talked about their decision-making. These background assumptions – which we have coded for the moment as tacit knowledge – may be potentially beneficial or problematic for clinical practice. The client was sufficiently surprised and excited by this insight to explore it further.

Meeting client expectations

Reporting qualitative findings for commercial clients can be challenging, especially since many clients still want to put a quantitative spin on qualitative findings.

I do use a lot of quotes to support my observations/analysis, but it’s also helpful to be able to share a framework matrix (as a report appendix for instance) to show the client how I’m interpreting text (and display, to a degree, analytic process), or to use a chart to display relationships between certain nodes and key classifications.

Lately I’ve been exporting NVivo data to Excel to have more flexibility in chart design (Figure 1).

Figure 1. Coding Frequency for Thematic Node X by Clinical Setting

Retaining rigor and trustworthiness

Moving forward, I’m excited to be using qualitative analysis software again as a routine resource in my work. It helps me retain rigor in my process and assures my clients about the trustworthiness of findings.