Intensive Longitudinal Data in Behavioral Medicine: A Q&A with Dr. Genevieve Dunton

Dani Arigo, PhD, LP^✉; Behavioral Informatics and Technology (BIT) SIG Chair

Genevieve Dunton, PhD, MPH

Dr. Genevieve Dunton is an Associate Professor of Preventive Medicine and Psychology at the University of Southern California (USC). She has been instrumental in promoting the utility of intensive assessment (i.e., multiple assessments of the same person over short time frames) for advancing behavior change science and the use of sensor technology to capture behavior and its determinants in real time. She recently published influential papers on these topics in JAMA and Translational Behavioral Medicine.

 

You’ve been an advocate for using intensive longitudinal data (ILD) to study “micro-temporal processes.” What is this approach and what can it do for behavioral medicine?

Micro-temporal processes are sequences of events, exposures, or experiences underlying human behavior that unfold across acute timescales such as minutes or hours. I’ve been a big advocate for studying these processes because they can tell us so much more about when, where, and how people behave than by looking at change across traditional timescales (i.e., months and years). Understanding micro-temporal processes can be particularly important for studying health behaviors that need to be repeated in order to have health benefits (such as physical activity and healthy eating). Sustaining these behaviors can be challenging due to variations in how people feel, who they interact with, and the environments they encounter. When factors that influence that behavior vary over short time periods and across settings, maintaining consistent behavior can be difficult. Intensive longitudinal data (ILD) methods are critical for capturing micro-temporal processes because they collect many measurements (hundreds or thousands) over time, often at frequencies of seconds or minutes. ILD can help us learn important things about health behaviors such as their temporal specificity (e.g., Does the influence of a factor on behavior vary over time?), situational specificity (e.g., What specific environments influence specific behaviors?) and person specificity (e.g., What unique sets of factors are predictive of behavior for a given individual?). ILD offer an enormous opportunity to fine tune, enhance, and in some cases even scrap existing theories that guide behavioral medicine intervention(s?). Further, ILD can form the basis for the development and evaluation intensively adaptive interventions that incorporate real-time feedback into decision rules about what type of intervention is needed for whom, at what time, and in what situation.
 

What types of technology have you used to capture ILD?

In my Real-time Eating Activity and Children’s Health (REACH) lab at USC, we’ve used smartphones, accelerometers, external GPS devices, ultraviolet dosimeters, Bluetooth-enabled asthma inhalers, and wearable air pollution monitors. In one of our ongoing studies, Dr. Stephen Intille and I are combining assessments from smartphones and smartwatches to try to get the? most complete picture of micro-temporal processes underlying maintenance and relapse of physical activity, sitting time, and sleep over 12 months. We’re collecting continuous passive ILD from including acceleration, GPS, noise, light, voice and SMS messages, and app use. We are also intermittently collecting active forms of ILD through real-time self-reported Ecological Momentary Assessment (EMA) surveys of feelings, motivation, and cognitions. An innovative feature of this study is the use of micro-EMA through the smartwatch interface. Micro-EMA allows us to deploy brief questions to participants very frequently (up to 6 times per hour) with limited burden.
 

You and your colleagues also developed new statistical methods and software to analyze ILD. What makes this approach to analysis novel?

We’re very excited about our new MixWILD software! MixWILD stands for “Mixed Effects Modeling With Intensive Longitudinal Data.” Funded by grants from NHLBI and NCI, Dr. Don Hedeker and I have been able to develop this user-friendly (and free!) product for the analysis of mixed models. What’s unique about MixWILD is that it can be used to examine how subject-level means, variances, and slopes of time-varying variables may predict subject-level behavioral outcomes. ILD can be aggregated into conceptually and theoretically relevant indicators at higher-level units of analysis (e.g., week, month, year, person), including means (e.g., how unhappy is a subject, on average, across occasions?), variances (e.g., how erratic is a subject’s mood across occasions?), and slopes (e.g., is a subject’s mood related to feelings of energy across occasions?). MixWILD uses a two-stage modeling approach to test how these subject-level means, variances, and slope predict subject-level outcome. For example, MixWILD can help us to determine whether fluctuations in mood influence smoking or moderate the impact of a smoking cessation intervention. MixWILD has a user-friendly graphical interface with drop-down menus, so it is super easy for the non-expert. It also can be used to test regular multilevel linear and logistic regression models if you don’t want to worry about any of the fancy stuff described above.
 

What new or underused technologies and methods might help us make the biggest strides toward understanding micro-temporal processes?

A big challenge that we face in mHealth is low user engagement. Individuals download apps and are enthusiastic users for a while, but engagement drops off steeply after a few months. We face the same challenges when collecting ILD. We need to continue to push ourselves to innovate to optimize engagement. Doing so may mean harnessing new technologies and machine learning methods to passively assess constructs such as affective and physical feeling states that we could only previously collect through active input methods. This includes facial expression recognition algorithms, eye-tracking sensors, and predictive modeling of heart rate variability and galvanic skin response. We also need to think creatively about how to integrate features of gamification, entertainment, humor, social connection, and relaxation into our health behavior change and data collection applications to make them more engaging.
 

You also invest in the next generation of behavioral medicine professionals beyond mentoring your own trainees. What are some of the most common questions you get?

This is one of my most favorite parts of what I do. As a trainee, I benefitted immensely from attending mentoring sessions with experts during SBM annual meetings. After I came back from sabbatical last summer, I started hosting “Ask Me Anything” sessions. I post available times on Twitter and anyone who is interested can sign up for a slot using Calbird. We then do a 15-min one-on-one video call using Zoom. I can’t tell you how inspiring and fulfilling this experience is for me. We talk about a variety of topics including the job market, post-docs, and grant-writing. We also talk about problems with specific theories and technologies, challenges, and hang-ups that we face. I most commonly get questions about my own professional path, and I explain how haphazardly I got to where I am now. I also get questions about how behavioral medicine professionals can get more structured training on EMA. I wish there were more EMA courses or workshops available right now. It makes me think about trying to put together an online non-credit course that anyone could take.
 

How have your involvement in SBM and your use of social media platforms such as Twitter contributed to your career success or satisfaction (or both)?

I started my Twitter account in 2013 and ended up taking a break for a while, to focus on other priorities. When I came back, I was shocked at how much it had taken off as a platform to communicate with other professionals in behavioral medicine. Today, I cannot imagine how out of the loop I would be if were not on Twitter. I am constantly hearing about policy updates, new reports and guidelines, meetings, and hot-off-the-press papers. I also get to hear commentary from other professionals on the relevance, meaning, and importance of these advancements. There is no other outlet for sharing candid opinions about scientific issues in real time (given the slow pace of published commentaries and the infrequency of face-to-face scientific meetings). Lastly, Twitter helps me meet and connect with early stage professionals who may be a great fit with my lab. So, my take home message is if you’re not on Twitter, you probably should be!
 

What other advice do you have for early-career professionals, particularly as it relates to the use of technology to advance behavior change science and practice?

My biggest piece of advice is to start to reach out and work with people in other disciplines as soon as you can. It might mean going to talks in other departments or attending conferences outside of your field, which I did as a graduate student. As a post-doc, I was introduced to wonderful colleagues in computer science, whom I’ve now worked with for 12 years. Five years ago, I started collaborating with environmental epidemiologists. In each of these interdisciplinary situations, there were many times that I felt uncomfortable because I had no idea what they were talking about, and I thought they would not be interested in what an applied behavioral scientist had to say. In order to make true progress in what we are doing, we need to get ourselves out of our comfort zones quickly.
 

Contact Dr. Dunton at dunton@usc.edu and find her on Twitter at @GenevieveDunton.