The Dawn of Scaleable Behavioral Diagnostics in Health Care

Personalized nudges have become the norm across the digital landscape, with companies such as Netflix, Amazon, and Spotify at the forefront. It is time to apply similar behavioral diagnostics to health care.

Two years ago, Pharmacy Times graciously allowed me to write an article about the need for a digital diagnostic tool for patient behavior. My point back then was that because patients with chronic disease only take about half of their prescribed medications, we need to diagnose the behavioral risk of each patient and understand the drivers of that risk to better support them. I then described our own efforts at Observia to do just that, via the SPUR behavioral framework.

Since then, a lot has happened. For one thing, when I wrote that I remember wondering just how bad this new virus breakout in Wuhan was going to be...well, we know how that turned out. In fact, as I write this follow-up, I’m recovering from a bout with COVID-19, which was luckily mitigated no doubt by the vaccines rather remarkably developed and distributed in the interim. Chalk one up for modern medical science.

Another thing that happened is that a whole lot more work was done on the SPUR approach, notably, the elaboration and validation of a digital tool that allows a full patient diagnostic via an interactive patient experience. While the COVID-19 epidemic slowed down that work, it did at the same time accelerate the acceptance of digital interfaces in health care.

Throughout the world, patients and health care professionals alike have become better acquainted with digital tools as face-to-face interactions became increasingly problematic. That positive factor at least will remain even after the epidemic subsides.

When we combine the work on SPUR with the accelerated acceptance of digital health tools, we find that the entire concept of digital patient behavioral diagnostics has quite suddenly become de jour. In the 2020 article, I talked about blended patient support initiatives as something more or less of the future, but now we’re talking about now.

And think about it—we have been subject to fully digital, personalized behavioral modification tools for quite some time now in other domains. When you look at Netflix’s home page, it makes a bunch of suggestions to you about what you might want to watch.

Although the format of that home page is the same for everyone, its contents are entirely different. Not only does Netflix suggest a different mix of individualized programming—even if some of the films or shows are the same—they are likely not using the same trailers and are describing them differently because Netflix knows what makes each of us tick.

Of course, we each can choose whatever we like, Netflix isn’t removing any options, they’re just making different suggestions. This is right out of Thaler and Sunstein’s book Nudge—let people choose whatever they like, but make it easier for them.

Thaler and Sunstein primarily presented their approach with an eye to public policy, and “nudge units” have since shown up in governments around the world to help leaders shape policies that assist their populations in making healthy, prosperous choices. The idea of “nudging” is old hat in marketing circles, and personalized nudges have become the norm in the digital landscape on just about any platform we interact with regularly.

Hopefully, at this point, you’re already wondering why we aren’t doing this on digital health platforms for chronic patients. Well, we are to a degree, as many digital health platforms are already determining behavioral risk.

This is particularly the case for platforms that contain usable information about behavior, whether through regular clinical information—for example, glycemic control in diabetes—or treatment behavior, whether through disbursement records or via connected devices. Fundamentally, you can get an idea of a behavioral risk—such as non-adherence—the minute you have a set of data that include dependent variables—such as how often someone uses their inhaler—as well as potential independent variables, which are just about anything else—for example, use of the platform, age, and how long they use certain features.

You feed all the variables into a predictive analytics mill and it comes up with a magic formula to predict what you’ll do. This is easy to understand for Netflix.

Netflix has independent variables for what you watch, as well as dependent variables for what you do on the platform. The magic predictive analytics mill links consider what you have already watched with what you then watch to predict what you’ll like.

However, it looks at way more than that. It also takes into account everything from the time you spend looking at specific trailers to your age to your blood type (OK, that’s a bit of hyperbole, but believe me, if Netflix knew your blood type they’d throw it into the predictive analytics mill to see if it was significant). You can do the same thing with an app for chronic patients the minute you gather data on an interesting dependent variable and you have other, independent variables.

This is useful, but it doesn’t tell you why. It doesn’t tell you what’s behind the behavior.

Predictive analytics can help... well...predict, but they can’t give you any insights into how to nudge an individual to help them adapt their behavior in favor of their health. It can raise flags, but it can’t tell you anything about the flag bearer.

For that, you need to turn to deeper elements of behavioral science. You need to know why that particular patient has a behavioral risk and what is driving them to potentially not take their medication or to follow the treatment plan as a whole.

There are a number of frameworks that can help to reflect on this. The most widely used are probably the Health Belief Model and its subsequent variants, COM-B, the Transtheoretical Model, and most recently, SPUR.

The Health Belief model was developed in the 1970s and is an extremely useful way to break down the rational elements of patient behavior. It assumes that people will weigh the risks and benefits of following treatment and make a choice on the basis of this information.

That’s all well and good, and most patient support interventions are inherently based on this assumption, but the revolution in behavioral science that started with Kahneman and Tversky was entirely about the fact that we’re way less rational than we imagine. It’s important to provide information of course, but don’t rely on that to solve the problem.

The Theory of Planned behavior, developed in the 1980s, wasn’t created with health care in mind, but it does examine the less purely analytical aspects of decision-making by taking into account the impact of social considerations and perceived control, or ability to change behavior. It has been successfully used in designing patient support in numerous cases. COM-B, developed in 2011, explicitly considers capability, opportunity, and the motivation to change behavior in the domain of health care.

It is perhaps the most widely used framework and can help considerably to think about how to nudge any individual. Lastly, the transtheoretical model is more of a procedural tool. It considers the process of changing health-related behavior, postulating several specific steps people go through as they modify their behavior.

Digital engagement capabilities allow us to move to a grand unified theory of a sort, designed not for health-related behavior in general, but specifically for patients treated for chronic disease. Because of all the previous work, there wasn’t a need to necessarily bring any new ideas to the party.

The opportunity to bring together well-established behavioral models into an overarching framework represented a way to move forward with a combination of well documented tools. The ability to postulate numerous specific drivers for adherence behavior and grouping them into 4 broad categories (as follows) represents a very meaningful step forward:

• Social drivers. These include the degree to which the person feels that their disease and their treatment affects their place in society as a whole, and their relationships with people around them.
• Psychological drivers. These include specific factors that affect attitudes about treatment, including self-image, reactance to authority, and the degree to which the person values future health benefits.
• Usage factors. Here, we see the practical elements, including financial burden, self-efficacy and forgetfulness.
• Rational drivers. Here, we break down those cognitive elements about the patient’s understanding of the disease and the treatment.

I believe—and sermonize about it to the chagrin of most of my colleagues—that the most mission critical element of any new digitally administered behavioral tool is that we move well beyond segmentation strategies. Just like Netflix, we don’t need to create segments of people.

Each of us who is a chronic patient has some mix of the different drivers, leading us to our own health behavior. In a digital environment we can consider a complete patient profile, take into account the particular mix of drivers for any patient at any point in time, link that to their clinical and socio-demographic profiles and determine exactly what that patient needs, and when, in order to change their behavior: take their medication properly, modify their diet, quit smoking...all the things that their physician has outlined for them.

Two years ago, this was a bit more of a distant dream, because that degree of personalization requires a digital platform and not everyone was using digital health platforms. However, COVID-19, though, accelerated the uptake of digital tools dramatically.

Every time a patient does a remote visit with a physician or interacts with a remote pharmacy platform to renew their medication, there is an opportunity to nudge them. For that matter, they’re being nudged whether it’s intentional or not, the question is, in which direction?

If the patient’s behavioral profile is used then those subtle cues, the wording they see, the default content suggested is all going to come together to help them in the best possible way and dramatically increase their engagement with the platform and with their own health.

There are a few big differences, though, between the kind of personalization algorithms used by companies like Spotify or Netflix and those used for behavioral diagnostics in health care. One of those is the application of a robust scientific methodology. Netflix can be satisfied with the application of predictive analytics through a machine learning approach—a black box method to automatically determine what to suggest to people.

We owe patients and providers something better, something more robust, something grounded in solid behavioral science. There are several well-established behavioral diagnostic tools used in health care, including the Morisky questionnaires (MMAS-4 and MMAS-8), the Patient Activation Measure, and many more.

All of them have been subjected to rigorous scientific testing and any new tool has to be subjected to the same. All that is to say that the time has come to build patient support initiatives that go beyond the dry, purely clinical, and educational interventions that dominate today.

Regardless of which of the available, well-validated behavioral tools you use, you don’t need to settle for an undifferentiated approach. We can increasingly see this philosophy at play in new, digital therapeutic offerings around mental health, but outside of that domain, behavioral science is much less visible in patient support initiatives.

Sure, most content creators do their best to incorporate the kind of nudge approach championed by Thaler, but they pointedly do not use behavioral diagnostics to adjust it to individual patient needs.

What physician wouldn’t consider every patient’s unique clinical profile before making treatment decisions? We believe that 5 years from now, it will be as unthinkable to take that one size fits all approach in patient support as it would be for Netflix to suggest “Duck Dynasty” to me or a cooking show to my resolutely kitchen-averse wife.

References

1. Dolgin K. The SPUR Model: A Framework for Considering Patient Behavior. PPA. 2020; Volume 14:97-105. doi:10.2147/PPA.S237778

2. de Bock E, Dolgin K, Arnould B, Hubert G, Lee A, Piette JD. The SPUR adherence profiling

tool: preliminary results of algorithm development. Current Medical Research and Opinion. 2022;38(2):171-179. doi:10.1080/03007995.2021.2010437

3. Wells J, El-Husseini A, Jaffar A, Dolgin K, Hubert G, Kayyali R. A cross-sectional study to evaluate the validity of a novel patient-reported outcome measure of medication adherence in Type 2 Diabetes. International Journal of Pharmacy Practice. 2021;29(Supplement_1):i30-i30. doi:10.1093/ijpp/riab015.036