Health Data Investments Potentially Wasted?

Digital Health adoption accelerated by three years as a result of the COVID-19 pandemic. Corporate funding increased by 103% between 2019 and 2020, capping at $21.9 billion in the US alone. So why are engagement metrics – particularly around health behaviour change – still so disappointing?

The influx of resources creates an abundance of opportunity for digital health research and business, but also casts a spotlight on the limits of its efficacy.

Through smart watches, mobile phones and a plethora of other devices, we can seemingly provide a person with every measurement they might need to get a full picture of their current health status and future health risks. But when we design digital health products, our hope is that people will not just engage with them. We also want people to feel empowered to make changes to their behavior, to live happier and healthier lives.

Much of the digital health and healthcare data science culture is based on the assumption that personalised Health Data itself can inform and motivate, or "nudge", digital health users to make healthier lifestyle choices. However, we might be experiencing a little bit of "shiny object syndrome", in the hope we place on personalised health data bringing about real life behaviour change. In reality, we often get stuck behind low adoption and engagement rates. Only a fraction of the digital health products out there are implemented by users in any way that would give them enough input to successfully change behaviour.

In fact, before the pandemic hit in 2020, adoption of digital health had actually dropped, from 48% to 35% from the previous year. This seems to be partly due to what users perceived as "low quality digital experiences". To empower change in people’s health behaviour, we must first engage them fully. Engage them enough to adopt the digital health tool. And then engage them more to continue using it for long-term benefit.

Is data enough?

The question is, why are all the health data analytics available to digital health creators and users not effectively engaging enough people?

The answer might lie in another assumption we often make about human behaviour.

Behavioural science is a discipline that grew out of psychology and economics and endeavours to predict and explain human behaviour. How people act is often seen to be determined by whether they are being "rational" or "irrational". For example, maybe you’ve heard of "System 1 and System 2", the automatic and conscious thought processes brought to us by winner of the Nobel Prize Daniel Kahneman in his seminal text "Thinking Fast and Slow"?

Healthy lifestyle choices are seen as rational, unhealthy ones as irrational. Almost as if unhealthy choices are made when people have somehow lost control of themselves and their decision-making. Rational choices are seen as informed decision-making, most likely based on reliable data.

If we apply this behavioural science framework, health data must therefore help people become more rational so they can make healthier decisions. And what could increase rationality more than more data to base rational decisions on?

But the low adoption and engagement rates of digital health solutions suggest that this is not being the full story.

Emotional elephant in the room

The existence or even the promise of accurate and personalised health data is not enough to empower people to make and maintain healthy choices. Accurate prediction of human behaviour in response to access to digital health data requires deeper determinants of that behaviour to be built into our predictive analytics.

Emotions are more often in the driver’s seat of human behaviour than cognitive problem solving. Emotions also characterise much of how we experience our days and are an integral part of the story of our lives. What’s more, emotions are often operating at the subconscious or unconscious level of human functioning and direct our actions, decisions and behaviour patterns without intentional awareness from us.

Perhaps what is missing from our predictive data models in digital health is an inclusion of emotion metrics, which might help us account for a bigger proportion of human behavior drivers, predict behavior more effectively and engage users in digital health experiences more sustainably.

So, let’s write emotions into our prediction models. How does that work?

Metrics of Emotions

Integrating emotions into our predictive models of health behaviours might get us closer to realising the potential of digital health, but how can we measure emotion and at scale for big data analytics?

In digital health, emotions have mostly been conceptualised and measured as facial expressions. This stems from the developmental psychology theory of universal facial expressions from which we can infer what other people are feeling. Great use cases of this kind of emotion metric are tools like clinical decision-support systems, where a health professional could more accurately identify what a patient might feel based on their facial expression. Sometimes, we don’t feel like we can safely express emotion and sometimes we might not know how we feel, especially when in the vulnerable role of being a patient.

However, digital health technology aimed at helping people change their health behaviour patterns will not glean as much value from its users identifying other people’s emotions. One general approach to incorporating emotions into digital health tools would be to try to predict people’s emotional responses when they use the tool. What would be helpful for individual level behaviour change is having one’s own emotions brought into conscious awareness.

Creating implicitly positive emotional experiences in digital health products would be even better, which we could do by taking a leaf out of customer loyalty building techniques in brand creation and user engagement through in-depth emotional design in user-centred technology design.

Google HEART Framework

If you’re considering including emotions in your digital health data models predicting human behaviour change, you might want to use the Google HEART framework as a starting point.

Developed to increase user engagement in web-based products, the framework can be adapted for other types of technologies. The HEART framework structures our thinking about user engagement prioritising user "happiness". The full acronym stands for: Happiness, Engagement, Adoption, Retention and Task Success.

For each category of measurement, the framework also asks you to identify a goal, signal and metric aligned with that measurement.

When thinking about how to include emotions into your predictive models, use the Happiness category from the framework, but expand the definition of user satisfaction beyond the usual 5-point Likert scale question of "how satisfied are you with this feature/product?".

Instead, identify the values your users align with, the kind of information they trust and build your Happiness metrics based on those.

You’ll already be one step closer to increasing your user engagement, creating positive emotional experiences and the loyalty to your product that will help people achieve long-term health behaviour change.

Recommendations

You don’t need to have all the answers.

That can be a difficult statement to accept. As scientists, as data analysts or as designers of technology, we’re curious and we want answers.

We wrote this article to help data scientists create more effective models of human behaviour, and help them improve the adoption potential of new data-driven technology. But we don’t have all the answers either. We do believe that we can make things better by asking the right questions.

We recommend that you start by asking yourself the following questions:

• Do the predictions of your model rely on people making strictly rational – logical, sensible, reasoned – decisions? (If so, you do not currently have an accurate behavioural model. You need to go back to the drawing board with the whole model)
• Does adoption or implementation of your model in practice, rely on people making strictly rational – logical, sensible, reasoned – decisions? (If so, you do not currently have a model with the potential to change anything or help people at scale. You need to go back to the drawing board with your plan for adoption and implementation.)
• What’s one emotional experience your users/research population goes through and how might you express that as a metric in your data?
• Which colleagues, perhaps interdisciplinary and international collaborators, might be able to help you further integrate emotional experience into your models?
• In your next funding application, how might you demonstrate your understanding of the importance of emotional experience in determining behaviour?
• How might you emphasise user/participant engagement in your funding applications to emphasise efficacy/effectiveness potential?

Regardless of whether or not you can answer these questions positively right now, this can be your framework for improving the way that you address the role of emotion in data science.

Given the momentum created by the pandemic, Investment in academic and commercial research and development of digital health tools is likely to keep growing at an unprecedented rate. Personalized health data analytics and health behaviour change remain important goals of the digital revolution. Now is the time to pay more attention to the role of emotions in engagement and behavior change, if we’re to avoid the disappointment of a wasted investment.