Preventative Care: A Data Science Problem

At Amazon, when a new idea was presented, one of the questions we asked was, “What needs to be true for this idea to work?” For Hobbes, a wellness coach designed to prevent chronic illness, the answer to that question is: preventative care is a data science problem.

Today, the healthcare system addresses preventative care in a few ways: annual physical exams and screenings, seasonal vaccinations, and public health campaigns & standardized guidance. If you’re a software engineer, this is analogous to running tests on your code once a year. Or like driving your car and only finding out you’re out of gas when the car stops. Healthcare systems around the world are designed and optimized for acute care - to take really good care of you when you fall sick. They are not great at preventing you from falling sick in the first place.

We need three things to be proactive about our health:

1. Continuous monitoring. Monitoring devices, such as smart watches and fitness bands & rings, are ubiquitous now and are gaining updated sensors every year. Continuous glucose monitors are now available without a prescription in the US. Beds track your sleep, blood tests are affordable, body scanners to measure fat and screen for cancers are becoming more common. We are entering an age where we can know how we eat and live affects our health, in real-time. It is not possible for any human, however qualified, to assess all these data, find correlations, patterns, and anomalies. We need data science techniques to make sense of these and generate meaningful insights.
2. Personalized recommendations. Each of us is different. How we live our days, what we eat, when we sleep, what stresses us out, what we are allergic to, what we enjoy doing, and what we dread - everyone is unique to each of us. The insights from our monitoring devices need to be converted to recommendations that fit within our lives. Don’t recommend that I eat fish to increase my protein intake if I am vegetarian, don’t tell me to follow the circadian rhythm to improve my sleep if I work the night shift, or don’t tell me to follow an hour-long workout program five days a week when I am stressed about meeting a work deadline. We need a system that understands who we are as individuals and adapts recommendations to our life.
3. Behavior change. The key to changing our lifestyle to be more healthy is to change our lifestyle. Behavior science has various strategies - understand cues, reward good behavior, create atomic changes, and build streaks for consistency. Yet, most of us don’t know what really works for us. What we really need to do is run short experiments - dozens of them - to figure out what works. Can I really wake up at 6am everyday and go for a walk? Can I fast for 16 hours? Do I like yogurt in the mornings? We might not know the answers to these questions until we try them. Then we learn, and try something else, until we find something that really works for us. That, then, becomes a habit and, subsequently, leads us a step closer to a healthier lifestyle. We need a system that encourages experimentation, analyzes the results and synthesizes the lessons learned, and suggests adaptations.

At the core of preventative care is understanding the individual - their lifestyle, its real-time impact on their health, their motivations, and how to facilitate positive change. Our current healthcare system cannot solve this problem. I posit that the only way to solve it is by using data science & machine learning.

I acknowledge that data science is not a magic wand. Socioeconomic factors, access to healthy food sources, privacy concerns, and bias & accuracy problems play very crucial roles in making a big dent in the problem of chronic illness. We also need doctors and healthcare professionals to ensure that the recommendations are valid and to provide support and motivation. My argument is that data science is an indispensable component.