Fitness & Data Science
If you took a weekend run through your favorite park a couple of years ago, you might have spotted an occasional runner with a fancy watch, tracking not only distance but also pulse. What was a rare instance back then has become a commonality now. With the rise of wearables, people who engage in sports on a regular basis have started to track their fitness data in quite a professional manner. But with fitness tracking wristbands, for example fitbit and jawbone, people also began tracking their physical state throughout the entire cycle of activities: sleeping, eating, walking, running. Naturally, there are vast amounts of data points aggregating and some of the best data scientists worldwide are tasked with the challenge to generate insights, which have the potential to massively impact the way we think about healthcare and fitness.
Understanding the relationship between health metrics
For individuals, the beauty of tracking their fitness and health condition comes from its convenience. If equipped with the wearable technology, data is collected automatically and will be nicely displayed as a dashboard on your computer’s screen. Naturally, tracking is the basis for improvement and individuals can use their data to enhance motivation and enforce self-discipline through supervision. For the individual, though, it’s still rather a nice-to-have item than an actual necessity – it improves keeping track of health metrics, but little is known about the actual interdependencies between them. Data science, though, can – when data sets of millions of users are available – generate insights, which can solidify the understanding of the relationships between different health metrics. This can be especially eye-opening, because a lot of the fitness trackers allow to log for an individual assessment of energy level, quality of sleep and overall “happiness”. Those relationships might be established in observational data, but are to be taken with caution. Correlation in a lot of cases – and this is a debate, which has repercussions all across academia – is not causation, and randomized controlled trials (RCTs) would probably be needed to effectively challenge the observed relationships.
Nutrition & health metrics – what diet could work?
Another exciting avenue is the connection of health metrics with nutrition data. As people enter information about their daily diet (literally: breakfast, lunch, snacks, dinner), this data – aggregated over millions of users – can be merged with their their health metrics and weight developments. Ever read a column in one of those trashy magazines about the new diet which promises weight loss within a few weeks? Well, fitness tracking devices might be the key to finally solving that problem (and putting a lot of shady diet-columnists out of business).
In the case of nutrition, though, results have one major disadvantage: users need to manually fill out their daily diet. Some might miss a chocolate bar here or there, or omit a glass of wine. This bias is inherent in the data and is something that needs to be considered and statistically be accounted for. Promisingly, though, this field allows for the conduction of randomized controlled tests. One group with similar health metrics is split, and one sub-group is randomly assigned to a diet plan, while the other continues with their regular nutrition (control group). Assuming that participating users adhere to the diet plan with stringency, it could reveal new insights into what really drives weight loss – an insight, which would be of great value for insurance companies and fitness companies alike.
Unlocking motivation – a behavioral science perspective on what drives people’s self-discipline
Data science on fitness tracking data could similarly contribute to research on motivation. The more insights into relationships between health metrics evolve, the more valid will recommendations for individual improvements become. The “fitness app” could recommend users to spend at least 20 minutes per day walking, in order to have a healthy daily activity level. Once users commit to this goal, it is fairly easy to track their progress in the upcoming days and weeks (well, that’s what fitness trackers are all about in the end). In the process, various experiments – informed by behavioral science – could be run on the users to enhance their motivation. They could get a reminder in the morning, they could get comparison scores, they could be “rewarded” – there are various ways in which these strategies could be applied. Being able to measure the direct impact of those strategies on physical behaviour across thousands of individuals bears the potential to not only improve the health of the participants, but also shed light on motivational strategies for physical exercise. Thus, it is one of the most exciting areas where data science and behavioral science cross-pollinate each other.
The road ahead?
Fitness and health data bear enormous potential, and benchmarked against this vision the above mentioned insights generated by data science appear rather small. As this might be realistic assessment, this work builds the foundation for future research and insights. Being able to fully understand the drivers of behavioral change and the impact of physical activity on health metrics is essential to turn onto more promising avenues. Reliably predicting diseases and illness of all kind and thus conceptualizing adequate counter-strategies is still a lofty vision, but with the amount of fitness data increasing by the minute, and machine learning models advancing at similar speed, real progress might soon kick in.