The accurate prediction of marathon race times using only training-run distance and time data collected from the smartwatches of around 14,000 runners, is reported in a paper in Nature Communications. The study demonstrates that mathematical modelling may be used to estimate physiological parameters, using data collected non-invasively from wearable technology.
Real world data, such as that collected routinely in hospitals and point-of-care institutions, but also those now available from wearable devices, is projected to change healthcare dramatically. The data collected by wearable exercise trackers hold the potential to enhance our understanding of the interplay between training and performance.
Thorsten Emig and Jussi Peltonen linked smartwatch-collected running data from approximately 14,000 runners to their individual physiological parameters, via a mathematical model of human physiology. The authors used this framework to predict marathon race times accurately and identify key predictive parameters of running performance, such as lactate threshold, using only information on the distance and times of their training runs (around 1.6 million exercise sessions in total). Moreover, they provide insights into how features of training sessions are associated with race performance.
Being able to relate key physiological parameters to non-invasively collected continuous data presents opportunities for human health monitoring. For instance, exercise sessions could be tailored to an individual. The authors conclude that their findings hint at new ways to quantify and predict athletic performance under real-world conditions.