Human running performance from real-world big data

Wearable exercise trackers provide data that encode information on individual running performance. These data hold great potential for enhancing our understanding of the complex interplay between training and performance. Here we demonstrate feasibility of this idea by applying a previously validated mathematical model to real-world running activities of approximate to 14,000 individuals with approximate to 1.6 million exercise sessions containing duration and distance, with a total distance of approximate to 20 million km. Our model depends on two performance parameters: an aerobic power index and an endurance index. Inclusion of endurance, which describes the decline in sustainable power over duration, offers novel insights into performance: a highly accurate race time prediction and the identification of key parameters such as the lactate threshold, commonly used in exercise physiology. Correlations between performance indices and training volume and intensity are quantified, pointing to an optimal training. Our findings hint at new ways to quantify and predict athletic performance under real-world conditions. Laboratory performance tests provide the gold standard for running performance but do not reflect real running conditions. Here the authors use a large, real world dataset obtained from wearable exercise trackers to extract parameters that accurately predict race times and correlate with training.