Consistency of pacing profile according to performance level in three different editions of the Chicago, London, and Tokyo marathons.

Running pacing has become a focus of interest over recent years due to its relationship with performance, however, it is still unknown the consistency of each race in different editions. The aim of this study is to analyze the consistency of pacing profile in three consecutive editions of three marathon races. A database of 282,808 runners, compiled from three different races (Chicago, London, and Tokyo Marathon) and three editions (2017, 2018, and 2019) was analyzed. Participants were categorized according to their time performance in the marathon, every 30 min from 2:30 h to sub-6 h. The relative speed of each section for each runner was calculated as a percentage of the average speed for the entire race. The intraclass correlation coefficients (ICC) of relative speed at the different pacing section, taking into account the runner time categories, was excellent over the three marathon editions (ICC > 0.93). The artificial intelligence model showed an accuracy of 86.8% to classify the runners' data in three marathons, suggesting a consistency between editions with identifiable differences between races. In conclusion, although some differences have been observed between editions in certain sections and marathon runner categories, excellent consistency of the pacing profile was observed. The study of pacing profile in a specific marathon can, therefore, be helpful for runners, coaches and marathon organizers for planning the race and improving its organization.