Can we predict long-duration running power output? A matter of selecting the appropriate predicting trials and empirical model

Abstract

When facing a long-distance race, athletes and practitioners could develop an efficient pacing strategy and training paces if an accurate performance estimate of the target distance is achieved. Therefore, this study aims to determine the validity of different empirical models (i.e. critical power [CP], Power law and Peronnet) to predict long-duration power output (i.e. 60 min) when using two or three time trial configurations. In a 5-week training period, fifteen highly trained athletes performed nine-time trials (i.e. 1, 2, 3, 4, 5, 10, 20, 30, and 60 min) in a randomized order. Their power-duration curves were defined through the work-time (CP_work), power-1/time (CP_1/time), two-parameter hyperbolic (CP_2hyp), three-parameter hyperbolic (CP_3hyp) CP models using different two- and three–time trial configurations. The undisclosed proprietary CP models of the Stryd (CP_stryd) and Golden Cheetah training software (CP_cheetah) were also computed as well as the non-asymptotic Power law and Peronnet models. These were extrapolated to the 60-min power output and compared to the actual performance. The shortest valid configuration (95% confidence interval < 12 W) for CP_work and CP_1/time was 3–30 min (Bias: 8.3 [4.9 to 11.7] W), for CP_stryd was 10–30 min (Bias: 4.2 [− 1.0 to 9.4] W), for CP_2hyp, CP_3hyp and CP_cheetah was 3-5-30 min (Bias < 5.7 W), for Power law was 1-3-10 min (− 1.0 [− 11.9 to 9.9] W), and for Peronnet was 4–20 min (− 3.0 [− 10.2 to 4.3] W). All the empirical models provided valid estimates when the two or three predicting trial configurations selected attended each model fitting needs.