Methodological Flaws in Aubry, RL, Power, GA, and Burr, JF. An Assessment of Running Power as a Training Metric for Elite and Recreational Runners. J Strength Cond Res 32: 2258-2264, 2018.

Abstract

Request for Clarification: In the study by Aubry et al., (1) the authors examine the effects of 3 variables: surface, speed, and subject to assess how well “metabolic demand” (p. 1) and “running power” (p. 1) correlate. However, their protocol and analysis contain major methodological flaws that interact to produce misleading conclusions. 1) Surface: How can the authors (1) determine what changes are due to surface vs. methodological differences when they collect the data at different interval periods for different conditions? The authors measured well before steady state for their treadmill tests, but much later overground, and thus likely conflated measurement error with surface condition–dependent variation. On the treadmill, the latest measurement began at 1:30 minutes, but overground, the latest measurement began at 3:30 minutes. It takes much longer than 1:30 for to reach steady state, leading to large differences in between the 1:30 and 3:30 measurement times, with larger differences for faster speeds (4). This leads to different errors for different speeds and subjects, errors that are large compared with differences in oxygen consumption across surfaces (5). These methodological errors preclude accurate analysis of correlation between oxygen consumption and Stryd power across surfaces. 2) Speed: How can the authors (1) determine the effect of speed when they completely remove its effect by normalizing 2 values by speed that are linearly dependent on speed? The authors report a speed-normalized power to speed-normalized correlation, thereby negating variation due to speed (2). Several readers have mistakenly understood their correlation numbers to represent power to correlation, and the article1 encourages the error by not canceling the time units in the numerator and denominator of cost of transport and using the phrase “metabolic demand,” (p. 1) rather than the accurate and physiologically accepted term “cost of transport” (p. 1 in (3)), which does not vary across speed (3). This could be addressed by replacing Figure 1 with the actual power- correlation. 3) Subject: How can the authors (1) assess an individualized training metric when they aggregate the data by subject before performing the correlation analysis? Within-subject correlation between and other variables is the relevant metric for training and racing (6). Aggregating data across subjects is statistically appropriate when the relationship that one is trying to determine is across all subjects. However, because the relationships here are subject-specific, data should only be aggregated across multiple measurements of the same subject.