Walking metabolic rate prediction equations are commonly used to estimate oxygen consumption, exercise intensity and energy expenditure across a wide range of ages and anthropometrics. Despite their widespread use, independent validations of these equations using metabolic data from a large number of individuals are uncommon.

PURPOSE:​ To assess the accuracy of the commonly used ACSM and Pandolf walking metabolic rate prediction equations, along with two new walking metabolic rate predictions equations developed by Weyand et al. and Browning et al., using data from a large number of adults.

METHODS:​ We used demographic, anthropometric, walking speed, and oxygen consumption data from several laboratories (N = 450 (164 Males, 286 females), 18-85 years old, 16.5-44 kg/m²). We estimated oxygen consumption using each prediction equation in 1,078 walking trials ranging from 0.55-2.18 m/s, and 0.5-12% grade. Comparisons between predictive methods were made for all walking trials, as well as among normal weight participants during level and gradient walking, and overweight and obese participants during level and gradient walking. We computed the mean prediction difference (MPD) as the difference between predicted vs. measured rates of oxygen consumption (ml/kg/min) for each trial, and examined the relationship between the MPD and measured oxygen consumption (ml/kg/min) using modified Bland-Altman plots. Linear regression was used to determine the intercept (fixed bias) and slope (proportional bias) for each equation. The absolute value of the mean prediction difference, and Root Mean Square Error (RMSE) values were also calculated for each equation and population.

RESULTS:​ For level walking, all prediction equations had mean prediction differences that were statistically different from zero (P ≤ 0.05) except for the Browning et al., equation when applied to normal weight individuals and the Pandolf equation when applied to overweight and obese individuals. Most importantly, all prediction equations had significant (P ≤ 0.05) fixed and proportional bias, and demonstrated large RMSE (7.8-23.5% of mean measured metabolic rate) that were similar across equations and population. In addition, prediction error increased as measured metabolic rate increased for all equations.

CONCLUSION:​ The metabolic prediction equations evaluated here each had considerable error when compared to measured values, regardless of the population in which the equation was created and/or validated. Improvements in prediction equations may require using approaches that aim to minimize RMSE and/or developing population/intensity specific equations.