Cross-Validation of VO2Peak Prediction Models in Adolescents

Background/Purpose: Cardiorespiratory endurance is a major component of health-related fitness testing in physical education. FITNESSGRAM recommends the use of the Progressive Aerobic Cardiovascular Endurance Run (PACER), or alternatively, the One-mile Run/Walk (1MRW) to estimate aerobic capacity. No research to date has cross-validated prediction models from both the PACER and 1MRW using current FITNESSGRAM standards. The purpose of this study was to cross-validate prediction models from the PACER and 1MRW against measured VO_2Peak. Models examined included the Cureton 1MRW model, the Mile-PEQ, a New PACER model, and both a Linear and Quadratic PACER model. Method: Cardiorespiratory endurance data were collected on 86 adolescents (34 Girls, 52 Boys; Mean Age = 14.7 ± 1.3 years) on three separate testing days. Each student completed the 1MRW and PACER in a counterbalanced order, in addition to a maximal laboratory treadmill test to measure VO_2Peak. Analysis/Results: Validity was examined using multiple correlations (R), standard error of estimate (SEE), and paired t-tests. Individual agreement was assessed using modified Bland-Altman Plots, which examined the relation between VO_2Peak residuals and estimates obtained from each algorithm. Criterion-referenced (CR) validity was examined using modified kappa (Κq), proportion of agreement (Pa), and a phi (ϕ) coefficient. Multiple correlations across models with measured VO_2Peak were considered strong (R = 0.74 to 0.78, p < .001). Prediction error (SEE) was 4.295 ml^.kg^-1.min^-1 using the 1MRW model, and ranged from 4.206 ml^.kg^-1.min^-1 (Quadratic PACER) to 5.907 ml^.kg^-1.min^-1 (New PACER) using various PACER models. All PACER models, except the New PACER, displayed systematic bias in the prediction of VO_2Peak by overestimating measured aerobic capacity at low levels and underestimating aerobic capacity at higher levels. Classification agreement with measured VO_2Peak ranged from (Κq = 0.37, Pa= 0.73, Φ = 0.52) using 1MRW to (Κq = 0.52, Pa = .79, Φ = 0.72) using the New PACER algorithm. Conclusions: All models examined in this study displayed a strong linear relationship with measured VO_2Peak and acceptable prediction error. There was systematic bias in the prediction of VO_2Peak except when using the New PACER model. Classification agreement with measured VO_2Peak was considered moderate among all models. The results suggest that the 1MRW and various PACER models’ predictive accuracy were similar, with the New PACER displaying the strongest CR agreement with measured VO_2Peak and an unbiased prediction of measured VO_2Peak across the range of estimated cardiorespiratory endurance levels.