Different Data Pre-processing Minimally Affects Algorithm-derived Threshold Values On Average But Reveals Inherent Algorithm Variability

Cardiovascular exercise intensity constitutes an important component of exercise prescription. Prior research shows that physiological threshold-based exercise prescription better improves fitness than standardized percentages of maximal anchors. Locating thresholds with ventilatory data typically requires pre-processing steps of outlier elimination and data averaging in addition to algorithm selection. However, the influence of data pre-processing steps on algorithm-derived thresholds remains unclear. PURPOSE: To quantify the effects of outlier removal and data averaging on the average value and variability in algorithm-determined exercise thresholds. METHODS: Adults at a clinic (n = 295) completed 350 total VO2max tests. Ventilatory data were processed using all combinations of three outlier removal approaches and seven data averaging durations. The respiratory compensation point (RC) and ventilatory threshold (VT1) were identified with common algorithms using the VE vs. VCO2 and VCO2 vs. VO2 relationships, respectively. The VO2 and other threshold values were regressed against the different outlier removal, data averaging, and algorithm conditions. The 95% limit of agreement (LOA) for VO2 and other threshold values between different data processing and algorithm combinations were calculated and regressed against the outlier limit comparison, algorithm comparison, and averaging duration difference. RESULTS: Main effects and several interactions between outlier limit, averaging duration, and algorithm conditions were significant for many variables at both VT1 and RC (p < 0.05). Effect sizes were often within the expected measurement error. There were also significant main effects and interactions for the LOA of VO2 and other variables at each threshold for the outlier limit comparisons, algorithm comparisons, and the averaging duration difference (p < 0.05). The algorithm comparison accounted for the greatest variability in LOA. The LOA for different algorithm comparisons ranged from ±13-32% VO2max. CONCLUSION: Data processing choices often exhibit negligible impacts on what algorithms select for VT1 and RC on average. However, different data pre-processing reveals that many algorithms may be inherently variable.