Test–retest errors and the apparent heterogeneity of training response

Abstract

Published reports have shown large apparent inter-individual differences of gains in maximal oxygen intake (V̇O_2max) in response to a standard 20-week programme of aerobic conditioning that progressed to 75% of the individual’s initial V̇O_2max. The observed gains of V̇O_2max ranged from 0 to 1,000 ml min⁻¹, with a coefficient of variation (CV) of 8.4%. The present analysis evaluates the potential contribution of test–retest errors to these apparent large inter-individual differences in training response. The 2-day test–retest CV for V̇O_2max readings in 742 healthy adults was initially 5.0%, dropping to 4.1% after training. Published training responses were estimated from the mean of paired measurements obtained before and after training if readings agreed by <5%, but from the highest of paired values if these differed by >5%. Taking account of the relative proportions of single and paired observations, the weighted V̇O_2max data for the entire sample had an effective 2-day CV of 4.3% before and 3.4% after training. Assumption 1: if the 20-week test–retest error remained similar to the 2-day figure, measurement error would contribute a CV of 5.5% to apparent training responses, or (for the stated initial mean V̇O_2max of 2,409 ml min⁻¹) an SD of 132 ml min⁻¹. Assumption 2: if the 20-week CV was similar to that in other long-term studies (~5%), measurement error would contribute a CV of 6.1%, or a SD of 146 ml min⁻¹. The published data show a total SD of 202 ml min⁻¹ for apparent inter-individual differences in training response, with age, gender, race and baseline V̇O_2max accounting for only 11% of this variance. After estimating the likely effect of test–retest measurement errors, the SD due to inter-individual differences would decrease to 138 ml minV̇O_2max (assumption 1) or 123 ml min⁻¹ (assumption 2). We conclude that when estimating the extent of inter-individual differences in training response, allowance must be made not only for the minor effects of recognized covariates (age, gender, race and initial fitness), but also for the larger influence of test-retest measurement errors. Nevertheless, substantial inter-individual differences persist after making such adjustments. The most likely explanation of these differences is a familial aggregation of training responses.