Self-reported Training Variables Are Poor Predictors Of Laboratory Measures In A Heterogenous Group Of Cyclists

Abstract

PURPOSE: Cycling is a predominantly aerobic activity that depends on a range of physiological attributes, as well as genetic, dietary and lifestyle factors. It is unknown to what extent laboratory-measured physiological and performance characteristics are predicted by individual training factors such as intensity, duration, distance, coach supervision, level of competition and training experience in cycling. METHODS: Fifty-two men and 18 women completed a training questionnaire and performance tests generating 14 outcomes (incremental cycling test [8 outcomes]; 30-s Wingate test [4 outcomes]; and 4-km cycling time-trial [2 outcomes]). LASSO (least absolute shrinkage and selection operator) regression was performed for each performance outcome including demographic and training factors as potential predictors. Continuous regression inputs were scaled by dividing values by two standard deviations to facilitate comparisons with binary predictors and assist with model interpretation. Models were generated using the glmnet package in R with associations described by regression coefficients and percentage inclusion in 10000 bootstrap samples. RESULTS: Laboratory measures indicated a heterogenous group of athletes, as demonstrated by the range of maximal oxygen uptake values (VO2max, range: 26.3 - 69.8 ml·kg-1·min-1). LASSO models identified that demographic factors were the most influential predictors of laboratory variables, with sex (76±37% inclusion), age (55±27% inclusion) and height (55±40% inclusion) featuring consistently in bootstrap samples across outcomes. In contrast, no discernible patters were identified for training factors. When training factors did appear consistently in a model, the regression coefficients were small and median estimates of the best training predictors were equal to 15.1±7.4% of sex or 30.6±14.5% of the next most influential demographic factor. CONCLUSIONS: Self-reported training variables were poor predictors of physiological and performance measures in a heterogenous group of cyclists, while demographics such as sex, age and height were greater predictors of these variables. A lack of a properly structured or implemented training program might explain the low predictive ability of training variables towards these laboratorial outcomes.