Associations between time-series features of training-load monitoring indicators and relative composite performance in elite speed climbers: a GEE and quantile regression study

Objective To examine associations between time-series features of training-load, biochemical, and subjective recovery indicators and relative composite performance in elite speed climbers, and whether these associations differed across conditional performance quantiles. Methods Nine elite speed climbers were monitored for 30 weeks, comprising 15 two-week monitoring periods. Six indicators were analyzed: training strain, training monotony, testosterone, creatine kinase (CK), perceived sleep quality, and soreness. Relative composite performance was represented by a sex-specific athlete-level score ranging from 0 to 1. Missing testosterone and CK values were handled using chained-equation imputation with predictive mean matching. Rolling means and coefficients of variation (CVs) were calculated over four consecutive monitoring periods, yielding 12 candidate time-series features and 108 rolling observations. Generalized estimating equations (GEE) assessed population-averaged associations, while athlete-clustered quantile regression explored associations at τ = 0.25, 0.50, and 0.75. False discovery rate correction and sensitivity analyses were applied. Results In the primary GEE analysis, CK rolling CV showed a nominal positive association with performance (β = 0.529, 95% CI 0.061–0.997, P = 0.027), but not after FDR correction (FDR = 0.322). In the six-predictor quantile-regression model, training monotony rolling mean was negatively associated with performance across three quantiles, whereas testosterone rolling mean was positively associated and CK rolling mean negatively associated. These directions were relatively consistent in the extended model and sensitivity analyses. Associations involving training strain, perceived sleep quality, and soreness were less consistent. Conclusions GEE and quantile regression provided complementary perspectives: GEE characterized population-averaged associations, whereas quantile regression identified differences across the conditional distribution of relative composite performance. Training monotony, testosterone, and CK showed consistent directional patterns. Rolling means and rolling CVs may provide complementary information for interpreting training-load, biochemical, and recovery monitoring in elite speed climbers. These findings require confirmation in samples with repeated objective performance outcomes.