Analyzing activity and injury risk in elite curling athletes: seven workload monitoring metrics from session-RPE

Abstract

Read online

ObjectiveThe study aimed to compare the differences in the performance of seven session-rating of perceived exertion (RPE)-derived metrics (coupled and uncoupled acute: chronic workload ratio (ACWR), weekly ratio of workload change, monotony, standard deviation of weekly workload change, exponentially weighted moving average (EWMA), and robust exponential decreasing index (REDI)) in classifying the performance of an injury prediction model after taking into account the time series (no latency, 5-day latency, and 10-day latency).DesignThe study documented the RPE of eight curlers in their daily training routine for 211 days prior to the Olympic Games.MethodsSeven Session-RPE (sRPE)-derived metrics were used to build models at three time series nodes using logistic regression and multilayer perceptron. Receiver operating characteristic plots were plotted to evaluate the model’s performance.ResultsAmong the seven sRPE-derived metrics multilayer perceptron models, the model without time delay (same-day load corresponding to same-day injury) exhibited the highest average classification performance (86.5%, AUC = 0.773). EMWA and REDI demonstrated the best classification performance (84.4%, p < 0.001). Notably, EMWA achieved the highest classifying accuracy in the no-delay time series (90.0%, AUC = 0.899), followed by the weekly load change rate under the 5-day delay time series (88.9%, AUC = 0.841).ConclusionEWMA without delay is a more sensitive indicator for detecting injury risk.

Keywords

• session-RPE
• injury classification
• neural network
• workload
• injury risk