Assessing individual response to training in sport and exercise

A conceptual and statistical review

Abstract

Researchers are increasingly exploring contexts where training causes meaningful differences in the changes experienced by participants across interventions. Where this occurs, the phenomenon is referred to as individual response or trainability and provides scope for personalising training to maximise improvements based on participant characteristics. The potential for training to cause individual response in a given population is commonly assessed by comparing the variability in observed change between an intervention and control group. Similarly, the most common statistic used to quantify the difference is the standard deviation of individual response (SD_IR). It has been recommended that preliminary studies estimate the SD_IR to identify areas where personalising training may provide substantive improvements over prescribing the same, usually standardised, training to all participants. The purpose of this review was to provide a detailed examination of the SD_IR including conceptual and statistical overviews. A series of different plausible data generating models were used to highlight where the SD_IR appropriately assesses individual response, and where the standard formulation may lead to erroneous conclusions. The review highlights the importance of expressing uncertainty in estimates, comparing three different approaches to creating confidence intervals. It is recommended that ‘melded’ confidence intervals be used, especially for studies investigating relatively small sample sizes. The review also shows how model misspecification in terms of different measurement error distributions between intervention and control, and variance heterogeneity in external factors may represent the most pressing threats to valid conclusions when estimating the SD_IR. It is recommended that future research assess the potential for model misspecification and variance heterogeneity. Repeated measurements pre- and post-training can be used to better estimate the SD_IR  and account for differences in group measurement error. The existence of variance heterogeneity should be relatively simple to identify, however, it will be important for research teams to consider the best measures to capture the wide range of external factors that may influence observed change in outcomes included pre- and post-training.