LAUSR

To the Editor: We would like to thank Dr Sparks for her comprehensive review of research on individual differences in the responses to exercise training [1]. It was especially helpful to see some of the statistical issues touched upon in this review [2]. We believe that these statistical factors are crucial for answering the fundamental question of whether there are true and clinically important individual differences in the response to exercise. By ‘true’, we mean individual response differences that are not merely random trial-to-trial variability in disguise. By ‘clinically important’, we mean individual differences that exceed a well-rationalised minimal clinically important difference (MCID).We alsomaintain that, in the ‘roadmap’ for researching this topic, true and clinically relevant individual response differences should be confirmed empirically before any moderators and mediators of the exercise response are explored [3]. The definition of ‘non-response’ was given in the review by Lauren Sparks [1] as, ‘the lack of a difference between a control and a treatment condition with respect to a specific variable’. This definition implies that researchers can identify non-responders simply by looking at their data from a twocondition (control and exercise) experiment and concluding that those participants with a treatment − control difference close to zero or in the opposite direction to that expected are identified as ‘non-responders’ (the latter are sometimes labelled ‘adverse responders’). The fallacy of this approach was hinted at by Barker and Schofield [2], but the full implications of this issue were not explicitly described. We have provided a full account of the pitfalls in non-responder identification [3, 4], and think that they can complement the useful review by Sparks [1]. An observed response is comprised of the true response as well as random trial-to-trial within-individual variability [3]. Therefore, observed non-response to exercise, or any other treatment, does not necessarily mean that there has been a true non-response. In Fig. 1, we present some simulated data that appears to show that individual participants differed substantially in terms of their acylated ghrelin response to exercise. It is well documented that exercise causes a reduction in the mean concentration of acylated ghrelin [5]. For the simulated data in Fig. 1, the mean ± SD reduction in acylated ghrelin was 18.1 ± 23.1 pg/ml (95% CI 1.6, 34.7). Nevertheless, it appears as though there are four non-responders in this sample of ten participants, according to the definition provided in Dr Sparks’ review [1]. In reality, true individual differences in response to exercise do not exist in the data presented in Fig. 1. In our simulation we subtracted exactly 25 pg/ml of ghrelin from each participant’s control condition measurement. We then added the component of a typical magnitude of random trial-to-trial variability. The trial-to-trial correlation coefficient was 0.77 (95% CI 0.27, 0.94). This random variability in biological measurements from day-to-day or week–to-week is always present and uncontrollable. Importantly, it is this component of * Greg Atkinson [email protected]