LAUSR

Nicholson et al. [1] recently reported that several different training methods improve short-sprint performance in football code athletes, an area of significant interest for practitioners. We commend them for pre-registering their protocol and reporting their methodology in detail, because rigorous and comprehensive reviews that determine the impact of an intervention are vital for evidence-based practice. However, we believe their conclusions about the potential effectiveness of these training interventions are mistaken and practitioners/coaches should heed caution when interpreting and applying the results of the review. While the results indicate statistically significant improvements in sprint performance, we cannot determine whether these have been caused by the training interventions due to the statistical methodology used in the review. Typically, a pairwise meta-analysis synthesises the between-group effect from comparative (ideally randomised) trials to determine the effectiveness of one intervention compared to another, such as a control group. In a trial, if the comparison groups are sufficiently similar at baseline (again, ideally achieved through randomisation), then we can infer that a difference between the groups following the trial was caused by the intervention being tested [2, 3]. This approach was not used by Nicholson et al. [1], who instead conducted their meta-analysis using the preand post-intervention data, pooling the within-group change for each of the 220 intervention groups and 64 sport only groups from the 121 included studies. Analysing a single group’s change in a trial, as opposed to the between-group effect, is misleading because several statistical phenomena are ignored and not accounted for, such as regression to the mean and natural history [2, 3]. In doing so, the data are now equivalent to cohort studies, which provide very limited inference about causal effects [4]. Similarly, when these within-group changes are pooled in a meta-analysis, as Nicholson et al. have done, we can only infer from the pooled estimate that sprint performance may have changed (though we cannot be sure of this), but we cannot determine that this has been caused by the training intervention. Therefore, we cannot confirm whether specific training interventions improve short-sprint performance in football code athletes. The more appropriate statistical method for meta-analysis in this scenario to infer whether specific training interventions cause improvements in short-sprint performance in football code athletes is to analyse the between-group difference from each trial (e.g., strength training vs control). This would allow us, for example, to determine whether using tertiary training interventions (such as strength training) to improve short-sprint performance is more effective than not doing so, one of the questions Nicholson et al. sought to answer with their review. We examined the 121 included studies in Nicholson et al. [1]. Eighty-five of these were randomised trials (54 of which contained an appropriate control group according to the methodological quality assessment by Nicholson et al.); the between-group effect from this study design is ideal to synthesise in meta-analysis when trying to determine whether an intervention caused an effect. Another 29 were non-randomised comparative trials; while not ideal, there are methods for incorporating the betweengroup effect from this type of study design into meta-analysis [5], which can be pertinent in situations where it is difficult to conduct randomised trials. However, it is important This comment refers to the article available at https:// doi. org/ 10. 1007/ s4027902001372-y.