Authors' Reply to Li et al.: "Alternative Statistical Analysis Shows Exercise Training-Induced Improvements in Peak VO2 are Clinically Significant".

Abstract

We thank Dr. Li et al. [1] for their commentary on our recently published meta-analysis in Sports Medicine [2]. While there is no doubt that there is an argument for the use of a random-effects model in the presence of significant heterogeneity [3], such a model can introduce inherent problems of its own. Our fixed-effects model for our primary peak VO2 analysis yielded a mean difference (MD) of 0.62 mL kg min and a 95 % confidence interval (CI) of 0.47–0.77. One can observe how narrow the 95 % CI is. While the random-effects analysis of Li et al. [1] produced a larger MD of 1.31 mL kg min, the 95 % CI of 0.94–1.69 was much wider—almost triple the width of our 95 % CI. This 95 % CI width suggests much larger variation in how patients responded to exercise training. While we acknowledge the CIs do not overlap in these two analyses, one should note that in the analysis of Li et al. [1] at least 5 %, and more likely 10 %, of patients still do not achieve a clinically significant benefit of [1 mL kg min improvement. One needs to be mindful of this and we therefore chose to take a more conservative approach, which we believe was supported by the Sports Medicine reviewers’ request for us to state the absolute improvements in peak VO2 in L min. While we agree the misuse of fixed-effects models can lead to devastating impacts on clinical practice, noting the rosiglitazone example given by Li et al. [1], in our case the approach taken was conservative. Point estimates, in this case MD, do not convey the complete message and CIs reveal more about the response in all participants, not just those close to the mean or median. In turn, we argue that narrow CIs give us exactly that, more confidence in predicting responses in the whole group, while wide CIs render the point estimates less meaningful. In addition, an improvement in peak VO2 of 0.62 mL kg min, rather than an improvement of 1.31 mL kg min, is not going to change one’s risk of death from cardiovascular causes, nor is it going to reduce the use of exercise as a potential prescription for this cohort. In fact, any improvement in peak VO2 is associated with a reduction in cardiovascular mortality in this particular cohort [4]. With regard to the sensitivity analyses, we could argue that removing the study by Crowther et al. [5] from the subanalysis by Li et. al. [1] appears a rather arbitrary choice. The work by Crowther et al. [5] is one of the longest duration studies (52 weeks) and scores 6 for study quality. We would argue this is one of the more robust interventions and perhaps indirectly reflects the 60 % analysis weighting given to this work. Several other studies scored less than 6 for study quality, and are therefore more deserving of removal during sensitivity analyses. For example, we removed Hiatt et al. [6], Hodges et al. [7], and Savage et al. [8] (all scored 5 on quality), with a negligible reduction in effect size for peak VO2. Regarding the vigorous subanalysis, the MD of 1.64 mL kg min (95 % CI 1.11–2.16; p\ 0.00001) reported by Li et al. [1] is not different from our metaanalysis results [MD 1.42 mL kg min (95 % CI 1.04–1.80; p\ 0.00001)] because the 95 % CIs do overlap. Again, we note that the 95 % CIs of Li et al. [1] are wider than ours. Regarding the program length subanalyses, we again note the results for the 12 weeks or fewer group are not B. J. Parmenter Faculty of Medicine, University of New South Wales, Sydney, NSW, Australia