Acute Exercise and Hormones Related Appetite Regulation: Comparison of Meta-analytical Methods
- PDF / 177,684 Bytes
- 2 Pages / 595.276 x 790.866 pts Page_size
- 76 Downloads / 176 Views
LETTER TO THE EDITOR
Acute Exercise and Hormones Related Appetite Regulation: Comparison of Meta-analytical Methods M. M. Schubert • B. Desbrow S. Sabapathy • M. Leveritt
•
Ó Springer International Publishing Switzerland 2014
Dear Editor, We thank you for the opportunity to discuss our recent paper [1] and the concerns brought forward by Professor Atkinson and colleagues [2]. We also thank Professors Atkinson and Stensel and Ms. Douglas for bringing these issues to our attention. These comments highlight potential issues in meta-analyses of crossover trials that may often be overlooked [3]. We acknowledge that we did not initially account for the crossover nature of the studies in our analysis. Thus, as Atkinson and colleagues have suggested, our results may have yielded a more conservative outcome in comparison to other methods because we treated the data as independent samples. Given the volatile nature and the relatively small levels of these hormones in the circulation, we felt it necessary to exercise caution [4]. However, in deference to these concerns, we re-performed the analysis using acylated ghrelin (AG) as an example with paired comparisons via three methods. First, correlation coefficients were calculated as per Elbourne and colleagues [3] from the reported means and standard
M. M. Schubert (&) B. Desbrow School of Public Health, Research Centre for Health Practice Innovation, Griffith Health Institute, Griffith University, Southport, QLD, Australia e-mail: [email protected] S. Sabapathy School of Rehabilitation Sciences, Research Centre for Health Practice Innovation, Griffith Health Institute, Griffith University, Southport, QLD 4222, Australia M. Leveritt School of Human Movement Studies, University of Queensland, St. Lucia, QLD, Australia
deviations. The second method was performed similarly, except that a paired t statistic was calculated for each study. If an exact p value was reported, the inverse of the t distribution (TINV) function in Microsoft Excel was used to calculate the t statistic from the p value and degrees of freedom. If no exact p value was reported, the t statistic was calculated as follows: (difference in means)/[(standard deviation of the difference)/(H(n)]. This t statistic was then applied to the formula (t2)/[(t2) ? (n - 1)] to estimate R2 [5]. Finally, the square root of R2 was taken to yield the correlation coefficient r. These estimates were then compared with the standardised mean differences we had reported previously. The median correlation coefficients for the first two methods were 0.68 (range 0.11–0.94) and 0.60 (range 0.11–0.92), respectively. The third method used the estimated paired t statistic, means, and sample sizes to estimate the effect sizes (ESs) and standard error. Results of the meta-analyses for AG are shown in Table 1 below. As seen in Table 1, the methods using the estimated correlation coefficients yielded estimates very similar to our original model, albeit with a tighter confidence interval. More individual studies had ESs that sig
Data Loading...
