Impact of progressive resistance training on lipids and lipoproteins in adults: Another look at a meta-analysis using prediction intervals
Introduction
Cardiovascular disease affects an estimated 80 million American adults (about 1 in 3) and is the number one cause of mortality in the United States, accounting for an estimated 864,480 deaths annually (Lloyd-Jones et al., 2009). In terms of expenditures, the annual total direct and indirect costs associated with cardiovascular disease in 2009 have been estimated to be $475.3 billion (Lloyd-Jones et al., 2009). Less than optimal lipid and lipoprotein levels, a common problem among more than 98.6 million (45.3%) American adults, have been closely associated with the development of atherosclerotic coronary artery disease and subsequent mortality (Lloyd-Jones et al., 2009).
In a previous issue of Preventive Medicine, we reported statistically significant average treatment effect reductions in total cholesterol (TC), the ratio of total cholesterol to high-density lipoprotein cholesterol (TC/HDL-C), non-high-density lipoprotein cholesterol (non-HDL-C), low-density lipoprotein cholesterol (LDL-C), and triglycerides (TG), with no statistically significant changes in high-density lipoprotein cholesterol (HDL-C), as a result of progressive resistance training (PRT) in adult humans (Kelley and Kelley, 2009). These results were based on the preferred random-effects model, an approach that is especially appropriate when statistically significant between-study heterogeneity exists (Hunter and Schmidt, 2000). While our results were encouraging with respect to the potential benefit of PRT for improving lipids and lipoproteins in adults, they were based on the traditional underlying mean treatment effect and their corresponding 95% confidence intervals. However, this approach does not provide any information regarding how treatment effects from new individual trials are distributed around the summary effect (Higgins et al., 2009). Consequently, this may be problematic since the treatment effect is assumed to vary between studies when a random-effects model is used. Since the time that our work was completed, a prediction intervals (PIs) approach has been developed and recommended for determining how treatment effects from new individual trials are distributed about the mean in a random-effects meta-analysis (Higgins et al., 2009). The calculation of PI is an approximation of the expected treatment effect in a new trial and is probably more practically relevant (Higgins et al., 2009).
Given that (1) our previous meta-analytic work was based on random-effects models (Kelley and Kelley, 2009), (2) the use of PIs for meta-analysis was not well-established at the time, and (3) the possibility that PIs may be the most appropriate and comprehensive statistical inferences to be derived from random-effects meta-analyses (Higgins et al., 2009), the purpose of this brief article was to calculate PIs for a random mean effect in a new (future) study using data from our previously published research dealing with the effects of PRT training on TC, HDL-C, TC/HDL-C, non-HDL-C, LDL-C, and TG in adult humans (Kelley and Kelley, 2009). To the best of our knowledge, this is the first time that PIs have been used in an exercise meta-analysis.
Section snippets
Data source
We used data from our previous meta-analytic work dealing with the effects of PRT on lipids and lipoproteins in adults, which have been described in detail elsewhere (Kelley and Kelley, 2009). Briefly, randomized controlled trials > 4 weeks dealing with the effects of PRT on lipids and lipoproteins in adult humans > 18 years of age and published between January 1, 1955, and July 12, 2007, were included. Primary outcomes included TC, HDL-C, TC/HDL-C, non-HDL-C, LDL-C, and TG. Treatment effects for
Results
The results for underlying mean treatment effects, their 95% confidence intervals, as well as 95% PIs for changes in lipids and lipoproteins are shown in Table 1. As can be seen, our random-effects meta-analysis resulted in statistically significant improvements, on average, for TC, TC/HDL-C, non-HDL-C, LDL-C, and TG, but not HDL-C. In contrast, the approximate PIs for the true effect in a new study crossed zero (0) for all lipid and lipoprotein outcomes.
Discussion
The purpose of this brief article was to calculate PIs from our previously published research dealing with the effects of PRT on TC, HDL-C, TC/HDL-C, non-HDL-C, LDL-C, and TG in adult humans (Kelley and Kelley, 2009). Our current analysis resulted in PIs for the true effects in a new study overlapping zero for all outcomes. From a practical standpoint, the results derived from PIs may be more practically relevant than those derived from confidence intervals. Consequently, caution may be
Conflict of interest statement
The authors declare that there are no conflicts of interest.
Acknowledgments
This brief report was funded by Grant-in-Aid 0755207B from the American Heart Association (G.A. Kelley, Principal Investigator).
The authors would like to thank Gerry Hobbs, PhD, Department of Statistics, West Virginia University, for his assistance in the preparation of this article.
References (8)
- et al.
Impact of progressive resistance training on lipids and lipoproteins in adults: a meta-analysis of randomized controlled trials
Prev. Med.
(2009) - et al.
Metan: fixed- and random-effects meta-analysis
Stat. J.
(2008) - et al.
A re-evaluation of random-effects meta-analysis
J. R. Stat. Soc., Ser. A Stat. Soc.
(2009) - et al.
Fixed effects vs. random effects meta-analysis models: implications for cumulative research knowledge
Int. J. Sel. Assess.
(2000)