. Author manuscript; available in PMC: 2010 Jan 26.

Published in final edited form as: Arch Intern Med. 2009 Jan 26;169(2):163–171. doi: 10.1001/archinternmed.2008.544

Table 2.

Summary of mediation analyses related to changes of body weight, physical fitness, and physical complaints. (Based on models without the interaction effect of treatment group by baseline HRQOL scores)

HRQOL Outcome of Interest	Putative Mediators	Spearman partial correlation coefficient				P-value from Sobel Test
		Regression parameter estimate ± standard error
		Path a Treatment effect on Mediator	Path b Effect of Mediator on HRQOL	Path c Treatment effect on HRQOL w/o mediation	Path c′ Treatment effect on HRQOL with mediation
PCS	weight change	−0.61^**	−0.17 ^**	0.23^**	0.08 ^**	<0.001
	weight change	−17.78 ± 0.46^**	−0.069 ± 0.0069^**	2.84 ± 0.21^**	1.62 ± 0.24^**
	fitness change	0.30 ^**	0.12 ^**		0.19 ^**	<0.001
	fitness change	0.76 ± 0.038^**	0.70 ± 0.084^**		2.31 ± 0.21^**
	physical complaint change	−0.14^**	−0.20 ^**		0.21 ^**	<0.001
	physical complaint change	−0.84 ± 0.091^**	−0.45 ± 0.034^**		2.47 ± 0.20^**
BDI-II	weight change	−0.62^**	0.07 ^**	−0.07^**	−0.01 ^ns	<0.001
	weight change	−17.78 ± 0.46^**	0.023 ± 0.004^**	−0.54 ± 0.12^**	−0.12 ± 0.14^ns
	fitness change	0.30 ^**	−0.07^**		−0.05^*	<0.001
	fitness change	0.76 ± 0.038^**	−0.27 ± 0.05^**		−0.33 ± 0.13^*
	physical complaint change	−0.14^**	0.13 ^**		−0.05^*	<0.001
	physical complaint change	−0.84 ± 0.091^**	0.21 ± 0.021^**		−0.36 ± 0.12^*

Table Note: Three statistical models were tested for each putative mediator: Model 1: Change in PCS (or BDI) = treatment arm + baseline PCS (or BDI score); Model 2: Change in mediator = treatment arm; Model 3: Change in PCS (or BDI) = treatment arm + baseline PCS (or BDI score) + Mediator. The designation of paths follows the illustration shown in Figure 5: Path a: ‘treatment arm’ effect from Model 2; Path b: ‘mediator’ effect from Model 3; Path c: ‘treatment arm’ effect from model 1. Path c ′: ‘treatment arm’ effect from Model 3. Three statistics are provided for all tests: Spearman partial correlation coefficient, regression parameter estimate, and standard error. Statistical significance of effects is shown by:

^**

p<0.001

p<0.02

^ns

p>0.05.

The last column summarizes the p-value from the Sobel test for the significance of the indirect effect ab, i.e., the amount of mediation. Results show that changes in weight, fitness, and physical complaints reduced the effects of treatment on PCS and BDI-II as indicated by the reduction in both regression parameter estimates from c to c ′ and Spearman partial correlation coefficients. In conclusion, changes in weight, fitness, and physical complaints partially mediated the effects of treatment on PCS and BDI-II, with one exception: changes in weight completely mediated treatment effects on BDI-II, as evidenced by the non-significant treatment effect under path c ′.