点击显示 收起
Carolina Population Center, CB #8120 University Square, University of North Carolina at Chapel Hill, 123 West Franklin Street, Chapel Hill, NC 27516-3997, E-mail: jstookey{at}sph.unc.edu
Dear Sir:
Starling et al (1) concluded that protein intake had "virtually no effect in modifying the age-related decline in appendicular skeletal muscle mass." This conclusion may be mistaken, considering that the analyses presented by these authors were not evaluating effect modification. In keeping with the technical terminology, if the effect of age does not vary by different protein intakes, then protein intake could be said to "not modify" the effect of age. To evaluate modification of the effect of age, Starling et al (1) would have needed to test the significance of an interaction term between age and protein intake in their model. Although several regression models predicting appendicular skeletal muscle mass with age as the main exposure and different combinations of covariates were fit in this paper, no models appeared to include interaction terms.
In their use of the term modifying the authors may have intended to communicate that protein intake did not explain away the effect of age. Here, they verge on the epidemiologic concept of confounding, which is evaluated by comparing changes in the estimates of effect from nested models, not by tests of statistical significance (2, 3). To make a statement about confounding, the authors would have needed to compare the magnitudes of age effects estimated from several nested models. As a general rule in epidemiologic analyses, covariates that produce changes 10% in the estimate of interest are considered potentially important confounders and are retained in models to avoid confounding bias. In Table 3, which presents results from nested models, we see that the effect of age on appendicular skeletal muscle mass controlled for height and protein intake was -0.127, whereas the effect of age controlled for height alone was -0.102. Removing protein from the full model produced a 20% change in the estimated age effect. Given that protein intake is inversely related to age (4), and may determine muscle mass via age-independent mechanisms (57), this result suggests that protein intake confounds (or partly explains) the relation of interest. The fact that protein intake appears to confound the effect of age on muscle mass and that analyses failing to account for protein intake may produce biased results contradicts the above message that there was "virtually no effect in modifying...." Clarification of the above statement may be necessary to prevent confusion and to avoid detracting interest from a potentially important risk factor for sarcopenia.
REFERENCES