Spurious association between telomere length reduction over time and baseline telomere length.

Ehrlenbach et al. found a linear relationship between baseline relative telomere length (RTL) and the decrease of RTL over 10 years (r1⁄4 0.674; P < 0.001). Similarly, others have reported on this relationship between the telomere attrition rate and the baseline telomere length and found largely similar associations. Likewise, when we correlated the RTL and delta RTL during 7 years of follow-up in 75 men from the Zutphen Elderly Study with an age range of 70–91 years, we found a largely identical association (r1⁄4 0.733; P < 0.001; EJ Giltay et al., unpublished results). We question whether this association is trivial. We used a random number generator to produce 510 baseline RTL values, similar to the number of pairs in the study of Ehrlenbach et al. A mean of 1.49 was aimed at for baseline values and 1.05 at 10-year follow-up, with distributions comparable to those presented in Table 1. Using these random numbers, we found a beta coefficient that was nearly similar to the beta coefficient that was presented in Table 2 (0.557 as compared with 0.589, respectively). Because baseline RTL (X) was used to calculate the RTL shortening rate (X–Y), the ‘dependent’ and ‘independent’ variables were functionally related. Pearson’s correlation coefficients using randomly generated factors can be estimated to be around 1/ˇ2, if baseline and outcome have equal variances. Therefore, it was to be expected that a linear regression model would best fit the data (as X was regressed on X–Y) and that an exceptional P-value of 2.3 10 90 was found (Table 2). We think that the slope of the regression line should have been tested against the slope of a no-effect line, instead of zero (i.e. a horizontal line). We think, therefore, that the reported association is explained neither by older cells having lower division rates nor by telomerase that acts preferentially on short telomeres as a special protection mechanism, as was suggested as potential explanations, but is merely a consequence of mathematical coupling. It seems more likely that the attrition rate of RTL is biologically independent of baseline RTL. Table 1 Hospital admissions for AMI (ICD-9: 410) in the Portuguese population during four international football competitions