Do leukocyte telomere length dynamics depend on baseline telomere length? An analysis that corrects for ‘regression to the mean’

Abstract

Leukocyte telomere length (LTL) shortens with age. Longitudinal studies have reported accelerated LTL attrition when baseline LTL is longer. However, the dependency of LTL attrition on baseline LTL might stem from a statistical artifact known as regression to the mean (RTM). To our knowledge no published study of LTL dynamics (LTL and its attrition rate) has corrected for this phenomenon. We illustrate the RTM effect using replicate LTL measurements, and show, using simulated data, how the RTM effect increases with a rise in stochastic measurement variation (representing LTL measurement error), resulting in spurious increasingly elevated dependencies of attrition on baseline values. In addition, we re-analyzed longitudinal LTL data collected from four study populations to test the hypothesis that LTL attrition depends on baseline LTL. We observed that the rate of LTL attrition was proportional to baseline LTL, but correction for the RTM effect reduced the slope of the relationship by 57 % when measurement error was low (coefficient of variation ~2 %). A modest but statistically significant effect remained however, indicating that high baseline LTL is associated with higher LTL attrition even when correcting for the RTM effect. Baseline LTL explained 1.3 % of the variation in LTL attrition, but this effect, which differed significantly between the study samples, appeared to be primarily attributable to the association in men (3.7 %).

Appendix

The change from the baseline measure X₁ to follow up measure X₂ is adjusted for the regression to the mean effect to yield a corrected value D as follows:

$$ D = \rho \left( {{\rm X}_{1} - \bar{\rm X}_{1} } \right) - \left( {{\rm X}_{2} - \bar{\rm X}_{2} } \right) $$

where

$$ \rho = \frac{{2rS_{1} S_{2} }}{{S_{1}^{2} + S_{2}^{2} }} $$

in which r is the correlation between X₁ and X₂.