Controlling for baseline telomere length biases estimates of the effect of smoking on leukocyte telomere attrition in longitudinal studies

Preprint, associated R scripts and dataset.

Summary: In analyses of longitudinal leukocyte telomere length (LTL) data it is common practice to adjust for baseline LTL when modelling the factors that potentially affect LTL attrition. However, the apparent dependence of LTL attrition on baseline LTL is largely brought about by regression to the mean as a result of measurement error. We used simulated LTL data to explore whether adjusting for baseline LTL results in biased estimates of the true effect of smoking on attrition. We show that when baseline LTL is shorter in smokers than non-smokers and LTL measurement error is non-zero, adjusting for baseline LTL results in overestimating the true effect of smoking on telomere attrition. The size of this latter bias increases with increasing LTL measurement error. Since it is a robust finding that smokers have shorter LTL than non-smokers and LTL measurement error is substantial, we conclude that the type 1 error rate for reports of effects of smoking on telomere attrition is likely to be above 5%. We therefore recommend that models of LTL attrition should not be adjusted for baseline LTL. Although we have couched our analysis in terms of the effects of smoking, our findings are likely to have general relevance to other lifestyle factors and exposures studied in relation to telomere attrition.