Letter to the editor, re “Bulky DNA adducts as risk indicators of lung cancer in a Danish case‐cohort study”

Abstract

I read with great interest the article “Bulky DNA adducts as risk indicators of lung cancer in a Danish case-cohort study” by Bak and colleagues in the recent issue of the International Journal of Cancer.1 However, I was concerned by their interpretation of the results after adjustment for smoking behavior (duration, intensity and time since stopping).1 Adjustment for smoking behavior produced the one statistically significant finding, changing the IRR for high adduct levels among smokers from 1.45 (0.96–2.19) to 1.61 (1.04–2.49) after control for smoking behavior, approximately a 28% change in beta. In the Discussion the authors write “The interpretation of the unadjusted results would be the effect of DNA adducts on lung cancer risk, whatever caused or modified the adduct levels, whereas the interpretation of the smoking-adjusted results would be the effect on lung cancer risk of adduct levels caused by anything but smoking”. However, as has been shown in the molecular epidemiologic literature, and in classic work by Susser this interpretation of the adjusted results is incorrect.2, 3 Statistical control for an antecedent variable(s) such as smoking behavior does not provide information on the effect on lung cancer of other exposures that cause adducts. In fact, control for smoking behavior provides an estimate of the risk associated with high adduct levels after control for other pathways through which smoking behavior impacts lung cancer risk that are unaccounted for by the adduct measure. The IRR after adjustment for smoking behavior indicates that high adduct levels predict lung cancer incidence among smokers after controlling for other mechanisms through which smoking behavior may impact risk. In actuality the change in IRR after control for smoking behavior may be too modest to really interpret, but the fact that the IRR increases after control suggests that either smoking behavior is inversely associated with high adduct levels or that smoking behavior has a protective effect against lung cancer once the impact of adducts is accounted for. Obviously these interpretations are contrary to current thinking on the role of smoking in lung cancer development and adduct formation. The change in the IRR may simply represent sampling error or perhaps that smoking behavior modifies the effect of adducts and that the variables were inappropriately modeled as confounders. The last possibility is consistent with the apparent effect modification by smoking status shown in Tables 2 and 3. The work by Bak and colleagues provides further prospective data on whether elevated adduct levels are associated with later lung cancer development, and suggests that risk may vary by smoking status at the beginning of follow-up.4, 5 The work provides no information on the role of other exposures that may also contribute to adduct burden.