Abstract
BACKGROUND: Due to variability in patterns of consumption as well as well-known difficulties in obtaining valid self-report from heavy drinkers, quantifying the effects of heavy alcohol consumption on mortality is challenging. Using a DNA methylation biomarker of chronic heavy alcohol consumption (HAC) named the alcohol T-score (ATS), we previously showed that chronic HAC was a strong predictor of mortality. However, whether there is a similar effect when measures of shorter-term heavy alcohol use, i.e., recent "binge" drinking, were used to predict mortality was not examined. This is a critical issue because most biomarkers of HAC assess only short-term HAC. METHODS: Therefore, we examined the prediction of all-cause mortality from a DNA methylation biomarker of smoking (cg05575921), the ATS and a short-term biomarker of recent heavy alcohol use (cg07375256) in 708 subjects from the Prostate, Lung, Colorectal and Ovarian (PLCO) Cancer Screening Trial using Cox Proportional Hazards Regression Modeling. Models were compared using Akaike's Information Criterion. RESULTS: The ATS was the best single predictor among three-variable models that included controls for sex and age. Of the possible four variable models, the model consisting of age, sex, cg05575921 methylation and ATS best predicted mortality. The addition of cg07375256 methylation did not improve model performance. In sensitivity analyses using only participants who provided alcohol SR (n = 639), the importance of the ATS and cg05575921 was replicated. We also found that ATS values were higher among those who declined to provide self-report alcohol use, indicating that missing self-report data about alcohol intake are not missing at random, and sometimes reflects elevated alcohol consumption. Finally, cg05575921 methylation was strongly associated with ATS values but only weakly with alcohol SR and not at all with cg07375256 methylation. CONCLUSIONS: Accordingly, this study indicates a strong effect of chronic HAC, but not short-term HAC, on mortality, further highlights the limitations of self-reported alcohol use in the prediction of all-cause mortality and indicates the value of assessing HAC in addition to smoking.