Epigenetic Predictors of Smoking: External Validation and Association With Smoking‐Related Cancer Risk

ABSTRACT Tobacco smoking has a strong influence on DNA methylation (DNAm), enabling the development of DNAm‐based smoking predictors, yet external evaluations of these have remained limited. We aimed to assess six DNAm‐based smoking predictors in a large sample of middle‐aged and older Australians for (i) their accuracy and reliability, and (ii) their association with risk of smoking‐related cancers. We used blood DNAm data from the Melbourne Collaborative Cohort Study to calculate six DNAm‐based smoking predictors ( EpiSmokEr , Elliott‐183 , McCartney‐233 , DNAmPACKYEARS , PCPACKYEARS , and mCigarette ). Accuracy with respect to self‐reported smoking pack‐years ( N = 2875, median age: 61 years) was assessed using Pearson's correlations, explained variance, and area under the curve (AUC). Reliability was assessed using intraclass correlation coefficients (ICCs) in 364 duplicated samples. Conditional logistic regression was used to assess associations with lung ( N cases = 332) and urothelial ( N cases = 426) cancer risk in two nested case–control studies. All smoking predictors showed excellent reliability (ICC > 0.9) and ability to distinguish current from never smokers (AUCs: 0.94–0.98), explaining 35%–50% variance of smoking pack‐years. The predictors were positively associated with smoking‐related cancer risk, with strongest adjusted associations observed for mCigarette with lung cancer (per SD, rate ratio [RR] = 2.18, 95% CI: 1.57–3.02) and for PCPACKYEARS with urothelial cancer (RR = 1.62, 95% CI: 1.27–2.07). This translated into prediction gains compared with models solely based on questionnaire‐collected smoking variables. In middle‐aged and older Australians, DNAm‐based smoking predictors showed very good reliability and accuracy in predicting smoking exposure, and strong positive associations with smoking‐related cancer risk beyond self‐reported smoking information. This highlights their potential utility as an alternative or addition to self‐reported smoking information for cancer risk stratification.