Estimating the Influence of Obesity on Cancer Risk: Stratification by Smoking Is Critical

Tobacco use is well known to be the most common cause of cancer in populations in which its use is widespread. More recently, excess bodyweight has been identified as an important cause of cancer incidence and mortality (Table 1). The population-attributable fraction (PAF) of cancer incidence, including all cancers in Table 1 except those in the prostate and liver, as a result of high body mass index (BMI) has been estimated to range from 4% to 6% in highincome countries and remain lower than 1% to 2% in moderateor low-income countries by using BMI estimates from 2002 and cancer incidence data in 2012. In a recent summary, Flegal et al presented results for the proportion of all cancer incidence attributable to excess body weight, and estimates ranged from20.2% to 8%. To achieve maximal generalizability, these estimates were based on the entire population, with smokers and nonsmokers combined. In contrast, when stratified by smoking, PAF estimates tend to be much higher for nonsmokers. For example, in the Cancer Prevention Study II, a prospective study led by the American Cancer Society in 1982, the PAF of total cancer mortality as a result of high BMI was 4.2% in men and 14.3% in women in the entire population, but these estimates increased to 14.2% and 19.8%, respectively, in the population that was limited to nonsmokers. This difference is not trivial, especially for men, where a 3.5-fold difference in the magnitude of PAF is observed when including or excluding smokers. Clearly, tobacco use has a profound influence in considering the association between body mass and cancer risk. Is it more appropriate to consider the PAF in the entire population or to stratify by smoking status? Some have argued that estimation on the basis of the entire population is more legitimate because it preserves generalizability. For example, Flegal et al argue that not all cancers are associated with obesity and, therefore, estimates on the basis of the entire population are most appropriate. In this commentary, however, we argue that from biologic, epidemiologic, clinical, and public health perspectives, effect estimates for excess body weight on cancer risk should be given separately for smokers and nonsmokers, and results given for the entire population are generalizable only in the sense that they are wrong for everyone.