Overall cancer incidence in the U.S. has been falling over the past several decades [12]. This has been attributed largely to the reduction in smoking prevalence, especially among men. We were interested in investigating the role of changes in smoking habits on the incidence of cancer, and in particular we wanted to answer the hypothetical question: “Suppose smoking cessation were completely effective and smoking was entirely eliminated as a cause of cancer. How much cancer would still remain?” We focused therefore on cancer incidence at 12 sites known to be linked to smoking; cancer risk at other sites would not be affected by changes in smoking, and so we excluded them from our analyses.
We draw three main conclusions: First, by simulating the impact of the elimination of smoking on cancer rates, we estimated that about 60% of the cancers at tumor sites considered to be smoking-related would still occur (60% for males, 63% for females) in the absence of smoking. Second, under the same simulated condition of no smoking, the 11-year trend in the incidence rate of these 12 types of cancer would have been stable – neither rising nor falling – suggesting that smoking cessation is almost entirely responsible for the reductions in incidence that have occurred over this time period. Third, the degree to which smoking elimination would reduce cancer incidence rates would not be uniformly distributed across the country. We predict that some counties would have only modest reductions in cancer from eliminating smoking and by extension higher rates of cancers remaining. Future research might prioritize identifying additional factors explaining these elevated rates. Decision-makers should consider the potential contribution of other cancer risk factors in addition to smoking in designing comprehensive cancer prevention programs.
We compared our results to other estimates of tobacco’s contribution to cancer incidence. Islami and colleagues at the American Cancer Society recently published an extensive review of the potential for cancer prevention from addressing modifiable risk factors, including smoking [1]. Using an entirely different method, based on a synthesis of published estimates of the relative risks for different exposures, they concluded that in 2014, smoking was responsible for 19% of all incident cancers in the U.S. Parkin and colleagues [13] arrived at the same figure for the U.K. in 2010. Our estimate of 2016 data for the 40.9% of cancers that occur at the 12 smoking-related sites agrees very well with this: a reduction of 39.8% for 40.9% of all cancers translates to smoking being responsible for 16.3% of all cancers, not very different from their estimate of 19%. This gives us some confidence that our method, based on patterns in county-level data from the SEER registries, provided reasonable estimates for making national inferences.
Islami and colleagues also estimated the proportions of cancers in 2014 attributable to other lifestyle factors [1]. After smoking, the most important were excess body weight (7.8%), alcohol (5.6%) and physical inactivity (2.9%). These relatively minor contributions support our choice to focus this paper on smoking as the single most important behavioral risk factor for cancer.
The role of tobacco control in the decline in cancer rates has been widely recognized [14]. Thun and colleagues wrote: “The most striking success in [cancer] primary prevention is undoubtedly tobacco, where falling consumption has resulted in marked reductions in the incidence and death rates from … tobacco-related cancers among men …” [15]. Our analyses did not include cancers at the other sites, un-related to smoking, but Han and colleagues did [16]. Over a longer time period, 1975–2004, they found that for cancers unrelated to tobacco (or cancer screening which complicates the picture for several sites such as prostate and breast), cancer incidence increased in both men (0.88% annually for whites, 0.12% annually for blacks) and women (0.69% annually for whites, 0.52% annually for blacks).
Without detracting from the importance of tobacco control, these findings suggest that progress in the primary prevention of cancer must include many strategies, not only those focused on healthy lifestyles but also reduction of carcinogen exposures wherever possible [17, 18]. Such exposures include urban air pollution, particularly diesel exhaust, water pollution, pesticides in food, ionizing radiation and work-related chemicals [2]. Using the same SEER county cancer incidence data (2006–2010), Jagai and colleagues found a strong association between all-site cancer incidence and a measure of overall environmental quality, the Environmental Quality Index (which, developed by the U.S. EPA, combines measures of environmental quality in five domains: air, water, land, built environment and sociodemographic factors) [19].
As noted in the introduction, there are disagreements in the cancer prevention community about the importance of addressing environmental exposures [3]. Some argue that because the proportion of cancers caused by environmental exposures (often called the attributable fraction) is small relative to smoking and other health behaviors, control of the former does not warrant attention as a prevention strategy. There are several problems with this argument. First, cancers are known to have multiple causes and the very concept of apportioning different fractions of a disease rate to different single causes is misleading [20]. For a multifactorial disease, attributable fractions must add to more than 100% — probably much more, but calculations of attributable fractions for cancer causes often do not [1, 21].
A second major problem with the calculation of attributable fractions is that they measure the impact of the complete removal of the exposure. The complete elimination of smoking is probably impossible, at least in a democratic society. The most recent U.S. Preventive Services Task Force evaluation of the effectiveness of smoking cessation strategies found the best strategies have success rates of 28% or less [22]. Therefore, if we want to decide how much to prioritize tobacco control strategies, we should understand that 19% – Islami’s estimate of the percent of all cancers caused by smoking – is probably not a realistic goal for prevention. In our analyses, we chose to use the unrealistic assumption of smoking elimination because our intent was to place a lower bound on the proportion of the smoking-related cancers that would not be prevented by tobacco control. Also, we wanted to be able to compare our results to others, like Islami, who used a more conventional approach to estimating attributable fractions.
An important limitation of this study was the lack of individual level smoking data. Using lagged county-level smoking prevalence was an imprecise proxy for personal smoking data with individual information on amount smoked, duration, etc. Data availability prevented us from applying a 20-year lag to the model for the full 11-year period (Fig. 1), but the results for 2016 using the 10- versus 20-year lagged smoking data were very similar (35.2% versus 39.8%, respectively). We therefore doubt that the time trend in Fig. 1 would have been importantly changed if the longer lag could have been used. Despite the limitations in the smoking data, we were reassured that our estimate of the size of the reduction in cancer incidence assuming smoking elimination was in good agreement with others who have used entirely different methods to arrive at a similar result [1]. Another limitation is that smoking, whether measured individually or geographically, is often correlated with other cancer risk factors including air pollution, occupational exposures and dietary factors. We were unable to remove the potentially confounding or modifying effects of these factors on estimates of the effect of smoking elimination. This would likely lead to an overestimation of the reduction in cancers observed by simulating the elimination of smoking. A strength of the ecological assessment of smoking patterns is that it should capture variability in secondhand smoking as well as active smoking. Therefore, these estimates of the benefits of smoking cessation would also include cancer prevention from eliminating secondhand exposure.