The results of this study suggest that modifying risk factors later in life could substantially reduce postmenopausal breast cancer incidence in the United States. Approximately 40 percent of postmenopausal cases of invasive breast cancer could be eliminated if all women ceased or did not initiate postmenopausal hormone use, restricted alcohol consumption to less than one drink per week, avoided weight gain of more than 5 kg after age 18 years, and exercised more than 5 hours per week.
PAR estimates are highly dependent on the cutpoints chosen to represent exposed status for each factor (7
). Thus, cutpoints should be carefully noted in interpreting and comparing PAR estimates. In this study, extreme cutpoints were used, such that the PARs likely represent an upper limit regarding the proportion of cases of breast cancer that could be eliminated by changing the risk factor distribution in the population. Whether weight gain and low physical activity could be altered by such a radical degree in the United States, as suggested by our cutpoints, is debatable. However, it is instructive to know the potential effect of public health efforts directed toward these ends.
Multiple studies to date have reported summary PARs for sets of risk factors not amenable to modification (4
). Other studies have included modifiable risk factors combined with nonmodifiable factors, without conducting separate analyses for modifiable versus nonmodifiable factors (3
). In the only previous study known to report a summary PAR associated with a set of modifiable factors, Mezzetti et al. (8
) found a PAR of 33 percent for low β-carotene intake, alcohol consumption of more than 20 g per day, and low physical activity. A summary PAR of 40 percent was also reported for low physical activity and overweight body mass index.
A number of studies have reported the PAR associated with individual modifiable risk factors, including diet (8
), body weight (8
), physical activity (8
), alcohol consumption (8
), and postmenopausal hormone use (6
). In the only previous study known to evaluate a PAR for weight gain, Eliassen et al. (16
) found a PAR of 15 percent for a gain of 2 kg or more since age 18 years. Clarke et al. (17
) and Mezzetti et al. (8
) estimated PARs of 15 percent and 12 percent, respectively, for physical inactivity. In our population, we found PARs of 21.3 percent for weight gain of more than 5 kg and 15.7 percent for 5 or fewer hours of weekly physical activity. Our estimate of a PAR of 6.1 percent for alcohol consumption is within the wide range of estimates (2–25 percent) previously obtained (8
We found a PAR of 4.6 percent for current postmenopausal hormone use. We are unaware of previous estimates for this effect of changing current hormone users to former users. Our estimate of 8.5 percent for converting all women to never users is similar to the PAR of 11 percent observed by Clarke et al. (17
) in a population of California women. Notably, both studies collected data prior to publication in 2002 of results from the Women's Health Initiative indicating that the risks of postmenopausal hormone therapy outweigh the benefits (24
). Since this time, use of postmenopausal hormones has declined rapidly in the United States (25
). Conversely, the prevalence of obesity continues to increase (28
), suggesting that the PAR for obesity may also increase.
Between-study variation in the PAR for a risk factor can often be attributed to dependence of the PAR on the prevalence of exposure to that factor in the population. Therefore, results in one population should be extrapolated to another population with caution, and with particular attention to the prevalence of risk factors of interest in each population. This study was limited to postmenopausal women; notably, the direction and magnitude of effect for certain breast cancer risk factors depend on menopausal status (1
). Participants were 95.6 percent non-Hispanic White and therefore our results may not be representative of those for the United States as a whole or for specific areas with more diverse populations.
Study design may also influence PAR estimates and account for between-study variation. As with any case-control study, the potential for selection bias, recall bias, and measurement error must be acknowledged. Any bias in obtaining our odds ratio estimates could influence the PAR estimates in either direction. Nondifferential exposure misclassification would bias odds ratio estimates toward the null, and Hsieh and Walter (31
) have shown that this effect would result in an underestimate of the true PAR.
The formula used here for the PAR assumes that there is a causal relation between the risk factors and the disease, that exposure to the risk factor(s) can be eliminated while other variables remain constant, and that the disease is rare such that odds ratios can be used as relative risk estimates. Notably, we focused on risk factors that have been consistently associated with breast cancer risk. However, causality for most of these risk factors has not been definitively established, and thus caution must be taken in interpreting these results.
Because the PAR is often misinterpreted (12
), its limitations must be clearly acknowledged. PARs associated with single risk factors cannot be summed to derive the total PAR attributable to all the factors, nor should the individual PARs for all risk factors sum to 100 percent (2
). Thus, the PAR for established risk factors cannot be subtracted from 100 percent to indicate the proportion of disease risk explained by a yet-to-be-identified set of factors.
The relevance of PARs to public health is minimal when the factors under consideration are not modifiable (12
). We focused on factors that are modifiable later in life, using variables representing current or recent lifestyle behaviors. Many of these factors, including weight change, physical activity, and alcohol consumption, are also modifiable earlier in life. Notably, recent behaviors may be correlated with earlier life behaviors that also influence breast cancer risk. For instance, the increased risk associated with recent physical inactivity may be partially due to inactivity earlier in life, and this increased risk may not be totally eliminated by initiating better exercise habits. We were able to investigate this issue by using data on lifetime physical activity habits. In models adjusted for recreational physical activity between age 22 years and menopause, the PAR for recent (5-year) recreational physical activity was reduced somewhat to 11.6 percent (95 percent confidence interval: −6.3, 26.6). The increased risk associated with other factors, such as hormone use and alcohol consumption, is thought to dissipate relatively quickly after cessation, although duration of use likely is important (33
). In this regard, the PAR attributed to these factors appears to be a reasonable estimate of the potential, immediate impact of lifestyle changes later in life.
Invasive breast cancer is a heterogeneous disease (35
). We did not have data regarding hormone status or other phenotypic markers and thus were unable to consider such heterogeneity. A number of risk factors, including parity, age at first birth, and obesity, are more strongly associated with estrogen receptor–positive than estrogen receptor–negative tumors (36
). Thus, it is possible that we underestimated the PAR of these factors for estrogen receptor–positive tumors and overestimated the PAR for estrogen receptor–negative tumors.
Finally, caution must be taken when risk factors for one disease may be protective factors for other diseases. For example, although alcohol consumption is a risk factor for breast cancer, consumption of moderate amounts is associated with a reduced risk of cardiovascular disease (37
). Thus, public health efforts must weigh these competing outcomes.
In summary, a substantial proportion of postmenopausal cases of breast cancer could be prevented by modifying known risk factors. While these findings provide a target for public health strategies, dramatic alterations in the lifestyle of the majority of the population would be required. These results provide further evidence that cancer incidence can most effectively be reduced by population-based strategies to shift the entire distribution of risk factors, rather than focusing on high-risk subgroups (39