Research Interests
I. Cancer-related research interests
My recent research has focused on the relationship between population attributable fraction estimates for established breast cancer risk factors, and the goodness of fit and discriminatory accuracy of breast cancer risk prediction models. Because of the relatively low attributable fraction for the established risk factors, it is not uncommon to hear the statement that little is known about the causes of breast cancer, despite decades of research.


In my research I emphasize the distinction between the causes of cancer incidence and the causes of specific cancer cases. Much of my work is an attempt to demonstrate the reality of this distinction, even as public health scientists have moved into an era of complex statistical tools and genetic/molecular 'risk factors.' To date, I have done this mostly through the vehicle of breast cancer epidemiology. Breast cancer risk prediction models based on the established risk factors perform well in terms of predicting average risk, or incidence, in groups of women; i.e., breast cancer risk models 'fit' aggregate incidence data well, and make accurate predictions about expected numbers of cases in different groups of women. However, the established risk factors and risk models perform quite poorly in terms of individual-level discrimination; that is, the models and the risk factors contained therein say little about the causes of individual cases, about the futures of specific individuals.

It is straightforward to demonstrate that a risk factor (or a composite score based on a combination of risk factors) must be very strongly associated with disease if it is to serve as a worthwhile discriminator, or screening tool, at the individual level, that is, if it is to serve well at identifying the 'needles in the haystack,' those who will develop disease over a specified time period. Autosomal dominant mutations, such as mutations in BRCA1/BRCA2, are associated with much higher relative risks than many other breast cancer risk factors, and thus can serve relatively well in terms of identifying individuals at very high risk. However, the prevalence of such mutations is rare, and thus the vast majority of women are segmented into breast cancer risk categories on the basis of very weak discriminators. The current search for markers of genetic susceptibility to breast carcinogenesis is driven by the goal of improving discrimination at the individual level. However, to date, virtually all of the studies of common polymorphisms and their interactions with environmental exposures have produced relative risks that are of the same modest magnitude (i.e., <3.0) as the established reproductive/lifestyle risk factors.

In forthcoming research, I will explore the consequences of the failure of scientists to more openly discuss the paradoxical nature of risk prediction, that is, the failure to acknowledge the strong predictive ability of most risk factors/risk factor models at the population level coupled with their extremely poor predictive ability at the individual level. The language of individual risk implies an equating of the often modest predictors of incidence with the causes of cases, and this equating carries social and ethical implications. At the individual level, the language of individual risk may arouse unwarranted fear and anxiety. It can also lead to cynicism and a disregarding of health risk information when 'individual risk' prediction does not accord with observed reality in terms of who gets disease and who remains healthy. At the societal level, the privatization of risk may lead to disease prevention strategies that are ineffective because they fail to address the 'population' level nature of risk. These issues become increasingly important as the deterministic power of current genetic/genomic research is conveyed, sometimes misleadingly, by scientists and other opinion leaders.

Recent Accomplishments and Honors
I have recently completed a chapter for the forthcoming edition of the textbook Cancer Epidemiology (Schottenfeld and Fraumeni) on the issues discussed above.

Publications
1. Rockhill B, Willett WC, Hunter DJ, Manson JE, Hankinson SE, Spiegelman D, Colditz GA. Physical activity and breast cancer risk in a cohort of young women. J Natl Cancer Inst 1998;90:1155-1160.
2. Rockhill B, Willett WC, Hunter DJ, Manson JE, Hankinson SE, Colditz GA. A prospective study of recreational physical activity and breast cancer risk. Arch Intern Med 1999; 159: 2290-2296.
3. Rockhill B, Colditz GA, Rosner B. Bias in breast cancer analyses due to error in age at menopause. Am J Epidemiol 2000; 151: 404-408.
4. Rockhill B. The privatization of risk. Am J Public Health 2001; 91: 365-368.
5. Rockhill B, Spiegelman D, Byrne C, Hunter DJ, Colditz GA. Validation of the Gail et al. model of breast cancer risk: Implications for primary prevention. J Natl Cancer Inst 2001; 93: 358-366.
6. Rockhill B, Byrne C, Rosner B, Colditz GA. Log-incidence model of breast cancer risk: evaluation of accuracy. (In print, J Clin Epidemiol.)

E-mail: brockhil@email.unc.edu
Telephone: (919) 962-2756
FAX: (919) 966-2089
Address: Department of EpidemiologyMacGavran-GreenbergCB# 7435 Chapel Hill, NC

Click here to update this profile