Guest Editorial
Mammographic Density: A Breast Cancer Risk Factor or Diagnostic Indicator?1 Celia Byrne, PhD
In this issue of Academic Radiology, Heine and Malhotra (1,2) present an extensive review of the literature with regard to mammographic density and breast cancer risk, with particular emphasis on attributes that change. Mammographic density can be assessed in various ways (parenchymal patterns or measured percentage density) (3). Although the parenchymal patterns include both appearance and extent of the density in their definitions, the measured extent of density as a percentage of breast area has been shown to be a better discriminator of future risk with greater reproducibility (3,4). As discussed by Heine and Malhotra (1,2), mammographic density has been shown to be one of the strongest predictors of breast cancer risk (5). It has long been known that mammographic density varies on the basis of characteristics of the woman such as her age, body mass index, reproductive history, menopause status, and use of exogenous hormones (6). Furthermore, mammographic features also vary on the basis of characteristics of the test itself, such as the degree of compression, positioning, and image contrast (6). The authors described their goal to develop a comprehensive risk model to aid the mammographer’s decision process; the model would incorporate information on breast density, as well as other risk factors, with parametric statistics from multiresolution wavelet analyses, spectral characteristics, and fractal modeling (1). In the second part of the review, the authors expand this process further, recognizing the goal to understand and incorporate Acad Radiol 2002; 9:253–255 1 From the Channing Laboratory, Harvard Medical School and Brigham and Women’s Hospital, 181 Longwood Ave, Boston MA 02115. Received and accepted January 22, 2002. Supported in part by R29-CA 75016 from the National Institutes of Health, Bethesda, Md. Address correspondence to the author.
©
AUR, 2002
the features that impact on the changes in breast density during a woman’s life such as age, involution and breast development, and hormonal influences into the models (2). Any new computer features from digital mammograms should be evaluated both for the ability to aid in diagnosis (do they improve screening sensitivity and specificity?) but also in a prospective manner as predictors of risk. The multiresolution wavelet analysis, spectral characteristics, and fractal modeling may add information to breast cancer risk prediction beyond that of percentage density, but it can not necessarily be assumed that new features will independently predict risk to the same extent. Previous studies incorporating fractal dimension features have not been as informative regarding breast cancer risk as the crude measure of percentage breast density (7). Clinically, mammographic information is used to evaluate the immediate determination if there is a need for further work-up (additional imaging or biopsy). Informally, some radiologists already incorporate a woman’s risk information into their interpretations (age, exogenous hormone use, family history, etc). Radiologists also routinely look for changes from past images as an indicator of a potential problem. Although the risk factor information, as well as a mention of breast density, may be included in the mammogram report, this information has not been formally incorporated in a uniform manner into the diagnostic process. Heine and Malhotra (1,2) should be commended for their efforts to merge the information on determining breast cancer risk that has been presented in the epidemiologic literature with the current goals of the radiologic field to develop better tools for computerassisted diagnosis (CAD). Mammographic information can also play two separate roles for a woman. A woman undergoing mammography wants to know, “Does my mammogram indicate I might have breast cancer?” (She is asking if her current probability is zero or one.) Only when there is a negative re-
253
BYRNE
sponse (probability of zero) to that first question would a woman be interested in knowing how her mammogram might provide additional information regarding her future risk of developing breast cancer (range of future probabilities between zero and one). Although the same information may influence both roles (diagnosis and risk prediction), there are a number of distinctions between the two concepts. Primarily there is a difference in the relevant time frame. For diagnosis, the time is instantaneous (does the woman have breast cancer now?), while for risk prediction a time frame must be specified, such as 5-year risk or life-time risk (what is the woman’s probability of developing breast cancer during her lifetime?). When they stated that it was most likely yesterday’s density that puts a woman at risk today, Heine and Malhotra (1,2) recognized that the influence of breast density on predicting the probability of having breast cancer may not be the same as on predicting the probability of developing breast cancer in the future (1). Throughout their review, however, the concepts of risk (probability of developing cancer) and that of diagnosis of an existing cancer are intertwined. It may be possible to integrate these two concepts, but they must not be mistaken for being one and the same without a valid rationale or indication. There is good evidence that increased breast density is associated both with increased risk of breast cancer and with masking the detection of a current cancer. Although these multidimensional aspects appear to be in conflict, both influences of breast density would need to be incorporated into a diagnostic model. If the desired outcome of an analysis is a diagnosis (CAD), adding the “risk-profile” information to the mammographic information will be worthwhile only if recommendations or follow-up will differ given the same mammographic findings. Adding the risk-factor profile is essentially changing the “prior” probability of disease. Thus, for a woman at high risk due to her “risk profile,” it may take only a small indication to make her mammogram “suspicious,” while the same small indication may not be considered “suspicious” for a woman with a “lowrisk” profile. Thus the same “small indication” may result in the “high-risk” woman undergoing repeat mammography or a biopsy and the “low-risk” woman continuing with annual screening. If such a scenario is not plausible, then how such information will be used by a radiologist must be considered. The addition of breast density to existing (8 –10) or new breast cancer risk prediction models is likely to improve these models. The addition of breast density infor-
254
Academic Radiology, Vol 9, No 3, March 2002
mation to the Gail model has been demonstrated on a small scale to improve breast cancer risk prediction (11). However, one issue of concern is that these models have only modest discriminatory accuracy (12). Dense breasts are quite common (5). Therefore, although the proportion of women with breast cancer who have increased density is high, the reverse is not true (the proportion of women with dense breasts who have breast cancer is still low). Thus, although the relative risk and the attributable risk associated with increased breast density are high, the discriminatory accuracy may still be modest. Most previous studies of mammographic density and breast cancer risk have been limited by the use of one mammogram. Despite the limitations of having information from only one mammogram, the crude measure of percentage breast density has been shown to be one of the strongest predictors of breast cancer over 10 years past the date of the mammogram (5). It would seem reasonable that first attempts should be made to improve existing models of predicting breast cancer risk with information from one mammogram. In adding mammographic density to risk prediction models, one must also consider how other factors such as body mass index, reproductive history, and menopause status are related. Heine and Malhotra (2) point out that much more information would be available from serial images. As those authors describe, however, other aspects of a woman’s history that influence the mammographic appearance and possibly her breast cancer risk also change with time. Thus, the models with serial mammographic information may be complex and require extensive assumptions beyond our knowledge of the biologic relation between many of these parameters. Given current computer capacity, these models are easily created, but their interpretation may not be as readily derived. Incorporation of serial image information will certainly be the direction for future research, but there is still a lot to be learned from incorporating the information from one mammogram into risk modeling. Heine and Malhotra (2) also point out that the data libraries that can be created from serial digital mammographic screening have the potential to be great resources for future studies to understand better the temporal parameters that influence breast density and potentially breast cancer risk. These authors have recognized the vast number of risk factors that one would need information about at the time of each mammogram to be able to model serial changes and their influence on breast cancer risk. To be of any use, this risk factor information must be collected in a validated, uniform, and consistent manner.
Academic Radiology, Vol 9, No 3, March 2002
In summary, epidemiologic risk factor data can easily be incorporated into CAD systems as suggested (1,2), just as measures from “digital mammography” can easily be incorporated into multivariate epidemiologic models. Of most importance is how these models are developed and what assumptions are made with each modeling technique. Most likely, breast density information can improve the ability to both diagnose breast cancer (risk of having breast cancer) and determine future risk of developing breast cancer. However, it is important that the researchers creating these models keep in mind the distinction between each of these applications. As indicated by Heine and Malhotra (1,2), this is an exciting field of research that has the potential for providing great insight into the etiology of breast cancer, as well as the translational benefits for diagnosis. REFERENCES 1. Heine JJ, Malhotra P. Mammographic tissue, breast cancer risk, serial image analysis, and digital mammography. I. Tissue and related risk factors. Acad Radiol 2002; 9:298 –316. 2. Heine JJ, Malhotra P. Mammographic tissue, breast cancer risk, serial image analysis, and digital mammography. II. Serial breast tissue change and related temporal influences. Acad Radiol 2002; 9:317–335.
GUEST EDITORIAL
3. Byrne C. Mammographic density and breast cancer risk: the evolution of assessment techniques and implications for understanding breast cancer. Semin Breast Dis 1999; 2:301–314. 4. Boyd NF, Lockwood GA, Byng JW, et al. Mammographic densities and breast cancer risk. Cancer Epidemiol Biomarkers Prev 1998; 7:1133–1144. 5. Byrne C, Schairer C, Wolfe J, et al. Mammographic features and breast cancer risk: effects with time, age and menopause status. J Natl Cancer Inst 1995; 87:1622–1629. 6. Saftlas AF, Szklo M. Mammographic parenchymal patterns and breast cancer risk. Epidemiol Rev 1987; 9:146 –174. 7. Byng JW, Boyd NF, Fishell E, et al. Automated analysis of mammographic densities. Phys Med Biol 1996; 41:909 –923. 8. Gail MH, Brinton LA, Byar DP, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. J Natl Cancer Inst 1989; 81:1879 –1886. 9. Pike MC, Krailo MD, Henderson BE, Casagrande JT, Hoel DG. Hormonal risk factors, breast tissue age and the age-incidence of breast cancer. Nature 1983; 303:767–770. 10. Rosner B, Colditz G. Nurses’ Health Study: log-incidence mathematical model of breast cancer incidence. J Natl Cancer Inst 1996; 88:359 –364. 11. Benichou J, Byrne C, Gail M. An approach to estimating exposurespecific rates of breast cancer from a two-stage case-control study within a cohort. Stat Med 1997; 16:133–151. 12. Rockhill B, Spielgelman D, Byrne C, Hunter DJ, Colditz GA. Validation of the Gail et al model of breast cancer risk prediction and implications for chemoprevention. J Natl Cancer Inst 2001; 93:358 –366.
255