|
Correction
to
Colditz et al.,
|
| |||||||||||||||||||||||||||||||||||
In an attempt to obtain an overview of the data, I have calculated the absolute risk of breast cancer on the basis of 1935 cases of cancer among 69,586 postmenopausal women, as reported by the authors. Simple arithmetic puts the absolute risk at 2.78 percent. Since the cohort includes both treatment and control groups, this figure is an overestimate of the risk for the controls. On the basis of the reported relative risk (1.71) among older women taking hormones, the absolute risk would then be 4.75 percent (95 percent confidence interval, 3.73 to 6.06 percent). For patient and doctor, a crucial decision based on a 71 percent increase in the relative risk of a disease is quite different from a decision based on a 1.97 percentage point increase in the absolute risk.
My numbers are necessarily imperfect for the reasons stated above, and I would welcome similar calculations by the authors based on their original data. Relative risks are all well and good, but for most patients, the bottom line is the overall likelihood of disease.
Thomas Brunoski, M.D.
4 Ivy Knoll
Westport, CT 06880
References
Lynda H. Powell, Ph.D.
Rush–Presbyterian–St. Luke's Medical Center
Chicago, IL 60612
To our knowledge, none of the analyses in the Nurses' Health Study have accounted for the racial or ethnic composition of the cohort.1,2,3,4 It is possible that the reported association between hormone-replacement therapy and breast cancer is confounded by race, since black women, as compared with white women, are both less likely to use hormone-replacement therapy and at lower risk for postmenopausal breast cancer. That is, a higher incidence of breast cancer among women using hormone-replacement therapy would be expected simply because the risk of postmenopausal breast cancer is higher among white women.
It is also possible that the elevated risk among the women using hormone-replacement therapy in the Nurses' Health Study reflects a lower-than-expected occurrence among nonusers rather than an increased risk among users. We compared the incidence of invasive breast cancer reported by the National Cancer Institute's Surveillance, Epidemiology, and End Results (SEER) program for white women in the United states (from 1986 through 1990) with the incidence reported in the Nurses' Health Study according to age.3 The incidence of breast cancer in the Nurses' Health Study was 0.8 times that expected (0.7 among the women who never used hormones, 0.7 among past users, and 1.0 among current users). An additional concern is the generalizability of risk estimates reported for long-term users, since nurses with very early menopause or bilateral oophorectomy (or both) may be overrepresented in this cohort.
Finally, Colditz et al. did not present any information on the relation between mortality from all causes and the use of hormone-replacement therapy. It is not implausible, given the cardioprotective effects of hormone-replacement therapy in the Nurses' Health Study, that mortality from all causes may be reduced by 20 percent or more among current users, even though the risk of breast cancer may be elevated. In our opinion, published analyses of the disease-specific effects of hormone-replacement therapy should be accompanied by information on mortality from all causes, since this information is necessary to help women and physicians make decisions about using hormone-replacement therapy.
Janine A. Blackman, M.S.
Trudy L. Bush, Ph.D., M.H.S.
University of Maryland School of Medicine
Baltimore, MD 21201
References
It is intriguing that the results from the early part of the follow-up differ somewhat from those obtained subsequently. The earliest report (after four years of follow-up, with no updating of exposure2) showed relative risks of 1.0 and 1.3 for current and past use of hormone-replacement therapy, respectively. The current article reports relative risks ranging from 1.14 to 1.46 for current use and from 0.90 to 1.03 for past use. The relative risk of coronary disease also appears to have changed. Early follow-up yielded a relative risk of 0.30 among current users and 0.59 among past users.3 After 10 years of data accrual, the estimated relative risk for current users had almost doubled, to 0.56, and the relative risk for past users had also increased, to 0.83.4 Note that the cumulative nature of the data tends to blur differences in results between early and subsequent periods.
The implications of observing simultaneous changes in clinical practice and drug effects (as assessed through nonexperimental methods) have been discussed previously.1 Such time trends suggest that nonexperimental quantification of drug effects may be confounded by changes in clinical practice. Perhaps a better insight into these issues would be obtained if the study investigators provided results stratified according to calendar time (e.g., two-year segments of data accrual) for the relative risks of both breast cancer and coronary disease. Nevertheless, the design feature relating exposure status to outcome within a two-year period, although probably relevant for coronary events, does not appear to be appropriate for studying breast cancer.
K.S. Joseph, M.D., Ph.D.
Royal Victoria Hospital
Montreal, QC H3A 1A1, Canada
References
Editor's note: Dr. Dupont has been a paid consultant to Wyeth–Ayerst Laboratories.
William D. Dupont, Ph.D.
Vanderbilt University School of Medicine
Nashville, TN 37232-2637
References
From experimental studies, it has been concluded that the period between the initial development of a breast tumor and its growth to 1 cm in diameter is approximately 16 to 20 years, with a mean of 18 years.1,2 Given this very long latent period, it is likely that a considerable but unknown proportion of the women had undetected breast cancer throughout the study period.
The authors did not state whether screening for breast cancer was the same for the women receiving hormone treatment and those not taking hormones. It may be that the detection rates were higher in the groups of women who used hormones because they were screened more diligently — for example, by mammography.
What the authors are reporting is an increased detection rate among the women treated with hormones, not an increased incidence of cancer. It is by no means certain that this is a bad thing, since breast cancer is a curable disease and the proportion of women who are cured is dependent on the stage of the cancer at the time of its detection.3 Provided the metastatic potential of breast cancer is not increased by hormone treatment and provided concern about the effects of hormone treatment results in improved screening, the potential risk may be lower than anticipated.
D.R. Wigg, M.B., B.S., M.D.
Royal Adelaide Hospital
Adelaide, SA 5000, Australia
References
Rather than discuss this absence of an association, the article by Colditz et al. deals almost exclusively with a minority of the women who have ever used hormones, categorized as current users. The authors report that the adjusted relative risk of breast cancer among current users who had taken hormones for five or more years (as compared with the nonusers) was 1.46. What hypothesis is accepted or rejected by these results? The authors' conclusion appears to be related more to statistical manipulation than to any known biologic determinant of breast cancer.
Avrum Z. Bluming, M.D.
University of Southern California
Los Angeles, CA 90033
References
To the Editor: Dr. Bluming asks what hypothesis we address and implies that there is no known biologic determinant of breast cancer in our analysis. However, estrogen is clearly a key determinant of risk. Breast cancer develops in women at much higher rates than in men. Women with earlier menopause have a lower risk of breast cancer than those with later menopause. Longer-term use of postmenopausal hormones increases the risk of breast cancer, an effect similar to that of a delay in menopause. The key question is not whether a woman who has ever used hormones has an increased risk of breast cancer, but rather how long a woman can use hormones before the risk is appreciably elevated and how rapidly any excess risk decreases after she stops using hormones. Pooling data from previous studies that included predominantly short-term use (typically less than two years) gives the false impression that all use is safe.
To respond to Ms. Blackman, Dr. Bush, and Dr. Powell, the relative risk associated with current use of postmenopausal hormones was similar among women with family histories of breast cancer and those without family histories and among those who consumed alcohol and those who did not. In sum, there was no important variation in relative risk for women with these important risk factors.
Dr. Brunoski asks about the absolute risk of breast cancer. It is small, although the disease is a major cause of death, in middle-aged women. However, the lifetime risk of breast cancer is substantial. For a 60-year-old woman who has never used hormones after menopause, for example, the risk of receiving a diagnosis of breast cancer during the next five years is 1.8 percent. If she has taken hormones for five years and continues to take them, the risk is 3 percent. Alternatively, for every 80 women taking hormones for five years from the age of 60 to 65, 1 will receive a diagnosis of breast cancer.
Ms. Blackman and Dr. Bush compare our incidence rates with the SEER data from 1986 through 1990. This comparison is inappropriate, because the incidence of breast cancer has risen significantly over time.1 A comparison of the incidence of breast cancer from 1976 to 1990 with the SEER data from 1976 shows that our observed number of cases is within 3 percent of the expected number.2 Furthermore, the cohort is more than 95 percent white, so race cannot materially confound this result.
Dr. Dupont asks for a survival analysis; such an analysis is possible but complex. Sampling by risk group is essentially equivalent to the analysis suggested by Dr. Dupont. We are currently undertaking an analysis of hormone use and total mortality.
As we noted, the rates of mammography did not differ substantially between hormone users and nonusers, so biased detection cannot explain the results, particularly the increased mortality due to breast cancer among women using hormones for more than five years.
During further analysis of the data on postmenopausal hormone use and screening histories, we identified a minor programming error. A total of 1411 women previously categorized as past users in 1982 were actually current users. With this correction, there is a 7 percent increase in person-years of follow-up among current users (with 33 additional cases of cancer in this category), and a 2.8 percent reduction in person-years of follow-up among past users (with 10 fewer cases in this category).
The analysis of the relation between the type of hormone and the risk of breast cancer is not affected by these corrections. Repeated analyses of the data presented in Table 2 and Figure 1 in our article show no material difference in the results; all previously significant results remain significant. In Figure 1, the relative risks associated with nonuse, current use for less than five years, and current use for five or more years are as follows: for women between 50 and 54 years of age: 1.0, 1.41 (95 percent confidence interval, 0.95 to 2.10), and 1.53 (95 percent confidence interval, 0.97 to 2.42), respectively; for those between 55 and 59: 1.0, 1.30 (95 percent confidence interval, 1.03 to 1.67), and 1.45 (95 percent confidence interval, 1.12 to 1.87), respectively; and for those between 60 and 64: 1.0, 1.06 (95 percent confidence interval, 0.75 to 1.51), and 1.63 (95 percent confidence interval, 1.28 to 2.07), respectively. The largest change is in the group of current users who took hormones for 60 to 119 months; the relative risk in this group changes from 1.46 (95 percent confidence interval, 1.22 to 1.74) to 1.36 (95 percent confidence interval, 1.15 to 1.61) (Table 1).
|
Graham A. Colditz, M.B., B.S.
Walter C. Willett, M.D.
Frank E. Speizer, M.D.
Brigham and Women's Hospital
Boston, MA 02115
References
| |||||||||||||||||||||||||||||||||||
This article has been cited by other articles:
HOME | SUBSCRIBE | SEARCH | CURRENT ISSUE | PAST ISSUES | COLLECTIONS | PRIVACY | TERMS OF USE | HELP | beta.nejm.org Comments and questions? Please contact us. The New England Journal of Medicine is owned, published, and copyrighted © 2009 Massachusetts Medical Society. All rights reserved. |
Previous
