Postmenopausal Hormone Therapy and Breast Cancer
December 1, 1997
A team of epidemiologists invited all investigators who had previously studied the association of postmenopausal hormone use and the risk of breast cancer to submit their original data for a collaborative, combined reanalysis. This effort was supported by the Imperial Cancer Research Fund, a United Kingdom charity.
A total of 52,705 cases of breast cancer was available for comparison with 108,411 women without breast cancer. No statements could be made regarding type of estrogen or the effect of progestational agents. Information regarding what hormones were used was available in only 39% of the women, and only 12% of the hormone users used a combination of estrogen and progestin. This reanalysis reached the following conclusions:
• Ever users of postmenopausal hormones had an overall increased relative risk of breast cancer of 1.14.
• Current users for five or more years had a relative risk of 1.35 (CI = 1.21-1.49), and the risk increased with increasing duration of use.
• Current and recent users had evidence of having only local- ized disease (no metastatic disease), and ever users had less metastatic disease.
• There was no effect of a family history of breast cancer.
• There was no increase in relative risk in past users.
• The increase in relative risk in current and recent users was greatest in women with lower body weights.
(Collaborative Group on Hormonal Factors in Breast Cancer. Lancet 1997;350:1047-1059.)
COMMENT BY LEON SPEROFF, MD
The most compelling reason to believe that long-term use of postmenopausal estrogen increases the risk of breast cancer is the inherent biologic plausibility. Factors known to increase a woman’s exposure to estrogen are known to increase the risk of breast cancer (e.g., age of menarche and age of menopause). Indeed, in this publication, the authors make a point of demonstrating that the quantitative effect of their conclusion is similar to extending the age of menopause. According to their calculations, current and recent hormone use was associated with a 2.3% increase in breast cancer risk per year, and the effect of the age of menopause was equivalent to a 2.8% increase in risk per year of delay. Many clinicians are attracted by the logic in this comparison; however, the steady exposure to postmenopausal estrogen is not exactly the same as extended exposure to cyclic ovarian function.
There are several limitations and problems with this reanalysis. An accompanying editorial states that this reanalysis is based on nearly 160,000 women.1 For the sake of accuracy, it should be noted that the conclusions regarding postmenopausal women are based on 42% of the women, yielding for analysis 17,949 cases of breast cancer in postmenopausal women. But, that is still a relatively large number.
Of the 63 studies eligible for consideration, one refused participation, and 11 had data unable to be retrieved. This amounted to 10% of the data, and these studies did not indicate an increased risk of breast cancer. Did this contribute to an incorrect final estimate? If we are to accept the absence of this 10%, what are we to make of the fact that the conclusion regarding increasing risk and duration of use was heavily influenced by 15% of the hormone users?
A closer look at the analysis regarding duration of use indicates that only two categories reached statistical significance: the group using hormones 5-9 years and the group using hormones for 15 or more years. What does it mean that the group using hormones 10-14 years had no statistical evidence of an increase in risk (relative risk was 1.09)?
It has been my position that the lack of agreement, uniformity, and consistency in about 50 observational studies (case-control and cohort) indicates that the use of postmenopausal hormone therapy cannot be associated with a major effect on the risk of breast cancer; otherwise, we would have agreement among the studies. I like to compare this situation with three other conditions: the protection against ovarian cancer by oral contraceptives; the protection against cardiovascular disease by postmenopausal estrogen use; and the increase in lung cancer due to smoking.
We believe each of those three despite the lack of a single randomized clinical trial because all the studies say the same thing: an impressive agreement and uniformity among observational studies. I believe this new reanalysis supports my contention that the lack of agreement among the studies indicates either a small effect of estrogen use or the effect of biases that can only be eliminated by a large, randomized trial (such as the ongoing Women’s Health Initiative).
In my view, the strongest indication that the risk estimate in this reanalysis is subject to a bias is found in the conclusion that current and recent users had evidence only of localized disease. This is consistent with two influences: surveillance bias and hormone acceleration of tumors already present and, thus, detection at an early, less aggressive stage.
In the Nurses’ Health Study, postmenopausal hormone users had 14% more mammograms.2 In the editorial accompanying this reanalysis, it is noted that in the Group Health Cooperative of Puget Sound, there is a 28% greater rate of mammography use in hormone users. Besides mammography use, greater interaction with health care providers among hormone users would yield more breast examinations.
It is well recognized that breast cancers diagnosed in postmenopausal hormone users are less aggressive.3 Less aggressive tumors and better survival rates in hormone users argue in favor of earlier detection. In turn, this argues in favor of surveillance bias operating in observational studies. The increased risk confined to localized disease in current and recent users in this reanalysis is a strong indication that the slightly increased relative risk reflects a combination of surveillance bias and less aggressive tumors due to acceleration of small tumors already present.
If the conclusion of this reanalysis were correct, it would mean that there would be 135 cases per year of breast cancer in every 1000 women aged 50-80 (based upon a 10% risk in 50- to 80-year-old women) with 35 of the cases due to hormone therapy. This should lead to nine extra deaths per 1000 women per year. However, studies have indicated a decreased risk of breast cancer mortality in postmenopausal hormone users. For example, the American Cancer Society nine-year prospective follow-up documented a 16% reduced risk of fatal breast cancer.4 The mortality data also support the contention that the results of the reanalysis reflect tumor acceleration and surveillance bias.
Remember that a meta-analysis cannot correct for biases and errors. Is this reanalysis a meta-analysis? Technically speaking, I am not sure. However, the epidemiologists in the accompanying editorial label this report as a meta-analysis. If it is, it is yet another example of a meta-analysis magnifying problems and biases inherent in observational studies.
The editorial states that physicians and other health care providers now have the ethical responsibility "to advise women that accumulated epidemiologic data show an increased risk in those on hormone therapy for five or more years."
Then, in the next paragraph, the epidemiologists state that "the excess risk should be viewed as a possibility not as a certainty." We are in an age of informed consent, but the manner in which a patient is informed has a profound effect on the decision ultimately made. It is one thing to say long-term estrogen therapy may increase the risk of breast cancer and quite another thing to say we are concerned, but there is still no definitive evidence.
I believe it is appropriate to tell patients that the lack of agreement among more than 50 studies is good news, consistent with either a very small effect or the result of inherent biases. This reanalysis has not settled the issue, and we must wait about 10 years for the results of the Women’s Health Initiative to provide us with definitive data.
References
1. LaCroix AZ, Burke W. Lancet 1997;350:1042-1043.
2. Colditz G, et al. N Engl J Med 1995;332:1689-1693.
3. Bonnier P, et al. Obstet Gynecol 1995;85:11-17.
4. Willis DB, et al. Cancer Causes Control 1996;7: 449-457.
You have reached your article limit for the month. Subscribe now to access this article plus other member-only content.
- Award-winning Medical Content
- Latest Advances & Development in Medicine
- Unbiased Content