Menopausal Estrogen and Estrogen-Progestin Replacement Therapy and Breast Cancer Risk

Abstract & commentary

Schairer and colleagues reported the results of a cohort study. The population base was 46,355 postmenopausal women followed as part of the Breast Cancer Detection Demonstration Project, a nationwide breast cancer screening program. Data from this study were previously analyzed. To increase statistical power, Schairer et al added subjects and then reanalyzed the data for this report. To increase the size of the study population, they relaxed the inclusion and exclusion criteria. Therefore, not all data points were available for all study subjects. For instance, if age of menopause was not known, it was estimated to have occurred at age 57. If a hysterectomy was done, but there was no information on ovarian status, the age of menopause was assigned as the age at the time of hysterectomy. If women were known to have used a nonoral hormonal preparation, they were excluded. In phase 1 of the study (original cohort), no information on hormone type was collected. In phase 2, information on breast cancer screening was not included. In phase 3, Schairer et al attempted to collect missing information by interview, thus making part of the study retrospective. Using the relaxed criteria, Schairer et al expanded the population to 46,355 women. In this cohort, there were 2082 breast cancers. Pathology reports were obtained for 82%. Nodal status was available for 86%. Tumor size was available for 71%. The mean age at the start of follow-up was 58 years and the mean duration was 10 years, but some were followed for only one year and others for 16 years. A majority (86%) of the population was white. Hormone use was classified as follows: none, 42%; ERT, 38%; PERT 4%; both ERT and PERT, 6%; uncertain, 10%; and progestin-only use, 1%. The most commonly used estrogen was conjugated equine estrogen and the primary progestin was medroxyprogesterone acetate. (Schairer C, et al. JAMA 2000;283:485-491.)

The main limitation of the study was the paucity of cases of breast cancer for each category. There were 1209 cases of breast cancer among ever-users of any type of hormone, 112 cases in whom hormone use was unknown, and 761 cancers among never-users. Despite the few number of cases, subanalyses were done.

For ERT users, the greatest number of cases of breast cancer (309) was in the category of more than 6 years use. The relative risk (RR) of breast cancer in this category was 1.1 and the confidence interval was 0.9-1.2. While a relative risk of 1.1 could mean a slight increase in risk because the confidence interval includes 1.0, the "increase" is not statistically significant. For all current users of ERT, the RR was also 1.1, but the confidence interval was 1.0-1.3, again indicating that the "increase" was not statistically significant.

There were only 77 cases of breast cancers among PERT users. When taken together, the RR of breast cancer was 1.4, with a confidence interval of 1.1-1.8. This is barely significant. Needless to say, with so few cases and only 10% of the population having ever taken PERT, subanalysis by duration of use yielded RR that were not statistically significant, but the trend was for the RR to fall with increasing duration of use, such that in the category of more than 6 years PERT use, the RR was 0.6 (decreased risk), with a confidence interval of 0.3-1.6.

Another subanalysis partitioned risk according to body mass index (BMI). Based on 72 cases who had used estrogen ³ 16 years and 82 cases who had used estrogen more than 8 and less than 16 years, Schairer et al conclude that there is an increase in risk in thin women (those in the bottom two quartiles, i.e., with a BMI £ 24.4 kg/m2). There was no increase in thin women with less than eight years of use and no increase in risk in women with BMI > 24.4 regardless of duration of ERT use. Although there were only 26 cancers in "thin" women who had used PERT ³ four years, Schairer et al interpreted a RR of 2.0 and a confidence interval of 1.3-3.0 as showing that there was an increased risk in lean, long-term PERT users, but not in heavier PERT users. Furthermore, the biological mechanisms that would make thinness a risk factor for hormone-induced breast cancer were not discussed.

COMMENT By Sarah L. Berga, MD

The day after this JAMA article appeared in my mail, the newspaper headlines read, "Progestin-estrogen use increases the risk of breast cancer more than estrogen use alone." The Wall Street Journal featured this on its front page. The implication is that both PERT and ERT increase the risk of breast cancer, that the increase is definitive and of a biologically relevant magnitude, and that hormone use is therefore dangerous. By the time I got to the office that day, I already had one phone call from a patient taking PERT who now wondered if she should have a hysterectomy.

The newspaper headlines were derived from the last line of the structured abstract that read, "Our data suggest that the estrogen-progestin regimen increases breast cancer risk beyond that associated with estrogen alone." At best, this sentence is a great example of overstating the study findings. On reflection, however, I think that the conclusion is simplistic to the point of being both wrong and irresponsible. Most newspaper reporters missed the word "suggest." That this would happen is entirely predictable. Now millions of women are needlessly anxious and many are deciding to quit taking hormones, vowing never to start, or signing up for an unnecessary hysterectomy. While there is an understandable tendency to prefer to report "positive findings ," we must demand better reporting from our medical journals. We must not allow ourselves and our patients to be victimized by misinformation. Sadly, the editorial that accompanied the source report did little to dispel the hype.1

To review, the limitations of the study are many. Certainly, there is ascertainment bias regarding hormone use, age of menopause, and disease status. Thus, subject classification may be in error. Adding subjects was done to increase the statistical power, but it may have compromised the validity of the data set. Schairer et al acknowledge in their discussion that the associations were smaller when subjects for whom age of menopause was unknown were removed from the analysis. They kept them in because excluding them would result in "a substantial loss of information." Does this mean that the borderline statistical significance of the subanalyses was lost altogether? Nonetheless, despite the increase in the study population size, the number of events is still too few to expect the subanalyses to yield meaningful conclusions.

Importantly, the overall analysis is limited to determining risk for incidence of breast cancer rather than mortality from breast cancer. As Schairer et al acknowledge in their discussion, mortality from breast cancer is lower in HRT users, although they cannot explain it based on earlier-stage disease or more favorable histology (see also Willis DB. Maturitas 1997;27:105-108). In addition, the study population included only those women who apparently took oral medications. While there is no reason to suspect that other HRT regimens will yield substantially different risks in a general population, the study results are specific to those using oral medication. Given the multiple limitations of the data set, I do not believe there was a basis for drawing firm conclusions. If there is an increase in risk, it is small and it would not translate into a risk for death from breast cancer. It is this last point that patients so desperately deserve to understand.

The accompanying editorial suggested that women consider omitting hormones and rely instead on a combination of statins to decrease the risk of cardiovascular disease and bisphosphonates to prevent osteoporosis. SERMs were also suggested as a second line of defense. However, none of these medications, alone or in combination, yields a comprehensive strategy for chemoprevention of aging. None of these will ameliorate urogenital atrophy. None of these has any significant short- or long-term benefits for the brain. None will relieve hot flashes and night sweats or buttress speed of processing. Consider that a statin and a bisphosphonate will not an undemented mind make! The benefits of estrogen use are multiple and so critical that it is truly a crime to simplistically conclude that ERT and PERT pose an important risk for breast cancer.

What if PERT but not ERT increases the risk of breast cancer? Progestins do seem to have untoward effects on the brain and the endothelium. Should we do more hysterectomies? The approach I would like to see become more widely available is a progestin-containing intrauterine device that confines the progestin exposure to the endometrium. Sadly, the development of this technology has received far more attention in Britain and Europe than here.

As a consequence of this article and its reiteration in the media, practitioners will have a lot of explaining to do. Sadly, much hand wringing could have been avoided by more responsible reporting. It is difficult to read all that one must. It is tempting to just read the last line of an abstract. Even though we all want a short bottom line, conclusions need to be more than overly reductionistic soundbites. Not every one of us has the time and skills to tease apart the data and draw independent conclusions. I hope this analysis helps those of you seeking a second opinion.


1. Willett WC, et al. JAMA 2000;283:534-535.

Which of the following is true?

a. Postmenopausal use of progestins causes breast cancer.

b. Postmenopausal use of progestins is contraindicated in women with a strong family history of breast cancer.

c. The present data suggest that if more women used PERT postmenopausally, then more would die of breast cancer.

d. If the relative risk is greater than 1.0 and the confidence interval ranges from 0.8 to 1.5, then there is a statistically significant increase in risk.

e. The conclusions of the present study as promulgated by the media are likely to make women using PERT anxious.