Editor’s note: With the recent lay media attention on the WHI study and the HRT controversy, Dr. Speroff has focused the main part of this issue on the study in question. OB/GYN Clinical Alert subscribers receive Dr. Speroff’s comment on the matter, as well as additional expert comment from Dr. Sarah Berga.
On may 31, 2002, the data and safety monitoring board (DSMB) made its 10th interim review of the data accumulated by the Women’s Health Initiative (WHI). The DSMB made 2 recommendations that were made public on July 9, 2002: 1) To discontinue the trial arm administering daily 0.625 mg conjugated estrogens combined with 2.5 mg medroxyprogesterone acetate or placebo; and 2) To continue the trial arm comparing daily unopposed estrogen (0.625 mg conjugated estrogens) with placebo in hysterectomized women. The combined estrogen/progestin arm was discontinued after an average of 5.2 years (range, 3.5-8.5 years) of follow-up because of an increase in invasive breast cancer trending toward, although not achieving, statistical significance. The conclusions highlighted by the WHI were derived from an unadjusted intent-to-treat analysis. The released results are shown in Tables 1, 2, and 3.
The WHI enrolled participants between 1993 and 1998 at more than 40 sites and was scheduled to end in 2005. The statistical parameters for benefit or harm were established in 1997 early in the study. When the increase in breast cancer exceeded the predetermined boundary, the DSMB was obligated to recommend discontinuation of this arm of the trial. The WHI concluded that this combination of estrogen/progestin should not be initiated or continued for the primary prevention of coronary heart disease, and that there is a substantial risk of breast cancer (Writing Group for the WHI Investigators. JAMA. 2002;288:321-333).
Comment by Leon Speroff, MD
The decision to terminate the combined E/P arm of the WHI was disappointing, disturbing, important, and necessary. The initiation and conduct of the first large clinical trial of postmenopausal hormone therapy generated high expectations, and I shared in the anticipation of solid, strong, and useful information. It is not without some trepidation that I challenge the image of WHI as an unflawed study. I would rather avoid placing myself in a position that can be viewed as defensive, but I cannot shirk my responsibility to formulate a response that hopefully will prove helpful for clinicians and patients in balancing the initial emotional reaction to the WHI.
The National Heart, Lung, and Blood Institute concluded in its press release that combined hormone therapy is unlikely to benefit the heart. In my view, the results do not justify a definitive conclusion. First of all, the WHI is heralded as a primary prevention clinical trial of postmenopausal healthy women. The average age of the participants is 63, and the age range is 50-79 (45% were in their 60s and 21% in their 70s). Although only 7.7% of the women reported cardiovascular disease upon entry, a significant number of the participants, because of their age, already had existing atherosclerosis, and we are increasingly aware that the beneficial effects of hormone therapy on the cardiovascular system are progressively diminished with increasing atherosclerosis.1
Does the increase in cardiovascular events in the treated group reflect an effect concentrated in older patients with significant atherosclerosis? The WHI answers this criticism by pointing out a lack of interaction with age; at a presentation of the WHI data, the investigator presented a graphic that indicated a similar difference between the treated and placebo groups in participants in their 50s, 60s, and 70s.2 However, the critical factor (results according to duration from menopause) has yet to be analyzed. Women with significant menopausal symptoms (especially hot flushing) were excluded from the WHI, which means that the numbers of women close to menopause had to be relatively small.
There remains, therefore, an important issue in regard to cardiovascular disease: this may not be a pure primary prevention trial. The WHI investigator in his presentation claimed that "unhealthy" women would have been balanced between the treated and placebo groups, but if the cardiovascular effect of estrogen is diminished in the presence of atherosclerosis, this would have an effect on the results.
The information provided from the WHI does not indicate the prevalence of new statin and aspirin use in the participants. A year-by-year analysis of coronary heart disease is of interest and importance. (See Table 4.)
It is easy to see that the coronary heart disease results are influenced by the events in year 5. What happened in year 5? Does this reflect new statin/aspirin treatment in the placebo group, lowering the event rate and providing a falsely high rate in the treated group? Did a similar experience take place in years 1 and 2? It is well recognized that the beneficial effects of statins occur rapidly, acting to stabilize plaques within a few months. Although statin use and aspirin use at baseline were comparable in the treated and placebo groups, no information is provided regarding new treatment during the follow-up. There is good evidence that the beneficial effect of estrogen on the cardiovascular system is lost in women already being treated with statins.3 Keep in mind that the cardiovascular events did not cross the pre-determined boundaries set by the WHI requiring cancellation of the study. With the small numbers involved, a shift of a few cases would have a major effect on the conclusion.
The WHI identified 400 women with established coronary heart disease upon entry. Among these women, the hazard risk for cardiac events was 1.28, a risk that did not reach statistical significance with a confidence interval of 0.64-2.56 (19 vs 16 events). When the remaining women were analyzed separately, the hazard risk was also 1.28, and again the confidence interval was not statistically significant (1.00-1.65; 145 vs 106 events). These numbers emphasize how small the observed cardiac effect was, and how easily a shift of a few cases could change the result. At a presentation of the WHI data, the investigator revealed that the released results were based upon diagnoses in the field, and that central adjudication of the cardiac diagnoses was revealing what seems to me as an important level of disagreement in 16% of the cases.2 It will be important to keep an eye on the final calculations.
When is the effect of a randomized, double-blind trial compromised by the clinical behavior of the patients? In the WHI, 42% of the treated group stopped their hormone therapy and 38% stopped medication in the placebo group. This drop out rate "exceeded design projections." Women in both groups began hormone treatment provided by their primary clinicians sometime after the study began, 6.2% in the treated group and 10.7% in the placebo group. This "drop in" rate was also higher than design projections. Also, 40.5% of the treated group (3444 women) and 6.8% of the placebo group were unblinded, mainly because of vaginal bleeding. When is intent-to-treat analysis inadequate in the face of unblinding drop outs, and drop ins, especially when duration of exposure is a critical factor?
Intention-to-treat analysis compares all individuals in the treated group with all in the placebo group, regardless of drop outs or drop ins. This is said to be the best method of analysis for clinical trials because it accurately reflects the randomization. One can’t help but wonder how the long-term benefit of a treatment can be assessed if subjects receiving treatment for only a short period of time are included. The WHI performed an "as treated" analysis, and this produced "more modest changes." The numbers and confidence intervals are not provided. A high drop out rate affects the numbers remaining and available for an as treated analysis.
For several years, I have argued that the lack of agreement, uniformity, and consistency among more than 60 case-control and cohort studies is a strong reason that the risk of breast cancer associated with hormone therapy cannot be a large one. The WHI results support that conclusion, amounting to a 26% increase, 8 cases per 10,000 women per year, and even this conclusion had a marginal level of confidence. I have further argued that those studies reporting an increase in risk could be reflecting hormonal acceleration of growth in pre-existing tumors. Is that a possibility in the WHI results? Here again, analysis by year is helpful. (See Table 5.)
It is apparent that the breast cancer results are heavily influenced by years 4 and 5. Remember that the growth of breast tumors is slow (it takes 10 years for a malignant cell to become clinically detectable at 1 cm diameter). The WHI breast cancer results are consistent with hormonal stimulation of pre-existing tumors. Notice that the hazard risk returned almost to 1.0 in year 6!
It is important to emphasize that a positive family history of breast cancer did not affect the results.
Case-control and cohort studies have uniformly observed a reduced risk of dying of breast cancer in women diagnosed during the use of hormone therapy. This is not only due to greater use of mammography, but it reflects lower grade and stage disease in hormone users, a finding that is consistent with accelerated growth of pre-existing tumors. In the WHI results, there were only 3 deaths due to breast cancer in the treated group and 2 in the placebo group. The follow-up was not long enough to provide the outcome of the breast cancers in the participants. The health of the participants is supposed to be monitored until 2005, so hopefully we will learn more about breast cancer mortality.
There is some good news. The reduction in osteoporotic fractures answers those who emphasize the lack of randomized trial data for the effect of estrogen on osteoporosis and fractures. The size of fracture reduction in the WHI is substantial because this population was at low risk for osteoporotic fractures (for example, women with previous fractures were excluded). The reduction in colorectal cancer is consistent with a uniform story in a large number of case-control studies. It is important to emphasize that the trial arm with unopposed estrogen is continuing because no increase in breast cancer has been recorded. Also keep in mind that the risk of venous thrombosis is concentrated in the first two years of use,4 and thus there is no reason to be concerned over this infrequent side effect in long-term users.
This will be an on-going story. We can expect periodic publications from the WHI as the data are analyzed in greater depth according to specific diseases and risk factors.
What Are We To Tell Patients?
Almost every patient will already have learned the facts, heavily and prominently reported by the media. One cannot deny the WHI results and their importance. They will change clinical practice, but I have tried to highlight some meaningful observations that will provide clinical perspective. It is appropriate to point out that the risk of breast cancer is small, and there is no major effect, but of course that is little consolation to patients. Remember that 97.5% of the participants in the WHI never experienced an adverse clinical event. In regard to coronary heart disease, I don’t believe we should discard a large body of biologic (including the monkey experiments in Tom Clarkson’s group) and epidemiologic evidence and make decisions solely based upon the WHI. I will continue to maintain that there is good reason to expect a beneficial cardiovascular effect in younger postmenopausal women without apparent atherosclerosis. Nevertheless, we should aggressively encourage women at high risk for cardiovascular disease to be treated with statins. There continue to be good reasons to expect beneficial effects of hormone therapy on menopausal symptoms, brain function, the skin, and the WHI provides strong support for a reduction in osteoporotic fractures and colorectal cancer. It should be emphasized that despite the reported increases in clinical events in the WHI, there was no difference in the death rates comparing the treated and placebo groups. Hopefully with time, a more objective and less emotional understanding of postmenopausal hormone therapy will be reached.
Are the WHI results limited to one kind of hormonal formulation? Of course, there is no way to know the answer at the present time. Clinicians and women may react to the WHI results by choosing other progestins, other doses, and other routes of hormonal administration. These are reasonable decisions, but we must be frank in our patient dialogues that there are little, if any, data to guide us.
The late Trudy Bush always argued that the objective of both basic and clinical science is to know the truth. And every epidemiologic study, no matter how good or how large, gives only one view of the truth.5 She always cautioned that it takes many views to come close to seeing the truth. The WHI is only one view of the truth. Contrary to the impressions reported in the media, the statistical calculations for coronary heart disease, stroke, and breast cancer are not overwhelming in their strength. The cardiovascular results may reflect new statin/aspirin use and the effect of hormone therapy on pre-existing breast tumors (a so-called promoter effect) may be the reason for the breast cancer results. I think it is appropriate to share with patients these alternative explanations for the WHI results.
Comment by Sarah L. Berga, MD
Do you have the facts on the recent analysis of the WHI trial? First, please look at the absolute increases in the key categories. Then look at the hazard ratios, which convey the excess risk in each arm. It is helpful here to refer to Figure 4 in the JAMA source document. The hazard ratio was calculated, as were the confidence intervals, for various outcome measures. Notice that there are two columns for the confidence intervals, unadjusted and adjusted. Please look at the column with the adjusted confidence intervals. How many of the ranges cross 1.0? The answer is all. This means that none of the findings met criteria for statistical significance. This is because there are so few cases. Usually when both the adjusted and unadjusted confidence intervals are presented, it is because there is some doubt about the unadjusted ones representing "reality," thus greater weight should be given to the adjusted confidence intervals.
Now, for sake of argument, let’s assume that there are actually statistically significant findings. Look at the column with the unadjusted confidence intervals and find the hazard ratio for dementia. Sorry. That is not one of the outcome variables. Why? Because one cannot answer the question as to whether "HRT" reduces the risk of dementia in a study conducted for only 5 years in women younger than 75. If fact, the global index did not include any measures relevant to the effect of sex steroids on brain function.
What do you make of the fact that only the Prempro arm was stopped and that the Premarin arm was allowed to continue? First, it means that the risk of breast cancer was not as high in the Premarin arm as in the Prempro arm. Second, it likely means that there was cardiovascular benefit in the Premarin arm. How much? We have to wait. In the meantime, the media and others would have you believe that all HRT is bad, even though the only arm that was stopped was the Prempro arm. Another, more logical, interpretation is that Provera counteracted the benefits of the Premarin upon the cardiovascular tree. The concurrent use of Provera in a continuous manner may even account for the small, but statistically nonsignificant, increase in the hazard ratio for breast cancer. Remember when it was widely posited that Provera did oppose the beneficial effects of Premarin? Maybe that hypothesis is true. We await the results of the Premarin arm.
Let’s get a little more theoretical. The study compared Premarin, Prempro, and placebo. Would you agree with the following interpretation? Premarin is an estrogen and Provera is a progestin, therefore all HRT formulations must be equally risky. In other words, Prempro is a good proxy for all continuous HRT preparations. I have written before about the many reasons that not all estrogens are the same. Certainly, not all progestins are the same. Therefore, although it is indeed tempting, in strict terms, one cannot generalize from this study to all HRT formulations. In other words, there is no scientific basis for collapsing across categories and using the WHI Prempro data to generalize to all other forms of continuous combined HRT—much less other regimens. How do estrogens differ? Let me count the ways: route of delivery, conjugated or not, bioidentical or not, and metabolites. The same holds for progestins. And not all women are the same. And not all tissues within a given woman are the same. Are women to be reduced to the sum of breast, bone, and cardiovascular tree? Hopefully not. Did all women get the same dose regardless of body weight and concurrent conditions? Yes. Would we really give a 100-pound woman the same dose as a 250-pound woman? Would we typically give a woman many years off HRT the same initial dose as a woman who is just having menopausal symptoms? It seems unlikely. Were serum levels of estrogen conjugates or unconjugated estrone measured to see if the women who suffered adverse events had higher or lower serum levels? No. In summary, no one study can arrive at truth and this study provides only partial information. The recent WHI results suggest that the use of Prempro may minimally increase some risks. But it does not suggest that all HRT preparations are dangerous or that long-term HRT is dangerous. The WHI did not address long-term risks.
There is a bright side. Remember the pill scare of the 1970s? As a result of the initial worrisome results, what happened? We got better, lower-dose pills.
My prediction is that the results of the WHI will lead us to the following practices:
- We will learn to individualize HRT dose and we will develop guidelines for individualizing intelligently;
- We will discover exactly how estrogen preparations differ from one another so that we can individualize;
- We will discover whether the transdermal route of delivery gives a better risk/benefit profile;
- We will determine, which, if any, progestins are safer than Provera. If they all are risky, then we will more widely use progestin IUDs for protecting the uterus or we will develop better intrauterine alternatives.
My predictions are based on the premise that the concept of ameliorating age-related disability has inherent merit and that to do this we need to refine our approach rather than abandon the attempt. In other words, it is the present set of compounds rather than the concept that is lacking. There is nothing wrong with wanting to age gracefully. Sex steroids are likely to play a role in doing so, but we need to refine how we give them in order to minimize risk and maximize benefit.
Dr. Speroff is Professor
of Obstetrics and Gynecology, Oregon Health Sciences University, Portland.
Dr. Berga is Professor and Director, Division of Reproductive Endocrinology and Infertility, University of Pittsburgh.
1. Mikkola TS, Clarkson TB. Cardiovasc Res. 2002;53: 605-619.
2. Oral Presentation. Rossouw J, Half Moon Bay, California, July 16, 2002.
3. Hodis HN, et al. Ann Intern Med. 2001;135:939-953.
4. Grady D, et al. Ann Intern Med. 2000;132:689-696.
5. Bush TL. Int J Fertil. 2001;46:56-59.