Hormones And Heart Disease

PROBLEMS WITH TREATMENT GROUPS: The study initially included two treatment groups in addition to its placebo control: .625 mg Premarin alone and .625 mg Premarin plus 2.5 mg Provera. Three years after the study started, the results of another study (PEPI, 1996), indicated that the Premarin-alone regimen was not advisable. The 331 women in this group were then started on Provera and were added to the Premarin plus Provera group. The authors say this had no effect on the results, but they either did not do or did not report the statistical tests that would justify combining these groups. The pooling of these 331 subjects with the roughly 8000 Premarin+Progesterone subjects contaminated the sample. Considering the weakness of the findings (see below), this decision seems rather "fishy." I doubt seriously that the data were unaffected. Irrespective of the above problem, there are serious problems with the interpretation of the data. To understand these problems requires a bit of Statistics 101.

STATISTICAL BACKGROUND: When scientists test a hypothesis, they calculate a probability that their results were the result of chance. For instance, if a coin is flipped 10 times, it may come up "heads" 7 of the 10 times. Is there a reason for this, or is it just dumb luck? If a scientist calculates a probability of 5% or less that the results are pure chance, then he or she concludes there is a real effect. (For instance if a coin comes up "heads" consistently enough, then it's not just dumb luck. Maybe it's a trick coin.) Now assume that a scientist is testing a long list of hypotheses. The chance that at least one of his tests will show a false result (i.e. from dumb luck) are greatly increased. The more tests that are performed, the greater the likelihood of a false result. For this reason, scientists use probabilities much lower than 5%, called Bonferroni probabilities, to determine whether a result is attributable to dumb luck. If a scientist is testing 20 hypotheses in the same experiment, he or she will use a test probability of 0.25%, such that the sum of the "dumb luck" probabilities is the conventional 5%. This way, the scientist can attest that he or she is at least 95% certain of his or her results. If results are at least 95% certain, they are said to be "significant" (a very important buzz word in statistics).

GENERAL STATISTICAL PROBLEMS: In the WHI study, the investigators examined a total of 22 effects, some of which were independent from each other, and some of which were interrelated. These included several measures of coronary heart disease and one statistic for invasive breast cancer, which were considered "primary" outcomes and which were examined without the application of Bonferroni probabilities. There were three additional cancer statistics and four osteoporosis-related statistics that were considered "secondary" outcomes. A Bonferroni probability of 5% / 7 = 0.71% was applied to those statistics. There is actually no justification for one measure being more "primary" than another, and a more conservative Bonferroni probability of 5% / 22 = 0.23% should have been applied to the entire slate of 22 statistics. The authors actually show "nominal" (unadjusted) results and Bonferroni-adjusted (7-test) results for all 22 of their tests.

CORONARY HEART DISEASE CONCLUSIONS: 8.6% of the subjects were tested for LDL ("bad") cholesterol, HDL ("good") cholesterol, and triglycerides. The investigators showed that HRT decreases LDL cholesterol by 12.7%, increases HDL cholesterol by 7.3%, and increases triglycerides by 6.9%. Statistical tests either were not done or were not reported; however, these findings agree with those of the HERS and PEPI studies and are likely valid.

The HRT group does appear to have a 1.29 times greater incidence of heart attacks without the use of Bonferroni probabilities, but even then, the effect is statistically on the borderline between "real" and "chance." With the proper application of Bonferroni probabilities, the effect is quite nonsignificant ("chance"). The same is true of the supposed 1.41 times greater incidence of strokes, which is barely significant without the application of Bonferroni probabilities and becomes quite nonsignificant when the necessary Bonferroni adjustment is applied. At best, these results suggest that another study should be done to re-examine these effects with a tighter, more narrowly targeted experimental design. The only cardiovascular results that potentially hold water concern venous thromboembolic disease (blood clots), including deep vein thrombosis (usually in the legs) and pulmonary embolism (in the lungs). The results indicate approximately a two-fold increase in risk for the HRT group, and they are significant after application of Bonferroni probabilities adjusted for 7 tests. They might or might not still be significant with the correct Bonferroni adjustment for 22 tests. Taken as a whole, we already know that estrogen is a thrombogenic (blood clot promoting) agent, so it is not surprising to find that women on estrogen therapy have a higher incidence of blood clots, resulting in deep vein thromboses, pulmonary emboli, and ischemic events such as heart attacks. On the weight of this evidence, a prudent physician might advise his or her HRT patients to take blood thinning agents such as vitamin E or aspirin on a daily basis. Otherwise, this study (probably) replicates previous studies that show HRT is beneficial for cholesterol levels. On the whole, if the thrombogenic effects of HRT are prevented with blood thinners, the overall incidence of coronary heart disease may be decreased through HRT.

CANCER CONCLUSIONS: The HRT subjects had a 1.26 times greater incidence of breast cancer, a 0.83 time lower incidence of endometrial cancer, a 0.77 times lower incidence of colorectal cancer, and a 1.03 times greater risk of cancer overall. However, because the overall incidence of cancer was rather low, the results are all indistinguishable from chance. Only the invasive breast cancer and colorectal cancer rates bordered on statistical significance, but only when Bonferroni probabilities were not applied. An effect of HRT on breast cancer would not be surprising, since estrogen and progesterone both stimulate breast tissue. A prudent physician would probably recommend yearly mammography for his patients on long term HRT, would advise caution for those patients with a strong family history of breast cancer, and would not prescribe estrogen or progesterone for patients with breast cancer histories. Breast cancer is already a known and calculated risk of HRT.

BONE FRACTURE CONCLUSIONS: The HRT subjects had about 3/4 the number of bone fractures as non-HRT subjects. Although this effect was significant after the application of Bonferroni probabilities for 7 tests, it would not have been significant after the correct adjustment of probabilities for 22 tests. Even so, we already know from other studies that estrogen therapy prevents bone loss and therefore prevents bone breakage, so this finding is probably "real."

OVERALL CONCLUSIONS: Virtually every woman undergoing HRT has already considered the long-term risks and benefits of this treatment. We already know that: (1) estrogen therapy carries with it a risk for blood clots (2) both estrogen and progesterone therapy are likely risk factors for breast cancer, and (3) estrogen therapy reduces the risk of osteoporosis. This study offers very weak corroborative evidence to support these assumptions. There are additional benefits to HRT which are not addressed in the study, including an overall feeling of vitality and well being, psychological benefits, and prevention of dementia in old age. In my opinion, there is no reason for a healthy woman (e.g. no history of heart disease, clotting, or breast cancer) to avoid HRT, as long as the known or suspected risk factors are managed. Specifically, women should be watchful for early signs of breast cancer and should consider taking daily vitamin E or aspirin to prevent problems from blood clotting.

[Note from Becky: I do not recommend vitamin E to prevent blood clots. The data supporting aspirin are much better.]

Additionally, women should carefully consider which estrogens and progestins they use for HRT. Some carry greater risks than others. If all these risk factors are managed effectively, women probably benefit in many ways from HRT.

Sarah Fox, Ph.D.