Postmenopausal Hormone Therapy and the Risk of Ovarian Carcinoma
Source: Garg PP, et al. Obstet Gynecol 1998;92:472-479.
Garg and associates performed a meta-analysis, examining the relationship between postmenopausal hormone therapy and the risk of epithelial ovarian carcinoma. After identifying 327 published citations, seven articles representing 12 analyses of 21 individual studies were included in this meta-analysis. The final conclusion indicated a relative risk of 1.14 (CI = 1.04-1.24) for developing invasive or borderline ovarian cancer among ever-users of postmenopausal hormone therapy. The relationship of invasive ovarian cancer with increasing years of hormone therapy use was derived from data from six studies. Among women who used hormone therapy for more than 10 years, the final relative risk for the development of invasive cancer was 1.27 (CI = 1/00-1.61). Garg et al conclude that the use of postmenopausal hormone therapy, especially for more than 10 years, is associated with an increased risk of developing invasive epithelial ovarian carcinoma.
Comment by Leon Speroff, MD
In my opinion, this meta-analysis illustrates everything that is wrong with the application of this technique to observational studies. Meta-analysis is a technique developed to bring together small, randomized clinical trials in the effort to achieve greater statistical power. When applied to case-control cohort studies, it is subject to the same confounding biases that are present in individual studies. An overall increased risk of 14% in ever-users of hormone therapy in epidemiologic terms is slight and cannot be expected to reflect greater reliability when derived from case-control studies (as in this meta-analysis).
The conclusion that a 27% increased risk is associated with long-term use (more than 10 years) of hormone therapy is based upon six observational studies with data including long-term use. This conclusion, with a C.I. of 1.00-1.61, was by definition not statistically significant. Examining the individual conclusions of each of the six studies reveals that only one of these six reported significant increase with long-term use. This was a report from The Nurses’ Health Study in 1995 (Rodriguez, et al. Am J Epidemiol 1995;141:828-835) that found no significant increase in the relative risk of fatal ovarian cancer with the ever-use of postmenopausal hormone therapy. The link with long-term use achieved statistical significance with only 18 cases. Thus, the conclusion of this meta-analysis, regarding increased risk with increasing duration of use, is tenuous. Despite this, in the discussion of the meta-analysis, Garg et al imply that this is a risk that should be included in the clinician-patient dialog. I strongly disagree.
A paragraph in the discussion is an excellent example of epidemiologic thinking that is not helpful for clinicians and patients. In making an argument for estrogens in the etiology of ovarian cancer, they cite the publication in the New England Journal of Medicine (Rossing, et al. N Engl J Med 1994;331:771-776), which concluded there was a two- to three-fold increase in the risk of developing ovarian cancer with the long-term use of clomiphene. They don’t share with the reader the fact that this conclusion was based on five cases, and, not surprising, the confidence interval (1.5-82.3) was extremely wide, reflecting the imprecision of their conclusion because of the small number of cases. They further ignore subsequent studies that failed to confirm a link between clomiphene and ovarian cancer. It is precisely studies like the clomiphene study in which clinicians are justified in concluding that statistically significant epidemiologic studies with small numbers and imprecision probably have no clinical relevance.
In view of the controversy in the last few years regarding the appropriate use of meta-analysis, it is disappointing to me that our journals give credibility to conclusions of meta-analyses such as this one based upon weak observational data. I am repeatedly impressed that epidemiologists believe that all data can be presented to patients, allowing objective decision-making on the part of the patients. There are some studies that involve cancer that cannot be separated from the emotions that surround the fear of cancer, and when the studies are incredibly weak, it is better that they do not enter the clinical dialog. This meta-analysis is a prime example.