False alarm: postmenopausal hormone therapy and ovarian cancer

Abstract

Background In a follow-up study of 948 576 women, based on respective relative risk (RR) estimates of 1.23 and 1.20 for incident and fatal ovarian cancer among current users of postmenopausal hormone therapy (HT), the Million Women Study investigators have estimated that, since 1991, HT has resulted in 1300 additional cases and 1000 deaths.

Critique The association was almost entirely confined to hysterectomized women, some of whom would not have been at risk because their ovaries had been removed; the findings in that group were uninterpretable. Among non-hysterectomized women, the RR was 1.12 and compatible with chance. The response rate to a follow-up questionnaire was only 64%, and HT-exposed women who developed ovarian cancer may selectively have responded. The risk of ovarian cancer was no longer increased once women stopped using HT – an effect that was pathologically and clinically incompatible with causation. Symptoms of as yet undiagnosed ovarian cancer may have ‘caused’ HT use, rather than the reverse. The histological classification of the tumors was not centrally adjudicated. A meta-analysis of nine studies of current HT use, for which the aggregated RR was 1.28, was acknowledged by the investigators to be defective.

Conclusions Only the findings among non-hysterectomized women were to some limited extent interpretable and, among them, there was virtually no evidence to suggest that current HT use increases the risk of ovarian cancer. It follows that the estimated numbers of additional cases of incident and fatal ovarian cancer that were attributed to HT use were spurious, and arbitrary extrapolation back to 1991, which was many years before the Million Women Study, had no scientific rationale.

Key words:

• Postmenopausal Hormone Therapy
• Ovarian Cancer

Authors' response

V. Beral, J. Green, G. Reeves, on behalf of the Million Women Study Collaborators

Million Women Study Coordinating Centre, Cancer Research UK Epidemiology Unit, Oxford, UK

In a recent Lancet article, we reported that postmenopausal women using hormone therapy (HT) had a slightly increased incidence of ovarian cancer (relative risk, 1.20; 95% confidence interval (CI), 1.05–1.36; p = 0.0002) and that the excess risk disappeared soon after use ceased[[1]]. All analyses were restricted to women who, as far as could be ascertained, had intact ovaries. While it is possible, as Shapiro suggests[[2]], that some women who had had a hysterectomy might be unaware that they had also had their ovaries removed, this should not lead to a false-positive association between HT use and ovarian cancer. Women who have had their ovaries removed are unlikely to develop ovarian cancer, but are more likely to use HT than women who have had a hysterectomy[[3]]. Thus, if many hysterectomized women also had their ovaries removed (but were unaware of it), this would result in an artefactually lower incidence of ovarian cancer in hysterectomized HT users than non-users – but hysterectomized HT users had a significantly higher incidence[[1]].

Our findings are in broad agreement with other published results on HT and ovarian cancer. In studies other than our own, the combined relative risk for ovarian cancer in HT users compared to non-users is 1.42 (95% CI 1.28–1.56; p < 0.0001)[[1]]. The fact that the definition of HT use varies somewhat from study to study does not invalidate meta-analyses of published results, but means that much is still to be learnt by central pooling of the available evidence, so that similar definitions can be used across studies. An international collaboration to do so is now underway.

Use of data from epidemiological studies in defined populations to estimate the number of excess cancers in the general population attributable to a particular exposure is common in public health – this is the way that excess tobacco-attributable cancers are estimated, for example. It is, of course, important to use reliable information on exposure in the general population and, for HT use in the UK, such information is available only for the years 1991–2005[[4]]. Hence, the methods used to calculate attributable risk and the years covered were appropriate.

The reduction in ovarian cancer risk soon after HT use stops is similar to that observed for breast cancer. The apparently transient adverse effects of HT on both ovarian and breast cancer have important consequences – breast cancer incidence in the US declined rapidly soon after large numbers of women stopped using HT[[5]].