Menopausal hormone use and ovarian cancer risk: individual participant meta-analysis of 52 epidemiological studies

Summary Background Half the epidemiological studies with information about menopausal hormone therapy and ovarian cancer risk remain unpublished, and some retrospective studies could have been biased by selective participation or recall. We aimed to assess with minimal bias the effects of hormone therapy on ovarian cancer risk. Methods Individual participant datasets from 52 epidemiological studies were analysed centrally. The principal analyses involved the prospective studies (with last hormone therapy use extrapolated forwards for up to 4 years). Sensitivity analyses included the retrospective studies. Adjusted Poisson regressions yielded relative risks (RRs) versus never-use. Findings During prospective follow-up, 12 110 postmenopausal women, 55% (6601) of whom had used hormone therapy, developed ovarian cancer. Among women last recorded as current users, risk was increased even with <5 years of use (RR 1·43, 95% CI 1·31–1·56; p<0·0001). Combining current-or-recent use (any duration, but stopped <5 years before diagnosis) resulted in an RR of 1·37 (95% CI 1·29–1·46; p<0·0001); this risk was similar in European and American prospective studies and for oestrogen-only and oestrogen-progestagen preparations, but differed across the four main tumour types (heterogeneity p<0·0001), being definitely increased only for the two most common types, serous (RR 1·53, 95% CI 1·40–1·66; p<0·0001) and endometrioid (1·42, 1·20–1·67; p<0·0001). Risk declined the longer ago use had ceased, although about 10 years after stopping long-duration hormone therapy use there was still an excess of serous or endometrioid tumours (RR 1·25, 95% CI 1·07–1·46, p=0·005). Interpretation The increased risk may well be largely or wholly causal; if it is, women who use hormone therapy for 5 years from around age 50 years have about one extra ovarian cancer per 1000 users and, if its prognosis is typical, about one extra ovarian cancer death per 1700 users. Funding Medical Research Council, Cancer Research UK.


Patterns of use of HT over time
Over the last few decades the prevalence of HT use has varied considerably over time. † Detailed statistics on long term trends in HT use are available for the USA and the UK.
In the USA HT use began to increase in the early 1970s but declined in the late 1970s following reports of increased risks of endometrial cancer associated with use of oestrogen-only preparations. HT use began to increase again in the late 1980s, continued to increase during the 1990s and halved abruptly in the early 2000s ( Figure A). Use stabilized during the 2010s, with an estimated 5 million users. The prevalence of HT use among controls from Canada in this meta-analysis appear broadly similar to those seen in the USA.
In the UK there was little use of HT until the late 1980s. HT use increased rapidly during the 1990s, halved abruptly in the early 2000s, and stabilized in the 2010s, with about 1 million users ( Figure B).
In western and northern Europe and Australasia the patterns of HT use are broadly similar to those seen in the UK.

Estimated person-years of HT use since 1970
Assuming that the average duration of HT use was 5 years (as found for controls in this meta-analysis) it is estimated that have been about 600 million woman-years of HT use in high income countries since the 1970s, about half in North America and half in Europe and Australasia.

Search strategy and eligibility criteria
This collaboration began in 1998, and since then potentially eligible epidemiological studies have been sought regularly by searches of review articles and computer-aided literature searches in MEDLINE and PubMed, using combinations of the search terms 'ovarian cancer risk', 'ovary cancer risk', 'hormon*', 'HRT', 'HT', and 'menopause'. To be eligible for these analyses, studies needed to have collected individual data on women's use of hormonal therapies for the menopause and on their parity and past history of oophorectomy and of hysterectomy. Studies completed after 2006 needed to have included a total of at least 200 women with ovarian cancer (not all postmenopausal). Studies completed before then were eligible with fewer than 200 cases. Studies that had collected relevant data, but had not published on ovarian cancer risk in relation to use of HT, were sought by correspondence with colleagues, by discussions at collaborators meetings (in 2000, 2005 and 2011), and by electronic searches using additional terms 'cohort', 'prospective', 'women' and 'cancer risk'.
By January 2013, 58 eligible studies 1-58 had been identified and principal investigators from each had been invited to participate in the collaboration. Data from 52 of the eligible studies are included in these analyses.  Studies not included in the analysis Data from six eligible studies [53][54][55][56][57][58] were not included in these analyses data. Three [53][54][55] had not published on the relationship between HT use and ovarian cancer risk.
All three eligible studies that had published results [56][57][58] and could not contribute data to this analysis were retrospective studies done in North America. Mills et al 56 studied 256 women with ovarian cancer and the adjusted relative risk in ever vs never users of HT was reported to be 1.39, 95%CI 1.01-1.93. Moorman et al 57 studied 364 postmenopausal women and the adjusted relative risk in ever vs never users of HT was reported to be 1.2, 95% CI 0.8-1.6. Rossing et al 58 reported results for a subgroup of women with ovarian cancer who had either never used HT or had exclusively used either estrogen-only, continuous estrogen-progestin or sequential estrogen-progestin preparations; no estimate for ever versus never use of HT in all women was published.
For the two studies 56,57 that had presented data on ever versus never use of HT the combined relative risk was 1.3 (95%CI 1.0-1.5). No study published estimates of the overall relative risk of ovarian cancer in current or recent ex-users versus never users, but Moorman et al 57 reported relative risks of 1.1 (0.7-1.5) and 1.6 (0.9-2.8), respectively, for current use and for past users who had ceased in the previous 5 years (ie, 1.

Data collection and definitions
Individual participant data contributed by principal investigators were checked and collated centrally so that analyses could use definitions that were as similar as possible across studies. Apparent inconsistencies in the data were rectified, where possible, by correspondence with the investigators. After the records had been checked and corrected, investigators were sent summary tables and listings of the variables to be used in analyses for final confirmation.
Due to Danish data protection laws, data from one prospective study, the Danish Sex Hormone Register Study (DaHoRS) 45 could not transfer individual exposure data abroad, and could be accessed only by on-line submission to Statistics Denmark. Principal investigators adapted the Danish dataset according to the specifications for this meta-analysis and provided tabular results. Analyses were restricted to women aged 55 years and older and adjustment variables were age, past hysterectomy, and parity. The tabulated results were used to impute variance-covariance matrices of relative risk (Greenland S, Longnecker MP. Methods for trend estimation from summarized dose-response data, with applications to meta-analysis. Am J Epidemiol 1992; 135: 1301-09) and results were combined seamlessly with the results from the other 51 studies, weighted by the inverse of the estimated variance.

Definition of HT use
Information sought from principal investigators about every woman's use of HT included: ever use, current use, age at first and last use, total duration of use, constituents of each preparation used and duration of use of each preparation. Based on the information provided, HT preparations were classified as those containing oestrogen only, oestrogen-progestagen, or other/unknown types. A few women were recorded as having used both oestrogen-only and oestrogen-progestagen preparations and were classified by the preparation last use (ie, only 54 of the cases who were current or recent ex-users were recorded as having changed from use of oestrogen-only to oestrogen-progestagen HT; and only 150 were recorded as having changed from oestrogen-progestagen to oestrogen-only HT). Limited information was available about the specific constituents of the oestrogen-only and oestrogen-progestagen preparations used. For example, only about two-fifths of those who were recorded as having used combined oestrogen-progestagen preparations had information recorded on whether the progestagens had been used every day or less often. In most analyses current users were combined with recent ex-users who stopped <5 years previously, ie, "current-or-recent users". Follow-up in prospective studies was censored 4 years after HT use was last recorded and so the only possible source of misclassification in such analyses is if never-users start or if ex-users restart within 4 years. Data from one large prospective study suggest that such changes in use were relatively uncommon, in that before 2003 only 1% of never users became current users each year and 4% of past users became current users each year 38 (and after 2003 even fewer never and past users started HT). Sensitivity analyses explored cutoffs other than 4 years (Appendix p17).

Classification of ovarian cancers
All but 3 studies 15

Statistical Methods
When more than two groups were compared, the variance of the log risk was estimated for each group (Plummer M. Improved estimates of floating absolute risk. Stat Med 2004; 23: 93-104) and these group-specific variances were used to calculate group-specific confidence intervals. This method yields valid comparisons between any two groups, even if neither is the baseline group. It allows the relative risk estimates to be treated as approximately independent in tests of heterogeneity and trend.
Estimates from stratified analyses were combined, weighted by the amount of statistical information in each stratum (inverse of the variance of log risk). Comparisons across different subgroups of women were made using standard chi-squared tests for heterogeneity, calculated from the change in log likelihood on adding extra terms.

Estimated absolute risk of ovarian cancer in HT users
As an example, we estimated the absolute excess risk of ovarian cancer associated with 5 years and 10 years use of HT, starting at age 50 for women in England.   and the relative risks found here in prospective studies (Figure 1) to estimate incidence rates in never users of HT (table below). National rates for ovarian cancer include a relatively small proportion of women who had had an oophorectomy in the denominator and, for comparability, the population at risk for these calculations also include women who had had an oophorectomy.
The relative risks for HT-associated incident and fatal ovarian cancer are similar. 23 HT use for 5 years, starting at age 50 in England, is estimated to result in one additional ovarian cancer in every 1000 HT users, and one additional death from the disease in every 1700 users; and use for 10 years is estimated to result in one additional ovarian cancer in every 600 HT users, and one additional death from the disease in every 800 users (table below).
Ovarian cancer incidence in other high income countries is similar to that in England (International Agency for Research on Cancer. http://www-dep.iarc.fr, accessed June 23, 2014), so the excess risks calculated here would be broadly similar for women in other high income countries.

Effect on the main findings of additional adjustment by various factors (not including DaHoRS 45 )
Relative risk (95%CI) in current-or-recent users vs never users of HT Stratified by study, centre within study, age, and body mass index and adjusted for parity, past use of oral contraceptives, age at menopause, and hysterectomy (as in the main analyses) 31 (1.21-1.41) As above, and with additional adjustment for: Year of birth  23 1.37 (1.29-1.46) †Relative risks were stratified by study, centre within study, age and body mass index, and adjusted for parity, use of oral contraceptives, age at menopause and hysterectomy.
Relative risk of ovarian cancer in current or recent ex-users vs never-users of HT, by study design age at first use of HT (not including DaHoRS 45 ) ) Heterogeneity: p=0.02 †Relative risk in current-or-recent versus never users of HT, stratified by study, centre within study, age and body mass index, and adjusted for parity, use of oral contraceptives, age at menopause and hysterectomy.