• E-mail
• Print
• Comment
• Font Size
• Digg
• del.icio.us
• Discuss article

The Million Women Study - is It Believable?

Posted on: Sunday, 18 September 2005, 03:00 CDT

The risk estimates of breast cancer associated with hormone replacement therapy (HRT) found in the Million Women Study were amongst the highest reported in the literature. Nevertheless, the press release which accompanied its publication1 in August 2003 was uncompromising and made no attempt to place the results in the perspective of a large volume of earlier work. Amongst other claims, it stated that 'For every 1000 postmenopausal women who begin 10 years of HRT use at the age of 50, there will be five extra cases of breast cancer amongst users of oestrogen-only HRT and 19 amongst users of oestrogen-progestagen combinations. So combined HRT causes four times as many extra cancers as oestrogen only. The study also found that current users have a 22% increased risk of death from breast cancer. . .'. The principal investigator (Professor Valerie Beral) was quoted as saying that 'We estimate that over the past decade use of HRT by UK women aged 50-64 has resulted in an extra 20000 breast cancers . . .'. It is not surprising that the subsequent coverage in the general press had a profound and immediate effect on the confidence of both women and doctors in hormone replacement therapy. The licensing authorities were quick to take action. Indeed, on the day the paper was published, the Chairman of the UK Committee on Safety of Medicines (Professor Gordon Duff) wrote to all health professionals in the UK2. In his letter, he said that 'The Committee on Safety of Medicines (CSM) and its expert working group (EWG) on HRT have reviewed (the MWS) study. The key points are: The previously described small increase in risk of breast cancer with oestrogen-only products has been confirmed. The increased risk of breast cancer in association with use of combined . . . HRT is substantially higher than with oestrogen only therapy. . .. An increase in the risk of breast cancer becomes apparent within 1-2 years of starting treatment, irrespective of the type of HRT used. The risk . . . begins to decline when HRT is stopped and by 5 years reaches the same level as in women who have never taken HRT. No mention was made of the fact that the principal investigator for MWS and a memher of its Steering Committee were both members of the Expert Working Group of the CSM.

The subsequent publication of the endometrial cancer study on the Million Women cohort was not accompanied by the same adverse publicity except that tibolone was said to carry a greater risk than conventional combined products. However, both studies have the same serious design flaws and there are important aspects of the published reports that are inconsistent.

The most obvious limitation in the design of the study is that the exposure data were collected at the time the women were recruited to the study rather than when the cancer was diagnosed (or the study terminated). Recruitment could have been up to 6 years prior to the diagnosis of cancer (or the end of the study). Clearly, in the intervening period, the woman's use of HRT could have changed. In the breast cancer paper, the authors reported some findings from a sample survey of 12221 (1.12%) that was conducted part way through the study 'done on average 2.8 years after recording of baseline information'. Of the baseline never-users, 11% had become users, 22% of the baseline users had become non-users and 19% of the past users had re-started HRT. The publication on the Million Women endometrial cancelstudy3 also referred to the survey carried out part way through the study. The endometrial cancer study was based on the women who had not had a hysterectomy at the time of recruitment. A total of 716738 women were involved. The authors reported that 'Information on use of HRT after recruitment was available for 111201 women who completed a follow-up questionnaire an average of 2.8 years after recruitment. At follow-up, 95% of those who reported at entry that they had ever used HRT continued to report that they were ever-users, and, as expected, some ceased use during that period, so that fewer were current users 2.8 years after entry. Also, some women who were initially never-users of HRT began use subsequently, with 3 % of the never-users at entry reporting 2.8 years later that they were current users'. This follow-up survey appears to be the same as that referred to in the breast cancer paper. There is no explanation for there being almost ten times more women in the survey, as reported in the endometrial cancer paper, than in the breast cancer paper. Neither are the discrepancies between the two papers in respect of the proportion changing exposure category. For example, in the breast cancer paper, the proportion of the baseline never-users reported to have started HRT during the follow-up period was 11%; in the endometrial cancer paper it was 3%. It is also important to note that the women who were 'everusers' at baseline could never be in any category other than 'ever users', yet, in the sample survey, 5% of these women claimed not to have used HRT.

The published risk estimates were relative to 'never-users' at baseline, but, according to the breast cancer paper, at least 11% of those classified as 'never-users' had used HRT before the end of the study - presumably before the cancer was diagnosed.

It is clear that the inflation of the non-user denominator by the inclusion of users at the time of diagnosis would have inflated all the risk estimates. Because of the design of the study, it would not have been possible to estimate the risks of cancer relative to true non-users at the time of diagnosis. The only risk estimates that might be valid are those based on breast cancers diagnosed at the screening visit during which the women were recruited. This is because it is the only time at which the history of exposure to HRT was up to date. In a letter to The Lancet in October 2003(4), the investigators reported that the relative risk of breast cancer diagnosed at the screening visit for current users versus non- current users was 1.4 (98% confidence interval (CI) 1.3-1.5) and the risk amongst users of HRT for less than 1 year relative to never- users was 0.8 (95% CI 0.6-1.1). This is in stark contrast to the headline results in the published paper and the press release.

The endometrial cancer paper revealed another problem of classification. The breast cancer study had 17160 current users of tibolone at baseline (6% of the 285987 current users). In the endometrial cancer study, which involved fewer women because those who had a hysterectomy were excluded, there were 20 389 tibolone users at baseline. There were also more women classified as current users of Other' or 'unknown' HRT at baseline in the endometrial cancer study (12225) than in the breast cancer study (8580). It is unclear how these inconsistencies arose but, without doubt, they will affect the validity of the risk estimates for individual products.

Follow-up relied upon the relevant regional cancer registries yet no attempt appears to have been made to verify the completeness of the registries. More importantly, some registers were known not to be up to date. Amongst the 63 screening clinics that participated in the study, 36 (59%) were in regions where the cancer registry was deemed to be complete up to December 2001; 23 were in regions where the registry was complete up to December 2000. Four of the clinics were in Scotland. Although not reported in the breast cancer paper, it was reported in the endometrial cancer paper that the Scotland cancer registry was deemed complete up to December 1999. It must be presumed that reporting the completeness of the Scotland register was inadvertently omitted from the breast cancer paper. It follows that, in Scotland, there was no 'follow-up' for the women recruited in 2000 and 2001; likewise, for the patients recruited in 2001, in 23 of the clinics there was no follow-up data. It is unclear whether or not the women recruited after the date when the registries were deemed complete were included in the study. They should not have been because there was no outcome data.

The length of follow-up for the women recruited before the closing date of the registries in Scotland and those covering 23 of the clinics in England was less than that for the other 36 clinics. This is of importance. It will result in a systematic under- identification of cases in the regions where the registries were not up to date. This will be compounded by the fact that, in all centers, the period of 'follow-up' will vary according to the year in which the women were recruited. Thus, women recruited from 36 of the clinics in 1996 would have had 5 years of 'follow-up', whereas those recruited through the same clinics in 2001 would have at most 9 months of 'follow-up'. Clearly, the number of cancers would be higher amongst the early recruits than amongst the late recruits - because of the longer period of observation. No attempt was made to adjust for calendar year of recruitment.

The results of the studies on the Million Women cohort were accepted uncritically. Many commentators and the licensing authorities ignored earlier research, particularly that on the risk of breast cancer, and it was assumed that a statistical association proved a causal association.

It is clear that the authorities and others were impressed with the size of t\he Million Women Study - but large numbers never compensate for flaws in design, indeed they may well magnify their effects. It is equally clear that the large number of statistically significant associations impressed many people - but a significant association between an exposure and an outcome is only part of the evidence required to demonstrate cause.

Forty years ago, Professor Sir Austin Bradford Hill5 considered of which factors one should take account before interpreting an association as causal. His scheme is widely accepted amongst epidemiologists. The strength of the association is important. A relative risk of less than 2 is a weak association, and, even if statistically significant, could be the result of uncontrolled confounding or bias. If the association is causal, it is to be expected that there would be consistency between studies. Bush and her colleagues6 reviewed the published literature on breast cancer and HRT. It is clear that the risk estimates are not consistent indeed some studies showed no increase in risk. It is also clear that the results of the Million Women Study on breast cancer are outliers amongst the body of research available. Another criterion is 'temporality' - by this Bradford Hill meant that the exposure should pre-date the disease. In the investigation of cancer, it is essential to have some idea of the interval between the initiation of the tumor and its diagnosis. Some of the reported risks are not consistent with a lead time amounting to years. It is to be expected that the findings from observational studies should be biologically plausible if we are going to claim a causal relationship. From current knowledge of the growth rate of breast tumors, it is highly unlikely that a diagnosable tumor could arise within a year of exposure to HRT. Bradford Hill also looks for 'coherence' - that is, the association should not conflict with generally known facts. The complete disappearance of risk of breast cancer within a short time of cessation of HRT is in direct conflict with our knowledge of malignant tumors.

Although the Million Women Study showed a significant association between the use of HRT up to the time of recruitment and breast cancer, the study findings do not prove a causal association. The possibility of an increased risk of endometrial cancer amongst users of tibolone has been reported by other groups. At first sight, the relative risk of 1.79 (95% CI 1.43-2.25) could be seen as being in line with earlier work. However, the flaws in the design of the study and the appearance of more tibolone users in the endometrial investigation than were in the breast cancer investigation must raise doubts over the findings.

Under these circumstances, it is reasonable to ask why many commentators and licensing authorities place such weight on these data. Would their appraisal have been different had they known that there was no significant increase in risk in breast cancers detected at the recruitment screen amongst women who had used HRT for less than a year? Would their judgement have been affected by the knowledge that the last follow-up for Scotland was 1999 - not 2001, as implied in the original publication? Would they have been affected by the major discrepancies in the exposures in the endometrial and breast cancer studies? How would they have regarded the discrepancies in the report on the sample survey?

Observational epidemiological studies have an important place in medical research, but, in common with all scientific research, they must be designed, executed and reported with care. Even then, it is essential to be circumspect in the interpretation of their results. They form part of a jigsaw of scientific enquiry and should be consistent with other knowledge - if the results conflict, then alarm bells should ring. Finally, with all the uncertainties and complexities involved in research, one must ask whether it is reasonable for a group to issue alarming press releases before the work has been appraised by the wider scientific community. Generating widespread alarm by making unbalanced judgements on a single observational study has not been helpful. Thousands of women still have significant menopausal symptoms - they now have the additional concern that the treatment of their symptoms might kill them.

References

1. Beral V, et al. Breast cancer and hormone replacement therapy in the Million Women Study. Lancet 2003;362:419-27

2. Duff G. Hormone replacement therapy (HRT) and breast cancer - results of the UK Million Women Study. 8 August 2003 http:// medicines.mhra.gov.uk/ourwork/monitorsafequalmed/ safetymessages/ hrt803epinet.pdf

3. Beral V, Bull D, Reeves G. Endometrial cancer and hormone- replacement therapy in the Million Women Study. Lancet 2005;365:1543- 51

4. Beral V, Banks E, Reeves G, Bull D. Breast cancer and hormone replacement therapy (letter). Lancet 2003;362:1330-1

5. Hill AB. The environment and disease: association or causation? Proc R Soc Med 1965;58:295300

6. Bush TL, Whiteman M, Flaws JA. Hormone replacement therapy and breast cancer: a qualitative review. Obstet Gynecol 2001;98:498-508

Richard Farmer

Emeritus Professor of Epidemiology, Postgraduate Medical School, University of Surrey, UK

Correspondence: Professor R. Farmer, Postgraduate Medical School, University of Surrey, Stirling House, Surrey Research Park, Guildford, Surrey GU2 7DJ, UK

Copyright CRC Press Sep 2005

Source: Climacteric

More News in this Category