Jfprhc-2011-100229.indd

Does hormone replacement therapy
cause breast cancer? An application
of causal principles to three studies
Part 4. The Million Women Study
Samuel Shapiro,1 Richard D T Farmer,2 John C Stevenson,3 Henry G Burger,4 Alfred O Mueck5 Abstract
Part 1 we concluded that the CR findings Background Based principally on fi ndings in three
studies, the collaborative reanalysis (CR), the Town, Cape Town, South Africa2Emeritus Professor of Women’s Health Initiative (WHI) and the Million (E+P)] did not establish causality. In Part Women Study (MWS), it is claimed that hormone 2 we concluded that the WHI findings for replacement therapy (HRT) with estrogen plus progestogen (E+P) is now an established cause trast, in Part 3 we concluded that valid of breast cancer; the CR and MWS investigators claim that unopposed estrogen therapy (ET) also not increase the risk of breast cancer, and increases the risk, but to a lesser degree than may even decrease it; the latter possibility, does E+P. The authors have previously reviewed however, was statistically borderline.
the fi ndings in the CR and WHI (Parts 1–3). Objective To evaluate the evidence
Methods Using generally accepted causal criteria,
reported an increased risk of breast can- in this article (Part 4) the authors evaluate the cer in HRT users,21 and based on the com- fi ndings in the MWS for E+P and for ET.
Results Despite the massive size of the MWS the
fi ndings for E+P and for ET did not adequately E+P is an established and major cause of satisfy the criteria of time order, information bias, detection bias, confounding, statistical stability not the WHI investigators)15–18 claim that and strength of association, duration-response, ET also increases the risk, although to a internal consistency, external consistency or Correspondence to
Professor Samuel Shapiro,
biological plausibility. Had detection bias resulted in the identifi cation in women aged 50–55 years ples to the evidence from the MWS.21–24 In of 0.3 additional cases of breast cancer in ET the MWS the estimated levels of risk asso- users per 1000 per year, or 1.2 in E+P users, it Anzio Road, Observatory, Cape Town, South Africa; would have nullifi ed the apparent risks reported.
Conclusion HRT may or may not increase
of the impact the study had on regulatory authorities, and on the public perception of safety, it is especially important to eval- Background
The Million Women Study21–24
In Parts 1–3 of this series of articles we ological principles of causality1–4 to stud- mography at 3-year intervals.21 From May ies of the risk of breast cancer in users of tigators sent letters and questionnaires25 reported from the collaborative reanalysis (CR)5 (Part 16), and the Women’s Health questionnaires26 were sent 2–3 years after Initiative (WHI)7–18 (Parts 219 and 320). In Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 for breast cancer incidence and mortality in National the first two-thirds of the study population”, and there were “392 341 (38%) women for whom follow-up Below, except where otherwise stated, all 95% con- information [was] included in [the] analysis”.
fidence intervals (CIs) around the relative risk (RR) Among current users of HRT the respective RRs of estimates excluded 1.0, and for convenience they are in situ and invasive breast cancer were 1.55 and 1.74. The RRs were higher for invasive mixed ductal-lobu-lar or tubular tumours (2.13 and 2.66) than for duc- First report21 (2003)
tal tumours (1.63); the RRs were also higher among Among 828 923 postmenopausal women followed for E+P than among ET users, but for each type of cancer an average of 2.6 years the RRs of invasive breast cancer the RRs did not increase significantly with increasing for current and past users of HRT were 1.66 and 1.01 duration of use. For ductal and lobular tumours the (95% CI, 0.94–1.09). Among women currently using RRs declined with increasing body mass index (BMI) HRT at baseline the RRs for users of various types of HRT were as follows: ET, 1.30; E+P, 2.00; tibolone, The investigators concluded that “the risks of inva- 1.45; other or unknown HRT, 1.44. The difference sive lobular and tubular cancers associated with cur- between E+P vs ET was significant (p<0.0001).
rent use of both [ET and E+P] are higher than for For current ET use at baseline the RRs for <5 and invasive ductal cancer” and higher for E+P users than ≥5 years’ total duration were 1.21 and 1.34, and for E+P use, 1.70 and 2.21. For ET use the RRs for total durations of <1, 1–4, 5–9 and ≥10 years of use were Fourth report24 (2011)
0.81 (95% CI, 0.55–1.20), 1.25, 1.32 and 1.37; for Among 1 129 025 postmenopausal women followed E+P use they were 1.45, 1.74, 2.17 and 2.31.
until “the end of 2002 … two thirds of the partici- Among women who last used HRT ≤1 year pre- pants had been mailed the second questionnaire and viously the RR was 1.14; for exposures that ended the response was 65%”. During 4.05 million WY of 2–≥10 years previously the RRs approximated unity. follow-up 15 759 invasive and in situ breast cancers The average time to diagnosis was 1.2 years, and within 1.7 years of diagnosis the RR of fatal breast The RRs for current users of HRT, ET, E+P, tibo- lone and other and unknown HRT were 1.68, 1.38, The investigators estimated that the “use of HRT 1.96, 1.38 and 1.55, respectively, and the estimates by UK women aged 50–64 years … resulted in an were statistically heterogeneous (p<0.001). In the extra 20 000 incident breast cancers, combined [E+P] first 2 years after HRT ceased the RR was 1.16, after accounting for 15 000” of them. They also estimated which the RRs approximated unity. For durations of that HRT would “result in five to six extra cancers per use of <5 and ≥5 years the respective RRs among ET 1000 women with 5 years’ use and 15–19 … per 1000 users were 1.24 and 1.44; among E+P users they were with 10 years’ use”. They concluded that “current use of HRT is associated with an increased risk of incident For both ET and E+P users the RRs were lower for and fatal breast cancer” … [which is] … “substantially breast cancers diagnosed in the first 4 months after greater for [E+P] combinations than for other types recruitment than subsequently [ET, 1.19 and 1.50 (p<0.001); E+P, 1.41 and 2.32 (p<0.001)]. For ET users the RRs of ‘screen-detected’ and ‘non-screen- Second report22 (2004)
detected’ cancers were 1.16 and 1.59 (p<0.001); Among users of HRT at baseline the RRs at 0.1 (‘screen- for E+P the corresponding estimates were 1.64 and detected’), 0.7, 1.5, 2.5 and 3.4 years of follow-up 2.81 (p<0.001). Those comparisons “should [have were 1.37, 2.66, 2.16, 1.66 and 1.70, respectively. The included] virtually all breast cancers found at screening average durations of use ranged from 6.1 to 6.9 years. soon after the baseline questionnaire was completed”.
The RRs were higher for E+P than for ET users, and For current ET users whose use began <5 and ≥5 maximal at 0.7 years (ET, 1.72; E+P, 3.31).
years after the menopause, the RRs were 1.43 and For women aged 50–55 years who used HRT for 5 1.05 (p<0.001); for E+P users the estimates were 2.04 years the estimated absolute risks attributable to ET and 1.53 (p<0.001). “The proportionate increase in and E+P use were 1.5 and 6.0 per 1000.
risks of breast cancer associated with use of hormone therapy was greater among lean women than among Third report23 (2006)
obese women”, but within BMI strata (≥25 kg/m2 and Among 1 031 224 postmenopausal women followed <25 kg/m2) the HRT-associated RRs remained higher over 3.6 million woman-years (WY) for the incidence for those whose use commenced <5 years after the of invasive and in situ breast cancer “the mean time … from … last contact to the end of follow-up was 2.7 For both ET and E+P users the RRs declined with years [SD (standard deviation)1.1]”. “At the time of increasing tumour grade (Grades I–III): ET, 1.27, 1.16, the analysis follow-up information was available for 0.87 (p<0.001); E+P, 2.42, 1.67, 1.03 (p<0.001). Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 For estrogen receptor (ER)-positive vs ER-negative Information bias
status the RRs for ET users were 1.76 and 1.29 Information bias in a cohort study is unusual, but it can (p=0.005); for E+P users the estimates were 3.10 and occur, and in the MWS it was likely. At recruitment 1.37 (p<0.001). For node-positive vs node-negative HRT users already aware of as yet undiagnosed breast tumours among ET users the RRs were 1.19 and 1.09 lumps, or of suspect mammographic changes identi- (p=0.3); among E+P users they were 2.00 and 1.66 fied before recruitment (see: Detection bias), could have tended to overestimate the total duration of use. The investigators concluded that “risks were sub- Had women who already had breast cancer at base- stantially greater among users of [E+P] than estrogen line been excluded, that bias could largely have been only formulations and if hormonal therapy started at or around the time of menopause than later”.
A defect in the study design may also have facilitated the occurrence of information bias. Ethinylestradiol Evaluation of the MWS
(EE), listed as one of 34 memory-prompts in the ques- Below we evaluate whether the evidence in the tionnaire25 as an HRT preparation, is a synthetic estro- MWS accorded with generally accepted principles of gen present exclusively in oral contraceptives. Women causality.1–4 The principles are inter-related, and when who were aware of breast lumps at recruitment, or who had suspect mammographic changes (see: Time order and detection bias), could erroneously have identified Time order
EE as HRT. Soon after publication of the MWS report21 If allowance is made for the time from the diagnosis the authors stated in an erratum that what was meant of breast cancer to its recording in a registry, virtually by ‘ethinylestradiol’ was ‘estradiol’.28 Yet the error was all the cases identified at 0.1 years of follow-up (HRT: not corrected in the second questionnaire,26 adminis- RR, 1.37)22 or at 4 months (ET: RR, 1.19; E+P: RR, tered 2–3 years after the first questionnaire.25 1.41),24 were already present when the women were recruited (see: Detection bias) and time order was Detection bias
violated. In a properly designed cohort study breast The design of a study of the risk of breast cancer in cancers already present at baseline should have been relation to the use of HRT in which the women were recruited from a screening programme guaranteed that Time order was further violated in respect of the it would be biased. By definition, women who decided timing and duration of HRT use. In the third report23 to have mammograms were alerted to the possibility follow-up information on HRT use was unavailable for of breast cancer, as has also been acknowledged in an 62% of the women. In the fourth report,24 by December earlier study based on mammographic screening,29 2002 the follow-up questionnaire had been received by and concern that HRT may cause the disease has been about 66% of the women, among whom the response widespread, and has increased over time. The MWS rate was 65%. Hence follow-up information on HRT invitation was explicit in the first questionnaire:25 “We use [and on menopausal status (see: Detection bias) and have a unique opportunity … to learn about the way on confounders (see: Confounding)] was missing for different types of HRT … [affect] a woman’s health, about 57% [1 – (±0.66×0.65)×100] of the women. particularly her breasts”. That wording ensured that Following publication of the WHI findings7 there was a HRT users already aware of breast lumps, or of sus- rapid and marked decline in the use of HRT.27 For that pected breast cancer, would selectively participate reason, as well as for other reasons (e.g. HRT-induced breakthrough bleeding),7 since 66% of ever-users of There was quantitative evidence of detection bias. HRT at baseline were current users [our calculation: First, HRT users were selectively enrolled: 32% of the derived from Figure 1 (current use) and Figure 2 (past women who participated and 19% or those who did use) in Reference 21], a substantial proportion could not were HRT users.30 Second, the data suggested that have become past users by the end of 2002.
women already aware of breast lumps, or of suspected How unreliable were the data? Recruitment com- breast cancer, tended selectively to participate (see: menced in 1996 and follow-up ended in December Time order): whereas the incidence of breast cancer 2002.24 For about 50% of the women the time from in the MWS population was 2.8 per 1000 WY,31 in the last contact to diagnosis was >1.2 years,21 and to the population at large it was 2.0 per 1000 WY.21 Third, end of follow-up >2.7 years.23 For women enrolled the baseline RRs of 1.3722 or 1.4124 (‘screen-detected’ in 1996 that interval could have been as much as 6 breast cancer) indicated that women who both used years. Thus it is likely that much of what was defined HRT and who were also aware of breast lumps, or of in the analysis as current HRT use became past use suspect lesions, or of suggestive precancerous changes during follow-up. In addition, the duration data were identified in earlier mammograms, were the most likely incorrect (see: Duration-response), as were the data to participate. Fourth, the average time from recruit- on menopausal status and confounding (see: Detection ment to breast cancer diagnosis was 1.2 years,21 and 1.7 years thereafter the RR of fatal breast cancer was Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 1.22. An increased risk of fatal cancer among HRT out. Hence, it was to be expected that detection bias users within 2.9 (1.2+1.7) years of recruitment was would be greater for E+P users than for ET users.
not plausible (see: Biological plausibility), and it could The RRs for invasive lobular and tubular tumours have been due to the selective enrolment of HRT users were higher than for ductal tumours.23 Lobular and with pre-existing suspected or diagnosed breast cancer. tubular tumours are more highly differentiated, smaller, Fifth, the RRs declined with increasing BMI,24 a known and more slow-growing than ductal tumours,34 35 and risk factor for breast cancer in postmenopausal women the detection of lobular tumours by mammography is (see: Biological plausibility), and the larger the breasts,
also more difficult.36 More intensive scrutiny of mam- the less likely was it that otherwise occult breast can- mograms of HRT users than of non-users could have cer would selectively have been detected among HRT resulted in the selective detection of lobular tumours, especially in radiologically dense mammograms, that Detection bias could also have occurred during fol- might otherwise have gone undetected.
low-up, as previously described in our critique of the The RR for in situ breast cancer was 1.55.23 In situ CR.6 Briefly, HRT users are advised to have regular tumours are seldom clinically detectible, usually they breast examinations and mammograms, and in the are identified by mammography, and the investiga- MWS users more frequently underwent mammog- tors acknowledged that detection bias was likely (see: raphy than did non-users;30 when mammograms are Detection bias). Yet in the fourth report24 in situ and performed HRT use is routinely recorded, and about invasive breast cancers were considered together. In that 30% of breast cancers actually present go undetected;32 report the RRs were higher if HRT had commenced about 5% of postmenopausal women have ‘clinically within 5 years of the menopause than subsequently. silent’ breast cancer;33 and HRT diminishes the sensi- However, since the data for in situ breast cancer were tivity of mammography.32 The mammograms of HRT biased, the combination of in situ and invasive breast users could have been more intensively scrutinised than cancer was also biased. In addition, most of the women those of non-users, especially if they were radiologi- who were premenopausal at recruitment would have cally dense, and otherwise occult breast cancer could reached the menopause during follow-up, among the selectively have been detected among the users.
57% of women not followed that information was For both ET users and E+P users the RRs were missing, and there was substantial misclassification of lower during the first 4 months of follow-up than sub- menopausal status, and of the time since menopause sequently.24 The investigators stated that “it has been suggested that part of the increased hormone thera- The RRs declined with increasing tumour grade, py-associated risk … observed in this study may have and were higher for ER-positive than for ER-negative resulted from the selective recruitment of hormone tumours, and for node-positive than for node-neg- therapy users who already had symptoms of breast ative tumours.24 As shown in Table 1 (our calcula- cancer. If that had happened there would have been tions: derived from Figures 1 and 3 in Reference 24) a greater hormone therapy-associated excess of breast unknown values for tumour grade, ER status and nodal cancer soon after recruitment than subsequently. status among current users of ET and E+P, and among However, the opposite was found”. They argued that never-users of HRT, ranged from 49.5% to 74.1%. these findings “largely [reflected] the lower hormone Such high rates cast doubt on the validity of the evi- therapy-associated risks observed for screen-detected dence. In addition, the declining RRs with increasing breast cancers than for non-screen-detected breast can- tumour grade could have been biased if more com- cers”. That claim ignored the likelihood that during mon use of mammography by HRT users than by non- follow-up HRT users could more commonly have had users resulted in the selective detection of low-grade repeat mammograms than non-users (see: Duration- tumours; an association with ER-positivity could have response), and because of a further defect in the study occurred if breast cancers in HRT users were more design that possibility could not be assessed: informa- commonly tested, and if ER-positive, more commonly tion on repeat mammograms was not solicited in the documented in the registries; and the higher RRs for second questionnaire26 (see: Confounding).
node-positive than for node-negative tumours could There was further evidence to suggest detection bias. readily have been due to detection bias.
The RRs were consistently lower for ET users than for How much bias would it have taken to account for E+P users.21–24 Unopposed ET causes uterine cancer, the findings? In the first report21 the investigators esti- and ET is preferentially prescribed to hysterectomised mated that among women aged 50–64 years the use of women, among whom vaginal bleeding does not HRT would result in “five to six extra cancers per 1000 occur. By contrast, E+P is preferentially prescribed women with 5 years’ use and 15–19 … per 1000 with to women with a uterus, among whom breakthrough 10 years’ use”. Thus if detection bias resulted in the bleeding is common;7 and bleeding makes it manda- identification of 1–1.2 (5–6/5) otherwise occult cases tory to rule out endometrial cancer. HRT users alerted each year among 1000 women exposed for 5 years, to the risk of that cancer would have become worried or 1.5–1.9 (15–19/10) cases each year among women about breast cancer as well, and have sought to rule it exposed for 10 years, that bias would have nullified Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 it may be reasonable to judge that it might perhaps be Table 1 Breast cancer in the Million Women Study:
reduced, but not be obliterated, even if it were possible percentages of unknown values for tumour grade, to entirely eliminate all sources of bias and confound- estrogen receptor status and nodal status among current users of estrogen therapy, current users ing. But if an association is small it may be impossible of estrogen plus progestogen and never-users of to judge. In the latter circumstance ‘statistical signifi- cance’ may not equate with causality: given a massive amount of data, all that may be accomplished is to rule out chance as one possible explanation, but not bias or In the four reports the highest overall RR for HRT users was 1.74,23 and RRs in excess of 2.0 were identi- fied only in subgroups. For ET users the overall RR was 1.30, and <2.00 in all subgroups. Such small RRs could have been due to bias or confounding. For E+P users the overall RR of 2.00 was again small,21 but *Derived from Figure 1 in Reference 24.
significantly higher than the estimate of 1.30 for ET †Derived from Figure 3 in Reference 24.
(p<0.0001). Or put another way, the RR for E+P ver- E+P, estrogen plus progestogen; ER, estrogen receptor; ET, estrogen sus ET was 1.54 (2.00/1.30). Such a small association therapy; HRT, hormone replacement therapy.
could readily have been biased or confounded (see: Detection bias and confounding), illustrating how in a the findings. In the second report,22 among women massive study, virtually any deviation of RR from 1.0, aged 50–55 years the respective absolute risks for ET no matter how small, can yield a p value of <0.0001.
or E+P use for 5 years were estimated to be 1.5 and Dose/duration-response
6.0 per 1000. That is, if detection bias resulted in the Under a promotional hypothesis it might reasonably be identification of 0.3 (1.5/5) additional cases in ET users expected that the use of HRT would confer a greater each year, or 1.2 (6.0/5) additional cases in E+P users, risk of breast cancer, the higher the dose or the longer that bias would have nullified the findings. Absolute the duration of use (see: Biological plausibility).
risks ranging from 0.3 to 1.9 per 1000 women per year could plausibly have been due to detection bias.
Dose-response
Dose-response was not analysed.
Confounding
Confounding was incompletely controlled. In the
Duration-response
first,21 second22 and third23 reports the factors allowed In the first report,21 for women who were using HRT for included age, time since menopause, parity, age at baseline (defined in the MWS as current users) the at first birth, family history, BMI, region and socio- total duration of use of all episodes of use, current economic status. In the fourth report24 age at men- plus past, was analysed. That analysis was incorrect. opause and alcohol consumption were also allowed Since the RR approximated unity within 2 years of for. During follow-up factors such as menopausal stopping,24 the duration of past use was irrelevant, and status, time since menopause, age at menopause and only the duration of the current episode of use should BMI changed, and for about 57–62% of the women have been analysed. In addition, the analysis of dura- the information was missing (see: Time order). In tion of use, as represented at baseline, misrepresented addition, information on the receipt of a mammo- the actual duration of use, since follow-up information gram during follow-up was not solicited in the second was missing for 62% of the women (see: Time order).
questionnaire26 (see: Detection bias).
A further defect in the study design made it impos- Statistical stability and strength of association
sible to analyse the duration of current HRT use among In our critique6 of the CR5 we alluded to the relation- women who used more than one product. In the base- ship between the statistical stability and strength of line questionnaire25 five relevant questions were asked: any given association: if a RR is ‘large’ (say >5.0), a “32. Have you ever used [HRT]?”; “35. For about how 95% CI that excludes 1.0 (i.e. a ‘statistically signifi- many years in total have you used HRT?”; “36. Are you cant’ association) can be documented in a relatively now using HRT?”; “37. What is the name of the most small study. But if a RR is ‘small’ (say <2.0), usually recent HRT you have used?”; and “38. For how many it can only be documented in a massive study. The years did you use the most recent type of HRT?”.
difficulty however, is that “if a massive study is suf- Based on questions 32 and 35, among current HRT ficiently massive, any deviation of the RR from 1.0, no users at baseline who used more than one product the matter how small, becomes ‘significant’”; but it may total duration of ever-use could be analysed, but based be impossible to discriminate among bias, confound- on questions 36, 37 and 38 the duration of current ing and causation as alternative explanations. By con- HRT use could not be. To illustrate, consider a current trast, “in a well-conducted study, when a RR is large, HRT user who at baseline had used E+P for 9 years, Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 and following hysterectomy, ET for 1 year: the current effects of estrogens on estrogen-sensitive cells,37 1-year duration of ET use would have been recorded, or the excessive metabolism of estrogens to highly but the current 10-year duration of HRT use (E+P, 9 active compounds38 with strong proliferative as years + ET, 1 year) would not have been.
well as possibly genotoxic effects. However, estro- In the second report22 duration data were not given. gens also have antiproliferative and pro-apoptotic In the third report23 “there was no significant differ- effects,39 which could possibly reduce the risk of ence in the trends in [RR] with duration of use of breast cancer. In addition, estrogens can be metabo- either type of hormone therapy [ET or E+P] for ductal, lised not only to potentially genotoxic metabolites, tubular or lobular cancer”. In the fourth report24 the but also to carcino-protective metabolites, such as RRs for ≥5 years of use of ET and of E+P at baseline were higher than for <5 years of use, and higher for In short, some mechanisms could possibly increase E+P than ET users. Those differences could have been the risk of breast cancer in HRT users, and other mecha- due to detection bias; trends according to durations nisms could decrease it. However, under a promotional of <1, 1–4, 5–9 and ≥10 years were not presented; hypothesis, for the most aggressively multiplying cells it ‘total duration’ of use again referred to all episodes is generally accepted that on average it takes at least 10 of use, not to current use; and the duration data were years to attain a tumour diameter of about 1 cm, which again misclassified, because of follow-up information is about the smallest lesion that can be diagnosed clini- on HRT use was missing for 57% of the women.
cally.38 In the MWS the average total duration of HRT Finally, the RRs for increasing duration of follow- use at baseline was 6.1–6.9 years,22 and the duration of up were inconsistent with a duration-response effect current use would have been appreciably less. Since the (see: Internal consistency). Among women who used RR approximated unity within 2 years of discontinuing HRT, ET or E+P at baseline the RRs were highest at HRT use, among current users of HRT the duration of 0.7 years of follow-up, after which they declined.22 past use cannot have had any effect. It is implausible Yet under causal assumptions, the longer the duration that the current use of HRT at baseline for less than 6 of follow-up, the higher should the RRs have been. A years could have increased the risk of breast cancer. It plausible explanation of these inconsistent findings is is also implausible that cancer cells, once already pro- that violation of time order and detection bias could moted, and once already invasive, could have ‘unpro- have been greatest during the first year of follow-up.
moted’ within 2 years of stopping HRT.24 Obesity is a risk factor for breast cancer in postmen- Internal consistency
opausal women, perhaps because of increased endog- As described above, the RRs according to duration of enous estrogen secretion,37 and the RRs declined with follow-up were inconsistent (see: Duration-response).
increasing BMI.24 Under a causal hypothesis, however, although obesity itself increases the risk of breast can- External consistency
cer, the RRs among HRT users should have been higher For ET users the MWS findings were inconsistent with than among non-users within strata of BMI, and the those of the WHI clinical trial15 16 in which the evi- decline in the RR was explicable by the diminished dence suggested that unopposed ET does not increase sensitivity of mammographic screening with increas- the risk of breast cancer. In the MWS there was quan- titative evidence of bias, whereas in the WHI trial women were randomly assigned, ‘double-blind’, to ET Conclusions
or placebo, all participants were hysterectomised, vagi- The name ‘Million Women Study’ implies an authority nal bleeding did not occur, ‘unblinding’ was seldom beyond criticism or refutation. Many commentators, necessary, the ‘unblinding’ rate was <2.0%, and there and the investigators, have repeatedly stressed that it was the largest study of HRT and breast cancer ever For E+P users the MWS findings were inconsistent conducted. Yet the validity of any study is dependent with those of the CR. In the MWS the RRs approxi- on the quality of its design, execution, analysis and mated unity within 2 years of stopping HRT;24 in interpretation. Size alone does not guarantee that the the CR the RR only declined to unity 5 years after findings are reliable. The MWS was an observational study, and it had the attendant problems and uncer- Biological plausibility
tainties intrinsic to such studies. If the evidence was Elsewhere we have considered relevant pathological unreliable, the only effect of its massive size would and experimental evidence for and against the possi- have been to confer spurious statistical authority to bility that HRT may cause breast cancer.6 19 20 Briefly, the hypothesis is not that HRT causes genetic muta- Here we conclude that the evidence in the MWS tion (initiation), but that estrogens, and probably was indeed unreliable. There were defects in the study progestogens as well, accelerate the proliferation design, and the findings did not adequately satisfy the of otherwise slowly growing malignant cells (pro- principles of causation. In terms of time order, infor- motion). Possible mechanisms are the proliferative mation bias, detection bias, confounding, statistical Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 stability and strength of association, dose/duration- 14 Prentice RL, Manson JE, Langer RD, et al. Benefits and risks of
postmenopausal hormone therapy when it is initiated soon after response, internal consistency, external consistency menopause. Am J Epidemiol 2009;170:12–23.
and biological plausibility the study was defective.
15 The Women’s Health Initiative Steering Committee. Effect of
HRT may or may not increase the risk of breast can- conjugated equine estrogen in postmenopausal women with cer, but the MWS did not establish that it does.
hysterectomy: the Women’s Health Initiative randomized
controlled trial. JAMA 2004;291:1701–1712.
Acknowledgement The authors thank Helen Seaman
16 Stefanick ML, Anderson GL, Margolis KL, et al. Effects
of conjugated equine estrogens on breast cancer and Competing interests Samuel Shapiro, John Stevenson,
mammography screening in postmenopausal women with Henry Burger, and Alfred Mueck presently consult, hysterectomy. JAMA 2006;295:1647–1657.
and in the past have consulted, with manufacturers 17 LaCroix AZ, Chlebowski RT, Manson JE, et al. Health
of products discussed in this article. Richard Farmer outcomes after stopping conjugated equine estrogens among has consulted with manufacturers in the past.
postmenopausal women with prior hysterectomy: a randomized
controlled trial. JAMA 2011;305:1305–1314.
Provenance and peer review Not commissioned;
18 Prentice RL, Chlebowski RT, Stefanick ML, et al. Conjugated
equine estrogens and breast cancer risk in the Women’s Health Initiative clinical trial and observational study. Am J Epidemiol References
2008;167:1407–1415.
1 Hill AB. The environment and disease: association or causation?
19 Shapiro S, Farmer RDT, Mueck AO, et al. Does hormone
Proc R Soc Med 1965;58:295–300.
replacement therapy cause breast cancer? An application of 2 Susser M. Causal Thinking in the Health Sciences. New York,
causal principles to three studies. Part 2. The Women’s Health Initiative: estrogen plus progestogen. J Fam Plann Repred Health 3 US Depart of Health, Education and Welfare: Public Health
Care 2011;37:165–172.
Service. Smoking and Health: Report of the Advisory Committee
20 Shapiro S, Farmer RDT, Seaman H, et al. Does hormone
to the Surgeon General of the Public Health Service (Public replacement therapy cause breast cancer? An application of Health Service Publication No. 11030). Washington, DC: US causal principles to three studies. Part 3. The Women’s Health Initiative: unopposed estrogen. J Fam Plann Repred Health Care 4 Susser M. What is a cause and how do we know one? A grammar
2011;37:225–230.
for pragmatic epidemiology. Am J Epidemiol 1991;133:635–648.
21 Million Women Study Collaborators. Breast cancer and
5 Collaborative Group on Hormonal factors in Breast Cancer.
hormone replacement therapy in the Million Women Study. Breast cancer and hormone replacement therapy: collaborative Lancet 2003;362:419–427.
reanalysis of data from 51 epidemiological studies of 52 705 22 Beral V, Banks E, Reeves G, et al. The effect of hormone
women with breast cancer and 108 411 women without breast therapy on breast and other cancers. In: Critchley H, Gebbie A, cancer. Lancet 1997;350:1047–1059.
Beral V (eds). Menopause and Hormone Replacement. London: 6 Shapiro S, Farmer RD, Seaman H, et al. Does hormone
replacement therapy cause breast cancer? An application of 23 Reeves GK, Beral V, Green J, et al. Hormonal therapy for
causal principles to three studies: Part 1. The Collaborative menopause and breast-cancer risk by histological type: a cohort Reanalysis. J Fam Plann Reprod Health Care 2011;37:103–109.
study and meta-analysis. Lancet Oncol 2006;7:910–918.
7 Writing Group for the Women’s Health Initiative
24 Beral V, Reeves G, Bull D, et al. Breast cancer risk in relation to
Investigators. Risks and benefits of estrogen plus progestin
the interval between menopause and starting hormone therapy. in healthy postmenopausal women. Principal results from the J Natl Cancer Inst 2011;103:1–10.
Women’s Health Initiative randomized controlled trial. JAMA 25 The Million Women Study: A National Survey of Women 2002;288:321–333.
Invited for Breast Screening. First questionnaire. http:// 8 Chlebowski RT, Hendrix SL, Langer RD, et al. Influence of
www.millionwomenstudy.org/files/mws-web1.pdf [accessed estrogen plus progestin on breast cancer and mammography in healthy postmenopausal women: the Women’s Health Initiative 26 The Million Women Study: Confidential National Study Randomized Trial. JAMA 2003;289:3243–3253.
of Women’s Health. Second questionnaire. http://www.
9 Anderson GL, Chlebowski RT, Rossouw JE, et al. Prior
millionwomenstudy.org/files/mws-web2.pdf [accessed hormone therapy and breast cancer risk in the Women’s Health Initiative randomized trial of estrogen plus progestin. Maturitas 27. Gompel A, Plu-Bureau G. Is the decrease in breast cancer
2006;55:103–115.
incidence related to a decrease in postmenopausal hormone 10 Heiss G, Wallace R, Anderson GL, et al. Health risks and
therapy? Ann N Y Acad Sci 2010;1205:268–276.
benefits 3 years after stopping randomized treatment with 28 Million Women Study Collaborators. Errata. Lancet
estrogen and progestin. JAMA 2008;299:1036–1045.
2003;362:1160.
11 Chlebowski RT, Anderson GL, Gass M, et al. Estrogen
29 Morrison AS, Brisson J, Khalid N. Breast cancer incidence and
plus progestin and breast cancer incidence and mortality in mortality in the breast cancer detection demonstration project. postmenopausal women. JAMA 2010;304:1684–1692.
J Natl Cancer Inst 1988;80:1540–1547.
12 Prentice RL, Chlebowski RT, Stefanick ML, et al. Estrogen plus
30 Banks E, Beral V, Cameron R, et al. Comparison of various
progestin therapy and breast cancer in recently postmenopausal characteristics of women who do and do not attend for breast women. Am J Epidemiol 2008;167:1207–1216.
cancer screening. Breast Cancer Res 2002;4:R1.
13 Chlebowski RT, Kuller LH, Prentice RL, et al. Breast cancer
31 Shapiro S. The Million Women Study: potential biases do
after use of estrogen plus progestin in postmenopausal women. not allow uncritical acceptance of the data. Climacteric N Engl J Med 2009;360:573–587.
2004;7:3–7.
Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229 32 Greendale GA, Reboussin BA, Sie A, et al. Effects of estrogen
disease at diagnosis in 184 patients. AJR Am J Roentgenol and estrogen-progestin on mammographic parenchymal 1993;161:957–960.
density. Postmenopausal Estrogen/Progestin Interventions 37 International Agency for Research on Cancer (IARC).
(PEPI) Investigators. Ann Intern Med 1999;130:262–269.
IARC Monographs on the Evaluation of Carcinogenic Risks 33 Welch HG, Black WC. Using autopsy series to estimate the
to Humans (Volume 72): Hormonal Contraception and disease “reservoir” for ductal carcinoma in situ of the breast: Postmenopausal Hormone Therapy. Lyon, France: IARC, how much more breast cancer can we find? Ann Intern Med 1997;127:1023–1028.
38 Seeger H, Wallwiener D, Kraemer E, et al. Comparison
34 Li CI, Moe RE, Daling JR. Risk of mortality by histologic type
of possible carcinogenic estradiol metabolites: effects on of breast cancer among women aged 50 to 79 years. Arch Intern proliferation, apoptosis and metastasis of human breast cancer Med 2003;163:2149–2153.
cells. Maturitas 2006;54:72–77.
35 Porter PL, El-Bastawissi AY, Mandelson MT, et al. Breast tumor
39 Dietel M, Lewis MA, Shapiro S. Hormone replacement
characteristics as predictors of mammographic detection: therapy: pathobiological aspects of hormone-sensitive cancers comparison of interval- and screen-detected cancers. J Natl in women relevant to epidemiological studies on HRT: a mini- Cancer Inst 1999;91:2020–2028.
review. Hum Reprod 2005;20:2052–2060.
36 Krecke KN, Gisvold JJ. Invasive lobular carcinoma
40 Seeger MH. 2-Methoxyestradiol – biology and mechanism of
of the breast: mammographic findings and extent of action. Steroids 2010;75:625–631.
Shapiro S, Farmer RDT, Stevenson JC, et al. Family Planning (2011). doi: 1136/jfprhc-2011-100229

Source: http://www.thebms.org.uk/publicdownloads/Shapiro%20MWS%20critique%20JFPRHC%202012.pdf