Risk of Endometrial Cancer After Tamoxifen Treatment

ABSTRACT: A total of 18 studies have been publishedconcerning the possible relationship of tamoxifen to endometrial cancer.Findings range from a protective effect (RR = 0.47) to a risk ratio ashigh as 15.2. Most studies are based on previous clinical trials of thedrug. There are several recurring biases throughout almost all of the studiesreported to date. This paper provides a critical review of each of thestudies, including identification of bias sources and potential confoundingvariables. A causal association has not been proven (nor even stronglyindicated) for tamoxifen and endometrial cancer, and further investigation,with less bias, will be required to resolve the question. [ONCOLOGY 11(Suppl1):25-33,1997]

Introduction

There is consensus that tamoxifen is effective in preventing both metastaticdisease and contralateral breast cancer. Because of its effectiveness inbreast cancer treatment, the drug is now the subject of clinical trialsfor breast cancer prevention in healthy women at high risk for developingthe disease. However, there has been considerable controversy over thepossibility that tamoxifen either causes or stimulates endometrial cancer,and this has discouraged enrollment in the prevention trial.

The controversy began with a series of case reports of women diagnosedwith endometrial cancer after tamoxifen treatment. Since those initialcase reports, a variety of epidemiologic studies have been reported, althoughthe findings of those studies have been far from unanimous.

It is the purpose of this paper to critically examine the epidemiologicevidence that has accumulated to date. In particular, the issues of confoundingvariables and bias will be explored to attempt to reconcile the differingresults of the various studies.

In performing this critical analysis, I have chosen to use only studiesreporting original data, and have omitted from the review most case reports,review articles, and position papers concerning tamoxifen.

The Common Epidemiology of Breast and Endometrial Cancers

Breast and endometrial cancers share many risk factors. Risk is agerelated and increases with early menarche, late menopause, nulliparity,obesity, and estrogen replacement therapy (ERT), [1-4] although the magnitudeof the risk varies for each disease. Because of their common risk factors(mostly estrogen-related), it is not surprising that the two diseases occurmore frequently in the same individual than would be expected by chance.[5-8]

The risk factors common to both diseases can all be viewed as potentialconfounding variables in any study of the relationship between breast cancerand endometrial cancer. A confounding variable is a factor related to boththe independent and the outcome variable. Avoiding bias, and recognizingand controlling potentially confounding variables are issues in every epidemiologicstudy of the possible role of tamoxifen and endometrial cancer followingbreast cancer.

The epidemiologic approach to the question of possible carcinogeniceffects of tamoxifen involves a choice of either cohort or case-controlstudies. The most obvious cohorts of tamoxifen-exposed women are the participantsin the many clinical trials of the drug.

In designing a cohort study, the epidemiologist must define not onlythe exposed cohort but the reference cohort against which to compare theexposed group. Naturally, women in the trials who were randomized to notreceive the drug are a likely comparison group, although there may be someadvantage (eg, availability, sample size) to using other cohorts or populations(such as U.S. SEER data) for comparison.

It must be remembered, however, that even without tamoxifen treatment,breast cancer patients will be expected to have a higher incidence of endometrialcancer than will women without breast cancer, thus artificially inflatingrelative risk estimates based on the experience of non-breast cancer patients.

Surveillance Bias

If the comparison group or reference population is not under the samesurveillance as women in the clinical trial, there will also be an opportunityfor surveillance bias. Because tamoxifen-treated women are under closermedical watch than the general population, any second disease (eg, endometrialcancer) is more likely to be detected. This bias will be especially importantif the second disease is characterized by long duration, low mortality,and infrequent symptoms—all characteristics of endometrial carcinoma.

A high proportion of endometrial carcinoma may be diagnosed only afteran occult phase that has been estimated to average four to six years.[9]Because of shared risk factors, and also because of surveillance bias,one can predict that any standardized incidence ratio of a tamoxifen-treatedclinical trial cohort will be elevated if a non-trial population is thecomparison group.

Ascertainment Bias

If the comparison group is another arm of a randomized clinical trial,then surveillance bias should not be operative. However, tamoxifen is knownto induce several gynecologic symptoms. The nonmalignant gynecologic complicationsare well described in a review by Assikis and Jordan.[10] A woman undergoingtamoxifen treatment is about three times as likely to see a gynecologistbecause of symptoms such as hot flashes or bleeding due to endometrialproliferation.[11-13]

The frequency of gynecologic conditions leads to an increase in doctorvisits, which creates an ascertainment bias. Ascertainment bias will alsobe present, as will surveillance bias, in any comparison between a tamoxifen-exposedcohort and the general population. This bias has been recently discussed.[14]

Because of surveillance and ascertainment biases, clinical trial cohortsare not ideal groups for the study of any tamoxifen-endometrial cancerrelationship unless the non-tamoxifen group undergoes the same degree ofgynecologic investigation as the tamoxifen group. To date, this has notbeen the case. Because they were never designed to ascertain risk of endometrialcancer, the clinical trials were not planned with the same degree of gynecologicsurveillance for all trial arms. Neither was there appropriate documentationof possible confounding variables, especially estrogen replacement therapy(ERT), pre- or post-tamoxifen treatment.

Since many postmenopausal women would have received ERT prior to theirbreast cancer diagnosis and some may have received ERT afterward, it isprobably the most potent confounding variable of all, and it has been reportedto increase endometrial cancer risk by as much as ninefold[8,15-17] ormore with prolonged duration of use.

Case-Control Studies

As an alternative to following a cohort, one may use a case-controlstudy, either population-based or within a cohort (the nested case-controlstudy). Although cases of endometrial cancer are readily selected fromany population, one must be aware of possible bias. If case selection isinfluenced in any way by ascertainment due to tamoxifen treatment, thestudy will be biased, since the exposure variable influences identificationof cases.

Even if a group of endometrial cancer patients is selected without knowledgeof previous exposure to tamoxifen, ascertainment bias may be present butunrecognized.

Tamoxifen-induced gynecologic symptoms may have triggered the investigationthat led to the diagnosis of endometrial cancer. Since the gynecologicsymptoms from tamoxifen do not differ from common symptoms due to otheror unknown causes, there is no easy way to identify the proportion of casesthat may have been ascertained because of the tamoxifen-induced symptoms.Presumably, controls are selected from a population without an equallyhigh probability of gynecologic investigation. As a result, there is considerablechance that any case-control study of this nature will suffer from a bias,of unknown magnitude, that will elevate relative risk estimates.

It is also possible that because of tamoxifen-associated symptoms, somewomen may have been prescribed estrogen therapy, adding yet another possibleconfounder to the situation. Although previous breast cancer is generallyconsidered a contraindication for ERT, I know of several gynecologistswho have chosen to ignore this contraindication and who continue to prescribeoral estrogens for estrogen- deficiency symptoms. Thus, it can be expectedthat the case-control study, like the cohort study, will be somewhat biasedunless extreme care is taken in selection of cases and controls. Specificcase-control studies will be examined in detail later in this article.

Similarly, since it is an important epidemiologic principle that casesshould be selected without regard for exposure, any case-control studyoriginating with a series of endometrial cancer cases appearing after tamoxifentreatment will be biased, no matter how subsequent controls are identified.For that reason, a series of tamoxifen-exposed endometrial cancer patientsshould not be converted into a case-control "study" by selectinga group of controls to complement exposed cases already collected. Theconversion of a set of cases known to be tamoxifen-exposed into a case-controlstudy indicates that the investigator is testing the hypothesis on thedata which generated it, a clear violation of statistical principles.

Problems of ascertainment of appropriate cases carry over to the selectionof controls, since this group should resemble the cases in as many waysas possible, except for tamoxifen exposure.

In both cohort and case-control studies, the biases discussed aboveall tend to increase the apparent relative risk of endometrial cancer dueto tamoxifen. These biases could thus create either a spurious elevationof the relative risk or conceal any protective effect that would have beenreflected as reduction in relative risk. Surveillance and ascertainmentbiases can be overcome only by using populations in which the comparisongroup (for cohort studies) or control group (for case-control studies)has a frequency and intensity of gynecologic examinations that equal thetamoxifen-treated cohort or the cases in a case-control study.

A further problem for any epidemiologic study is the issue of latency.One must be careful to define latency, since to some it may mean time fromexposure to tumor initiation, and to others the time to clinical appearance.For the epidemiologist, the latter definition is the usual, but it doesallow for several phases of the tumor, including tumor initiation, promotion,and, finally, sufficient growth to create symptoms. Although the latencyperiod for induction of endometrial cancer is unknown, experience withother cancers and other substances suggests that as many as five yearsor more may be needed for cancer to develop and be detected.

It is still possible, however, that tamoxifen could either promote tumorgrowth or cause symptoms (eg, bleeding) from a pre-existing tumor, thusshortening latency. This issue will be addressed with reference to specificstudies.

Although tamoxifen has undergone many clinical trials, none of its publishedstudies give mortality rates for endometrial cancer, and few trials reporton incidence of endometrial or other non-breast cancers. This leads oneto suspect publication bias, wherein investigators in trials with an excessof endometrial cancer publish, while those without do not.

Clinical Trial Cohorts

The National Surgical Adjuvant Breast and Bowel Project B-14

This project (also known as NSABP B-14), described in several publications,involves 2,843 breast cancer patients randomly assigned to either tamoxifen(20 mg/day) or placebo from January 1982 to January 1988.[11,13] An additional1,220 women registered to receive the drug without randomization. Trialpatients were followed an average of eight years and registered patientsan average of five years. The end date for the follow-up has not been published.

For endometrial cancer, the authors report an average annual incidencerate of 1.7/1000 in the tamoxifen group, compared to a rate of zero forthe placebo group. The lack of any cases in the placebo group preventsany computation of relative risk, and throws into question the usefulnessof this reference group. Recognizing this, the authors compared the tamoxifengroup against two other populations--the NSABP B-06 placebo group and SEERdata, which cover about 10 percent of the US population.

On the basis of those comparisons, the relative risk for tamoxifen-treatedwomen is 2.3 (NSABP B-06) or 2.2 (SEER). In considering the SEER comparison,it should be remembered that breast cancer itself confers a relative riskfor endometrial cancer of 1.7 over a 20-year period, with most of the excessincidence in the first five years after breast cancer diagnosis.[7] Thus,there is little, if any, increase in endometrial cancer incidence overthat expected.

The original relative risk of 7.5 reported in this study was based upon15 cases of endometrial cancer, compared to the original placebo group.Of the 15, however, two were sarcomas, and one other was found not to becarcinoma on pathology review. Of the 12 true endometrial cancers, onenever received the drug and five others had fewer than three years of latency.Results for the registered group were similar to those for the clinicaltrial.

Within the randomized trial, there were five deaths attributed to endometrialcancer among the tamoxifen group with no such deaths in the placebo group.The authors give a detailed account of the errors in pathology in somecases, and of the total lack of tamoxifen exposure in one of the five.There is no statistical analysis of mortality from endometrial cancer.

As with the other trials, this study was not designed to examine therelationship of tamoxifen to endometrial cancer. The authors point outthat the tamoxifen group had more gynecologic attention than the placebogroup, creating both ascertainment and surveillance biases. Although thereis evidence from the cancer cases that some placebo women took the drugand some tamoxifen women never received the drug, the extent of this breakin randomization is not known for the great majority of participants.

There is incomplete information concerning use of exogenous estrogensbefore, during, and after tamoxifen treatment. To control for the effectof age, it appears that the randomization was stratified only on the basisof above and below age 50 at time of entry into the trial. This may betoo crude a stratification for a disease as age-dependent as endometrialcarcinoma.

In summary, the alternative relative risk of about two, as reportedby Fisher et al, is of a magnitude that can be explained by increased gynecologicsurveillance which increased the probability of diagnosis, as well theeffect of confounding by other variables such as age, and especially byERT, for which the authors had incomplete data.

Because the higher reported relative risk of 7.5 is based upon an intention-to-treatanalysis, and has the other problems discussed above, that estimate shouldprobably be disregarded.

The Stockholm Study

The Stockholm study results were originally presented in three papersby Fornander et al.[12,18,19] Postmenopausal, postsurgical patients wererandomized to either two years of 40 mg daily tamoxifen or no tamoxifen.From November 1976 to September 1986, 931 patients were allocated to thetamoxifen group and 915 to the control group. The patients also receivedeither radiotherapy or adjuvant CMF chemotherapy comprised of cyclophosphamide,methotrexate, and 5-fluorouracil.

In the 1989 report of follow-up through May 1987, the authors reporteda risk ratio of 6.4 for uterine cancers in the tamoxifen-treated group,based on 13 cases in the treated group, and two cases in the control group.

In their 1991 report of this trial, with slightly shorter follow-up(to January 1987), the authors report similar results for endometrial cancer(with minor discrepancies) and the important finding that tamoxifen-treatedpatients had a relative risk of 3.2 for admission to hospital for benigngynecologic disease other than prolapse or bleeding. This finding indicatesthat gynecologic symptoms were much more common in the tamoxifen group,leading to a greater likelihood of diagnosis of any previously silent endometrialcancer.

In 1993, the authors reported on a follow-up through 1990. The numberof endometrial cancer cases with tamoxifen treatment was now 17, comparedto five among those not receiving tamoxifen. Pathology review of the 17,however, found one patient with a mixed mesodermal tumor, not endometrialcancer. The findings translate to a relative risk of about three. However,of the 16 endometrial cancers among the tamoxifen-treated group, five werediagnosed in the first two years and 10 within the first three years.

In the series of Fornander reports, a nonsignificant increase in cancermortality was reported, with at least three deaths attributed to endometrialcancer in the tamoxifen group. The data presented are inadequate for aproper examination of the effect of tamoxifen on mortality.

Although the Fornander et al results are widely interpreted that tamoxifenincreases endometrial cancer incidence with a relative risk of about six,closer examination of the published data does not support this conclusion.Rather, the magnitude of risk is about three, similar to the risk of hospitaladmission for benign gynecologic conditions after tamoxifen treatment.Furthermore, most of the endometrial tumors occurred within the first fewyears, with insufficient latency to attribute carcinogenesis to tamoxifen.The data suggest one or more of four alternatives:

• Tamoxifen treatment increases the likelihood of diagnosis of pre-existingendometrial cancer by greater diagnostic activity (ascertainment bias);
• Tamoxifen increases the likelihood of uterine bleeding from pre-existingendometrial cancers;
• Tamoxifen greatly accelerates tumor initiation;
• Tamoxifen greatly accelerates tumor growth.

There are several caveats to any interpretation of the Fornander etal reports. The clinical trial was never designed to examine the endometrialcancer issue. The authors present no data concerning the major potentialconfounder of ERT. It is clear that some of the group assigned to tamoxifennever began treatment. Although the authors are certain of the tamoxifen/nontamoxifenstatus of patients diagnosed with endometrial cancer, there is no reportof any validation of the treatment status of the other women in the trial.Thus, the denominators of the tamoxifen exposed and unexposed are in doubt.Finally, there is the issue of reporting. It may be that endometrial cancerin tamoxifen-treated women is more likely to be reported (to a registryor to trial investigators) than cancer in cases without such treatment.

A subsequent report of this study by Rutqvist, as part of a meta-analysis,will be discussed later in this article.[20]

The Danish Trial

The Danish trial had three treatment arms;[21,22] after primary surgery,low-risk patients (n = 1,828) were given no further therapy, while thehigh-risk group was randomized into either radiotherapy alone (n = 846)or radiotherapy with tamoxifen (n = 864). Maximum follow-up was 12 years.

Among the low-risk group, 11 women developed endometrial cancer, fora Standardized Incidence Ratio (SIR) of 1.1. The radiotherapy only grouphad two cases, for an SIR of 0.6, while the tamoxifen group had seven caseswith an SIR of 1.9. None of the SIRs was statistically significant whencompared either to the Danish population or to each other.

The average latency for the low-risk group was 4.3 years after surgery,and for the radiotherapy plus tamoxifen group, only 2.3 years. This disparityin the time since surgery indicates that ascertainment bias may well havebeen operating. Among the tamoxifen-treated, all endometrial cancers occurredwithin the first five years, with six of the seven occurring in the firstfour years.

The Danish trial had problems with small numbers and lack of informationconcerning confounding variables. The tamoxifen group was also slightlyolder. Because the groups were compared to the Danish population for calculationof SIRs, both surveillance and ascertainment biases are probable.

Despite indicating a risk ratio of 3.3 (NS), the most recent data fromthis study offer little or no support for a relationship between tamoxifenand endometrial cancer.

The South Swedish Study

Problems with this study were described even before the results werepublished.[23,24] After radical mastectomy, 719 postmenopausal breast cancerpatients were randomized into three treatment arms: radiotherapy alone,radiotherapy plus tamoxifen, or tamoxifen alone. There were two, four,and five cases of endometrial cancer, respectively. Although the authorspresent no statistical analysis, they comment that the incidence of endometrialcancer in the three groups did not differ to a statistically significantdegree.

The Rutqvist et al Re-Analysis

In their 1995 re-analysis of three previous studies, Rutqvist et alcombined results from the Stockholm, South Swedish, and Danish studies.[20]The study received editorial criticism on statistical grounds, especiallyconcerning the possible relationship of tamoxifen to risk of colorectalcancer.[25]

Table 1 displays the numbers publishedfor the Stockholm trial, at various times. The size of this trial has beenconsistently increasing, because of the continued acquisition of patientsuntil 1990. Rutqvist et al, however, give no data concerning the latencyof the endometrial cancers, nor do they discuss pathologic findings. Thereis no explanation as to why the total number of endometrial cancers inthe controls appears to have fallen over time. One of the differences withthe previous Fornander et al analyses may be because Rutqvist et al havenow used an intention-to-treat analysis, which is a dubious technique whensecond primary cancers are the concern.

In using the data from the Danish trial, Rutqvist et al omitted anyconsideration of the 1,828 patients who received no treatment at all.[21,22]As discussed above, the radiation therapy-only group had a low endometrialcancer incidence (SIR = 0.6) while the untreated group had an SIR of 1.1.By selecting the lower incidence group as their referent population forcalculation of relative risks, Rutqvist et al artificially inflated therelative risk calculation for tamoxifen.

The third study included by Rutqvist et al, the South Swedish BreastCancer Study, has many well-documented problems of patient compliance andmorbidity and mortality reporting. [23,24] For these reasons, it is probablypreferable not to include this study in any attempted meta-analysis.

In summary, Rutqvist et al attempted a meta-analysis of three differentstudies, using numbers that do not match previous publications. Each studyhas, in and of itself, various methodological difficulties. Finally, thereare severe statistical problems as outlined by Simon.[25]

The International Breast Cancer Study

This study followed 320 postmenopausal patients, 167 of whom receiveda combination of tamoxifen with prednisone, while the remaining 153 wereonly observed.[26] The study reported a mean observation period of 96 months.None of the patients (treated or untreated) had an endometrial cancer appearduring the observation period.

The Edmonton-Toronto Study

In 1991, a group of Canadian investigators published an abstract describingthe results of a tamoxifen trial.[27] Of 1,874 women treated with tamoxifen,there were two sarcomas and one endometrial carcinoma of the uterus. Inthe 8,201 patients without tamoxifen treatment, there were 26 adenocarcinomasand two sarcomas. This gives a relative risk of 0.47 (CI = 0.14-1.54) forendometrial cancer.

The Christie Hospital Clinical Trial

Ribiero and Swindell reported on 10 years of follow-up of a clinicaltrial in which premenopausal postsurgical patients received either radiationor tamoxifen while postmenopausal patients received either tamoxifen orno further treatment.[28] Patients were randomized after stratificationby menopausal status, with 481 receiving tamoxifen and 480 receiving notamoxifen.

After 10 years of follow-up, each of the two groups had one case ofendometrial cancer. The small numbers of endometrial cancers limit thepower of this study, but the data fail to support any association betweentamoxifen and endometrial cancer.

The Scottish Trial

Stewart reported a set of four trials involving 1,312 patients, of whom747 received tamoxifen, about half as short-term therapy and half as long-termadjuvant treatment.[29] Each group recorded one new case of endometrialcancer. Although it is not clear as to how rates were calculated, the authorreported no increase in endometrial cancer incidence.

A follow-up to the Scottish trials was published as an abstract indicatingan increased risk of uterine (site not otherwise specified) cancer after10 years of tamoxifen therapy.[30] This information is difficult to interpretbecause of the paucity of data supplied within the abstract. There is noinformation concerning either the pathology of the tumors or the ERT statusof the women.

The Breast Cancer Adjuvant Chemo-Hormone Therapy Cooperative Group(GROCTA) Trial

This trial followed 504 patients, age 35 to 60 years, randomized intothree groups: tamoxifen, chemotherapy, or tamoxifen plus chemotherapy.[31,32]The patients were followed for a median period exceeding 60 months. Onlyone woman in the combination therapy group developed endometrial cancer.Gynecologic symptoms, however, were significantly more frequent in tamoxifen-treatedwomen.

The Eastern Cooperative Oncology Group (ECOG) Study

This publication consists of 168 postmenopausal women given either tamoxifenor placebo, followed for a median of 10 years.[33] There were 83 in theplacebo group and 85 in the tamoxifen group. One endometrial cancer occurredin each group.

The Royal Marsden Chemoprevention Trial

Although this is a pilot study for a massive chemoprevention study,it does provide data for 2,012 women receiving tamoxifen from October 1986to June 1993.[34] Mean follow-up was 35 months. Unfortunately, the authorsdo not give numbers for the incidence of endometrial carcinoma in the trialand predict that results will be available for determining that risk onlyafter the year 2002. To date, however, they have seen no evidence thatwould deter them from continuing the trial and expanding it to a largerstudy. The study did find a higher incidence of fibroids with tamoxifen,but there was no significant increase in the hysterectomy rate.

Clinical Studies

The Neven Prevalence Study

Neven et al performed hystero- scopies on 14 patients taking tamoxifenand 42 controls.[35] Both cases and controls had abnormal postmenopausalbleeding. Three controls and one case were found to have endometrial cancer.Although numbers are small, the study demonstrates an identical prevalenceof endometrial cancer in the two groups, and does not support the elevatedrisk reported in the Fornander and Fisher studies. However, the authorsnoted that the tamoxifen group had an increased incidence of endometrialpolyps and hyperplasia, both conditions that could lead to bleeding and,hence, more investigation, had they been left to become symptomatic.

The Cleveland Study

Hubay et al have reported the results of a randomized controlled trialcomparing tamoxifen plus chemotherapy to chemotherapy alone.[36] Of 311women receiving tamoxifen and followed for eight years, only one case ofnonfatal endometrial cancer was reported. No cases occurred in the chemotherapygroup.

The Athens Series

In 1992, Potamianou et al published a five-year follow-up of 288 breastcancer patients treated with tamoxifen and 108 not treated with tamoxifen.[37]They reported no cases of endometrial cancer in either group. Althoughthis is not a randomized clinical trial, and there is no information concerningwhy women were or were not given tamoxifen treatment, there appears tobe no increase in risk from tamoxifen.

The Lahti Cross-Sectional Study

Lahti et al performed hysteroscopies on 103 asymptomatic breast cancerpatients, 51 of whom had been treated with tamoxifen.[38] One of the tamoxifenpatients had endometrial carcinoma, as compared to two not exposed to thedrug. Like the study by Neven et al, this paper demonstrates that tamoxifen-treatedwomen do not appear to have a higher rate of endometrial cancer when thecontrols are subjected to the same intensity of investigation, such ashysteroscopy. The tamoxifen group did have a statistically significantincrease in the prevalence of endometrial polyps.

The Hardell Study

In a series of letters to the editor, Hardell described a collectionof cases of endometrial cancer in women with a history of breast cancer.[39-41]After the first publication, he selected a number of controls to attemptto change the case series into a case-control study. Because the caseswere first selected due to their tamoxifen exposure, the approach is invalidas a case-control study. Even so, exposure to tamoxifen alone (withoutpelvic irradiation) produced no significant elevation in the odds ratio(OR = 2.6, CI = 0.7-9.6).

Case-Control Studies

The Netherlands Case-Control Study

In 1994, vanLeeuwen et al published their case-control study of endometrialcancer following breast cancer.[42] Using registries, they identified 98such patients. They then selected 385 controls, matched on the basis ofage, year of breast cancer diagnosis, and survival time with an intactuterus. Using multivariate analysis, the authors reported a relative riskfrom tamoxifen treatment of 1.3 (CI = 0.7-2.4). Although there was no significantelevation in relative risk for any particular cumulative dose or durationof treatment, both duration and cumulative dose showed a significant trend(P = 0.049 and p = 0.046, respectively). For both duration and cumulativedose, the highest relative risk was 3.0 (based on three cases) and wasnot statistically significant. In both instances, the lower quartiles oftamoxifen exposure gave relative risks of less than unity.

Clearly, this study is open to differing interpretations, and can beconsidered compatible with either no effect or a mild effect. Althoughthe authors inserted several control variables, including ERT, the numbersare small for calculating dose-response relationships, especially in theabsence of any significant relative risk for any single dose or duration,or ever/never treated. Although it is a case-control study based on registrycases, there is still a possibility of ascertainment bias because of theincreased gynecologic symptoms and diagnostic testing among the tamoxifen-treatedwomen. For that reason, minor elevations of relative risk are compatiblewith ascertainment bias.

The Seattle Case-Control Study

In 1995, Cook et al published a population-based, nested case-controlstudy from a cohort of women diagnosed with breast cancer between 1978and 1990.[43] From the cohort, they were able to use 36 cases of endometrialcancer, that were then matched with 66 controls from the same cohort, basedon the following criteria: year of breast cancer diagnosis, age, and stageof disease. Potential controls were disqualified if they had a previoushysterectomy. Through review of previous medical records, the study teamalso gathered data on several variables, such as ERT, family history, andother risk factors. The authors also went to considerable lengths for qualitycontrol, including verification of tamoxifen use.

The relative risk for any tamoxifen use was 0.6 (CI = 0.3-1.9), andranged from 0.2 (CI = 0.1-1.04) for women using tamoxifen for more than12 months, to 1.6 (CI = 0.4-7.0) for less than one year of use. Clearly,there was no excess risk associated with tamoxifen use.

The Cook study, like the vanLeeuwen study, has the potential for ascertainmentbias, due to more frequent gynecologic exams among users of tamoxifen.Thus, even the relative risk of 0.6 may be biased upward, and it is possiblethat a significant protective effect of tamoxifen (against endometrialcancer) might exist and be hidden by the bias.

Although the numbers are not large, the study has considerable power.Using a = 0.05, the study has an 80% power to detect a relative risk of2.5 or greater.

The authors point out that their follow-up (latency) for many of thewomen is not adequate to rule out a long-term effect. However, since theother studies that conclude that tamoxifen increases endometrial cancerrisk are based upon the early-onset cases, the Cook study, with its carefulconsideration of the possible confounders (especially ERT), would seemto offset these earlier reports.

The Wilfred Hall Cohort

Although Robinson et al analyze and present this as a case-control study,the data collection is really more like a cohort study.[44] The authorsassembled, from medical records, a cohort of 1,017 breast cancer patientsdiagnosed and treated between 1978 and 1989. The patients were not partof any formal clinical trial. Of the cohort, 586 patients were judged tohave adequate records, a uterus in situ, and primary breast cancer. Ofthat group, 108 had received tamoxifen in varying doses, and 478 appearedto have had no tamoxifen. After an average of nine years of follow-up (range5 to 16 years), four endometrial cancers had occurred in each group.

The most appropriate analysis would have been to use incidence rateswith person-years as a denominator. Instead, the authors treated the entiregroup as a large case-control study. Although the crude relative risk calculatesout to 4.43 (CI = 1.12-17.42), the authors claim that by using a Mantel-Haenzelstatistic, they find an odds ratio of 15.2 (CI = 2.8-84.4). This dramaticincrease in the odds ratio by stratifying on the variables of hypertensionand diabetes is remarkable. Also remarkable is the authors' decision notto stratify on age, since the tamoxifen patients were an average of fiveyears older than the others. There is no information given as to latency(since breast cancer diagnosis) of the endometrial cancers, although meanduration of treatment was about 37 months. The study excluded from theanalysis women who took tamoxifen for less than one year.

This study is surprising in many respects. Although the authors gatheredconsiderable data on several variables such as diabetes, hypertension,and family history, they make no mention of ERT prior to breast cancer.They have chosen to ignore the opportunity to perform a cohort analysisand instead have used a case-control approach. The results published (relativeodds of 15.2) do not approximate the relative risk of 4.43 that can becalculated from their Table 3. Finally, this cohort is not a randomizedclinical trial and we know nothing of the factors that influenced the clinicians'decision to treat or not to treat with tamoxifen. For all of the abovereasons, this study should not be considered particularly useful in evaluatingthe tamoxifen-endometrial cancer relationship.

The Lyon Case-Control Study

Recently, Sasco et al published the results of a case-control studyconducted in Lyon.[45] Earlier, some of the same authors had publisheda series of cases of breast cancer with subsequent endometrial cancer aftertamoxifen therapy.[46] In the earlier paper, they mentioned that more definitivestudies were now under way. If the original 20 cases described by Mignottewere included in the study reported in 1995, there is a likelihood of severebias by using the cases with known tamoxifen exposure as a starting point.As the authors point out, the source of patients in the case-control studywas a combination of a registry and data from participating hospitals inthe region. There is no mention in the 1996 paper that the patients fromthe original case series were excluded.

The study is composed of 43 cases of endometrial cancer diagnosed atleast one year after diagnosis of breast cancer, with 177 controls matchedon age, region, year of breast cancer diagnosis, and survival from breastcancer. There is no mention of ERT.

The study reports an odds ratio of 1.4 (CI = 0.6-3.50) for ever-useof tamoxifen, with a dose of over 20 mg/day giving a slightly lower oddsratio than 20 mg or less. Castration, on the other hand, produced an oddsratio of 6.3 (CI = 1.5-26.1). A multivariate analysis increased the tamoxifenodds ratio to 2.8 (CI = 0.69-17.0) for five years or fewer and 4.4 (CI= 0.68-27.9) for more than five years. The multivariate analysis also reducedthe castration odds ratio to 4.2 (CI = 0.57-30.8).

The mild and not statistically significant elevations of relative oddsfrom tamoxifen treatment are compatible with both ascertainment bias (seeearlier discussions) and/or bias due to an inclusion of the cases thatprompted the study. There is also the possibility of more ERT in the tamoxifen-treatedwomen, although this variable was not discussed. The study does not providestrong evidence for a causal link between tamoxifen and endometrial cancer.

Discussion

In assessing a potential causal relationship between tamoxifen and endometrialcancer, it is important to look at the epidemiologic strength of the reportedassociation, its consistency, and whether the finding could be explainedby some factor other than tamoxifen. Other criteria for causality includedose-response and biologic plausibility.

Table 1 indicates that the relativerisks reported range from 0.47 (protective) to infinity. Most risks clusternear unity or under three, especially if appropriate controls are used.Epidemiologists would refer to this as a mild elevation in risk, at most.The studies by vanLeeuwen et al[42] and Cook et al[43] are probably thebest-designed studies, since they were able to control for the potentialconfounding effect of ERT. These two studies, with overlapping confidencelimits for odds ratio calculations, have compatible findings of no significantexcess relative risk for persons ever treated with tamoxifen. The studiesdiffer, however, in the question of dose-response, and there is no immediateexplanation for the differences.

As a group, the Table 1 study resultsare inconsistent, with most trials not replicating the increased risksreported by Fornander and Fisher, and none of the case-control studiesfinding any significant increase in risk.

For this question, there is a high probability that most (includingcase-control) studies suffer from confounding and from bias of varioussorts, as discussed above. The single greatest problem appears to be theincreased gynecologic surveillance of tamoxifen-treated women. This increasein surveillance (due to symptoms or concern over cancer) will almost certainlylead to a higher degree of endometrial cancer ascertainment among tamoxifen-treatedwomen. Both of the large studies that report a significant increase inrelative risk note that treatment and hospitalization for gynecologic symptomswere more frequent in the tamoxifen-treated group.[12,13]

In addition to the ascertainment and surveillance biases previouslydiscussed, only two studies incorporated any variable concerning ERT, whichmay be more frequent in the tamoxifen-treated women because of the menopausalsymptoms it induces.[42,43]

Because the clinical trials were never designed to answer the endometrialcancer question and because those trials all suffer from inadequate dataconcerning confounders (and even treat- ment compliance), they are of limiteduse in assessing the possible relationship.

The case-control studies have the advantage of being able to collectmore data concerning ERT and other possible confounding variables. Becausethe vanLeeuwen and Cook studies can be interpreted as either contradictoryor similar, they await confirmation, hopefully by studies with larger numbers.

Dose-response has been examined in only a few of the trials. The vanLeeuwenstudy found a significant trend with dose, but no statistically significantelevation in relative risk. The Cook study found lesser risk with higherdose.

The greatest lack of biologic credibility comes from the finding thatmany of the alleged tamoxifen-related cancers appeared within the firstfew years of starting therapy. Because of the probability that some ormany of the endometrial tumors preceded the tamox- ifen treatment, thereare serious questions concerning the temporal association of the tumorswith the drug. From the data, it is impossible to tell whether increasedrelative risks are due to the effect of increased surveillance, symptom-inspiredinvestigation, promotion of bleeding in existing tumors, stimulation ofa premalignant condition, or a combination of the above. It seems highlyunlikely that the results to date reflect tumor initiation by tamoxifen.

Since the epidemiologic evidence fails to meet conventional criteriafor association, the fairest conclusion is the Scottish verdict of "notproven" until further evidence is available. One would hope that futurestudies will be able to avoid or control for the severe confounding presentin many of the studies reported thus far.

References:

1. Key T, Pike MC: On the age-dependent association between cancer ofthe breast and endometrium. Br J Cancer 56:704-705, 1987.

2. Kelsey JL, LiVolsi VA, Holford TR, et al: A case-control study ofcancer of the endometrium. Am J Epidemiol 116:333-342, 1982.

3. Henderson BE, Ross RK, Pike MC, et al: Endogenous hormones as a majorfactor in human cancer. Cancer Res 42:3232, 1982.

4. Henderson BE, Casagrande JT, Pike MC, et al: The epidemiology ofendometrial cancer in young women. Br J Cancer 47:749, 1983.

5. Schoenberg BS: Multiple primary malignant neoplasms. The Connecticutexperience, 1935-1964. Berlin, Springer-Verlag, 1977.

6. Kelsey JL, Hildreth NG: Breast and gynecologic cancer. Epidemiology.Boca Raton, Florida, CRC Press, 1983.

7. Adami HO, Krusemo UB, Bergkvist L, et al: On the age-dependent associationbetween cancer of the breast and of the endometrium. A nationwide cohortstudy. Br J Cancer 55:77-80, 1987.

8. Koumantaki Y, Tzonou A, Koumantaki E, et al: A case-control studyof cancer of endometrium in Athens. Int J Cancer 43:795-799, 1989.

9. Koss LG, Schreiber K, Oberlander SG, et al: Detection of endometrialcarcinoma and hyperplasia in asymptomatic women. Obstet Gynecol 64:1-11,1984.

10. Assikis VJ, Jordan VC: Gynecologic effects of tamoxifen and theassociation with endometrial carcinoma. Int J Gynecol Obstet 49:241-257,1995.

11. Fisher B, Costantino J, Redmond C, et al: A randomized clinicaltrial evaluating tamoxifen in the treatment of patients with node-negativebreast cancer who have estrogen-receptor-positive tumors. N Engl J Med320:479-484, 1989.

12. Fornander T, Rutqvist LE, Cedermark B, et al: Adjuvant tamoxifenin early-stage breast cancer: effects on intercurrent morbidity and mortality.J Clin Oncol 9:1740-1748, 1991.

13. Fisher B, Costantino JP, Redmond CK, et al: Endometrial cancer intamoxifen-treated breast cancer patients: findings from the National SurgicalAdjuvant Breast and Bowel Project (NSABP) B-14. J Natl Cancer Inst 86:527-537,1994.

14. Baum M, Cuzick J: Adjuvant treatment with tamoxifen. Side effectsmay have been overstated. Br Med J 312:1036, 1996.

15. Ewertz M, Machado SG, Boice JD, et al: Endometrial cancer followingtreatment for breast cancer: a case-control study in Denmark. Br J Cancer50:687-692, 1984.

16. Hulka BS: Replacement estrogens and risk of gynecologic cancersand breast cancer. Cancer 60:1960-1964, 1987.

17. Hulka BS: Links between hormone replacement therapy and neoplasia.Fertil Steril 62:168S-175S, 1994.

18. Fornander T, Cedermark B, Mattson A, et al: Adjuvant tamoxifen inearly breast cancer: occurrence of new primary cancers. Lancet 1:117-120,1989.

19. Fornander T, Hellstrom AC, Moberger B: Descriptive clinicopathologicstudy of 17 patients with endometrial cancer during or after adjuvant tamoxifenin early breast cancer. J Natl Cancer Inst 85:1850-1855, 1993.

20. Rutqvist LE, Johansson H, Signomklao T, et al: Adjuvant tamoxifentherapy for early stage breast cancer and second primary malignancies.J Natl Cancer Inst 87:645-651, 1995.

21. Andersson M, Storm HH, Mouridsen HT: Incidence of new primary cancersafter adjuvant tamoxifen therapy and radiotherapy for early breast cancer.J Natl Cancer Inst 83:1013-1017, 1991.

22. Andersson M, Storm HH, Mouridsen HT: Carcinogenic effects of adjuvanttamoxifen treatment and radiotherapy for early breast cancer. Acta Oncol31:259-263, 1992.

23. Persson N, Adami HO, Lindgren A, et al: Reliability of endometrialcancer diagnoses in a Swedish cancer registry--with special reference toclassification bias related to exogenous estrogens. Acta Path MicrobiolImmunol Scand 94:187-194, 1986.

24. Ryden S, Ferno M, Moller T, et al: Long-term effects of adjuvanttamoxifen and/or radiotherapy: the South Sweden Breast Cancer Trial. ActaOncol 31:271-274, 1992.

25. Simon R: Discovering the truth about tamoxifen: problems of multiplicityin statistical evaluation of biomedical data. J Natl Cancer Inst 87:627-629,1995.

26. Castiglione M, Gelber RD, Goldhirsch A: Adjuvant systemic therapyfor breast cancer in the elderly: competing causes of mortality. J ClinOncol 8:519-526, 1990.

27. Champion PE, Nabholtz JM, Jenkins G, et al: Is tamoxifen increasingthe risk of uterine malignancy in women treated for breast cancer? BreastCancer Res Treat 19:195, 1991.

28. Ribeiro G, Swindell R: The Christie Hospital adjuvant tamoxifentrial. Natl Cancer Inst Monogr 11:121-125, 1992.

29. Stewart HJ: The Scottish trial of adjuvant tamoxifen in node-negativebreast cancer. Natl Cancer Inst Monogr 11:117-120, 1992.

30. Stewart HJ: Adjuvant tamoxifen duration in a randomized trial. Breast4:256, 1995.

31. Boccardo F, Rubagotti A, Amoroso D, et al: Chemotherapy versus tamoxifenversus chemotherapy plus tamoxifen in node-positive, oestrogen-receptorpositive breast cancer patients: An update at 7 years of the 1st GROCTAtrial. Eur J Cancer 28:673-680, 1992.

32. Boccardo F, Amoroso D, Rubagotti A, et al: Prolonged tamoxifen treatmentof early breast cancer: The experience of the Italian cooperative groupfor chemohormonal therapy of early breast cancer, in Long Term TamoxifenTreatment for Breast Cancer, pp 159-179. Madison: University of WisconsinPress, 1994.

33. Cummings FJ, Gray R, Tormey DC, et al: Adjuvant tamoxifen versusplacebo in elderly women with node-positive breast cancer: Long-term follow-upand causes of death. J Clin Oncol 11:29-35, 1993.

34. Powles TJ, Jones AL, Ashley SE, et al: The Royal Marsden Hospitalpilot tamoxifen chemoprevention trial. Breast Cancer Res Treat 31:73-82,1994.

35. Anonymous: Tamoxifen and the uterus and endometrium. Lancet 1:375-376,1989.

36. Hubay CA, Gordin NH, Pearson OH, et al: Eight-year follow-up ofadjuvant therapy for Stage II breast cancer. World J Surg 9:738-749, 1985.

37. Potamianou A, Tsikinis A, Papadiamantis J, et al: Tamoxifen forbreast cancer and endometrial cancer. Breast Cancer Res Treat 28:85, 1993.

38. Lahti E, Blanco G, Kauppila A, et al: Endometrial changes in postmenopausalbreast cancer patients receiving tamoxifen. Obstet Gynecol 81:660-664,1993.

39. Hardell L: Tamoxifen as risk factor for carcinoma of corpus uteri.Lancet 2:563, 1988.

40. Hardell L: Pelvic irradiation and tamoxifen as risk factors forcarcinoma of corpus uteri. Lancet 2:1432, 1988.

41. Hardell L: Tamoxifen as a risk factor for endometrial cancer. Cancer66:1661, 1990.

42. vanLeeuwen FE, Benraadt J, Coebergh JWW, et al: Risk of endometrialcancer after tamoxifen treatment of breast cancer. Lancet 343:448-452,1994.

43. Cook LS, Weiss NS, Schwartz SM, et al: Population-based study oftamoxifen therapy and subsequent ovarian, endometrial, and breast cancers.J Natl Cancer Inst 87:1359-1364, 1995.

44. Robinson DC, Bloss JD, Schiano MA: A retrospective study of tamoxifenand endometrial cancer in breast cancer patients. Gynecol Oncol 59:186-190,1995.

45. Sasco AJ, Ghaplain G, Amoros E, et al: Endometrial cancer followingbreast cancer: Effect of tamoxifen and castration by radiotherapy. Epidemiology7:9-13, 1996.

46. Mignotte H, Sasco AJ, Lasset C, et al: Traitement adjuvant du cancerdu sein par tamoxifene et cancer de l'endometre. Bulletin du Cancer 79:969-977,1992.