© 1999 by Oxford University Press
Journal of the National Cancer Institute, Vol. 91, No. 21, 1829-1846,
November 3, 1999
© 1999 Oxford University Press
SPECIAL ARTICLE |
Weighing the Risks and Benefits of Tamoxifen Treatment for Preventing Breast Cancer
Affiliations of authors: M. H. Gail (Division of Cancer Epidemiology and Genetics), R. Croyle (Division of Cancer Control and Population Science), National Cancer Institute, Bethesda, MD; J. P. Costantino, J. Bryant, University of Pittsburgh, PA; L. Freedman, Bar Illan University, Ramat Gan, Israel; K. Helzlsouer, The Johns Hopkins School of Hygiene and Public Health, Baltimore, MD; V. Vogel, University of Pittsburgh Cancer Institute/Magee Women's Hospital.
Correspondence to: Mitchell H. Gail, M.D., Ph.D., National Cancer Institute, EPS-8032, Rockville, MD 20892 (gailm{at}exchange.nih.gov).
 |
ABSTRACT |
|---|
 Top Abstract I. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
BACKGROUND: In response to findings from the Breast Cancer Prevention Trial that tamoxifen treatment produced a 49% reduction in the risk of invasive breast cancer in a population of women at elevated risk, the National Cancer Institute sponsored a workshop on July 7 and 8, 1998, to develop information to assist in counseling and in weighing the risks and benefits of tamoxifen. Our study was undertaken to develop tools to identify women for whom the benefits outweigh the risks. METHODS: Information was reviewed on the incidence of invasive breast cancer and of in situ lesions, as well as on several other health outcomes, in the absence of tamoxifen treatment. Data on the effects of tamoxifen on these outcomes were also reviewed, and methods were developed to compare the risks and benefits of tamoxifen. RESULTS: The risks and benefits of tamoxifen depend on age and race, as well as on a woman's specific risk factors for breast cancer. In particular, the absolute risks from tamoxifen of endometrial cancer, stroke, pulmonary embolism, and deep vein thrombosis increase with age, and these absolute risks differ between white and black women, as does the protective effect of tamoxifen on fractures. Tables and aids are developed to describe the risks and benefits of tamoxifen and to identify classes of women for whom the benefits outweigh the risks. CONCLUSIONS: Tamoxifen is most beneficial for younger women with an elevated risk of breast cancer. The quantitative analyses presented can assist health care providers and women in weighing the risks and benefits of tamoxifen for reducing breast cancer risk.
|
I. INTRODUCTION |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
The Breast Cancer Prevention Trial (BCPT) was a randomized, placebo-controlled, 6-year study of the effects of tamoxifen (20 mg once daily for up to 5 years) in a population of women at elevated risk of breast cancer. To be eligible for the trial, women had to be 60 years old or to have a projected 5-year risk of invasive breast cancer (IBC) equal to or greater than that of an average 60-year-old woman (1.66%). Women with previous breast cancer or ductal carcinoma in situ (DCIS) were excluded from the trial. Women with lobular carcinoma in situ (LCIS) who had been treated with lumpectomy and radiation were eligible for the study. On April 6, 1998, the National Cancer Institute (NCI) and the National Surgical Adjuvant Breast and Bowel Project (NSABP) released data from the BCPT demonstrating that tamoxifen treatment reduced the risk of incident IBC. Fisher et al. (1) subsequently published data from the BCPT showing a 49% reduction in IBC. Similar reductions were seen in all age groups, in each year of the study, and in women who were eligible by virtue of a baseline diagnosis of LCIS. There was also a 50% reduction in the risk of noninvasive incident breast cancer, and the incidence of fractures, including hip fractures, was reduced in subjects receiving tamoxifen (1).
Unfortunately, some women in the BCPT also experienced life-threatening adverse effects from tamoxifen, including excesses of endometrial cancer, pulmonary embolism, stroke, and deep vein thrombosis. Some excess incidence of cataracts was also noted.
The variety of effects from tamoxifen complicates the decision to use it in prevention. A good decision will depend on the particular risk profile of the woman and on her preferences. The balance of benefits and adverse effects from tamoxifen depends on age and on other factors, such as the woman's underlying risk of breast cancer and whether she has had a hysterectomy. The NCI held a workshop on July 7 and 8, 1998, to develop information to assist women and their health care providers in deciding whether or not to initiate preventive use of tamoxifen (see "Appendix: Workshop Program and List of Participants" section appearing before "References"). While this information would be based on the results of the BCPT whenever possible, workshop participants were also asked to consider the potential benefits and risks for high-risk women who were not eligible for the BCPT, such as women with DCIS.
This special article is based in part on the presentations and discussions at this workshop as well as on additional statistical analyses of risk/benefit indices and additional review of the literature.
Shortly after the workshop, reports from two smaller controlled studies (2,3) appeared that did not demonstrate a reduction in breast cancer incidence from tamoxifen use. Although the risk/benefit tables that we present are based solely on BCPT estimates of tamoxifen's effects, we discuss the impact of including data from other studies.
|
II. PROJECTING RISKS IN THE ABSENCE OF TAMOXIFEN TREATMENT |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
To assess the risks and benefits of tamoxifen, it is necessary to know the incidence rates for certain health outcomes in the absence of tamoxifen. Except for cataracts, we relied on data from outside the BCPT to estimate background incidence rates, although estimates of the effects of tamoxifen on these rates were obtained from BCPT data alone. Risks and benefits were calculated by comparing estimated incidence rates in the presence and absence of tamoxifen. We now review information on the methods used to estimate incidence rates in the absence of tamoxifen.
A. Breast Cancer
1) Invasive Breast Cancer (IBC) The primary end point of the BCPT was IBC. To be eligible for the BCPT, a woman had to be 60 years old or to be 35-59 years old with a projected 5-year risk of IBC of at least 1.66%, which is the rate for an average 60-year-old white woman in the United States. We, therefore, emphasize estimates of 5-year risks of IBC in organizing our risk/benefit tables.
We recommend the method used in the BCPT for calculating the projected 5-year risk of IBC for a woman with particular risk factors but with no history or current evidence of invasive cancer, DCIS, or LCIS. The method is based on the earlier work of Gail et al. (4), who analyzed data from the Breast Cancer Detection Demonstration Project (BCDDP) and developed a model for projecting total breast cancer incidence, including both invasive and in situ lesions. The model of Gail et al. (4) uses the age of the counselee and the following additional risk factors: number of affected first-degree relatives, age at menarche, age at first live birth, number of previous breast biopsies, and the presence of atypical hyperplasia in a biopsy specimen. Benichou (5) wrote a computer program, RISK, that allows one to estimate the absolute risk of developing breast cancer for a woman of any age from 20 to 80 years and with any pattern of these risk factors over any desired time interval. This model was intended for women in a program of annual screening with mammography. Although the model overpredicts risk in young unscreened women (4,6-8), it accurately predicted total breast cancer incidence for women under age 60 years in the placebo arm of the BCPT and underestimated total breast cancer incidence somewhat for women over age 59 years (9). The BCPT protocol specified annual mammography and semiannual breast examinations for all women in the trial.
NSABP statisticians, including Stewart J. Anderson and Carol K. Redmond, modified the model of Gail et al. to project only IBC (10). While retaining the relative risk features of the model by Gail et al., they employed composite age-specific rates of IBC for white women from the Surveillance, Epidemiology, and End Results (SEER) Program1 of the NCI, instead of the composite rates from the BCDDP used by Gail et al., and they modified the factors needed to convert composite rates to baseline rates accordingly. In addition, NSABP statisticians adapted the model to project risks for black women by using age-specific composite SEER rates for black women and by modifying the factors used to convert composite rates to baseline rates. Apart from black women, the same projections were made for all other U.S. women as for whites.
The ratio of age-adjusted rates of IBC comparing Hispanic women with non-Hispanic white women in the United States is 0.60 (11). Because we did not have data on the prevalence of breast cancer risk factors among Hispanic women, we could not modify the IBC risk model for Hispanic women. It is likely, however, that the rates of IBC predicted for white women by the IBC risk model overestimate the correct rates for Hispanic women.
The placebo arm of the BCPT, exclusive of women with LCIS at baseline, affords an
excellent opportunity to test the projections of the modified risk model for invasive cancer for
women in regular screening. The agreement of observed and expected incident invasive cancers is
excellent for all ages (Table 1).
Because only 1.7% of the BCPT
participants were black, however, no independent validation was possible for black women. In
fact, 1.7 cases were predicted among black women, and two cases were observed. Even though
the model predicted well for all ages, there is some evidence that the model underpredicts risk
slightly for those in the lowest quintile of projected risk (observed-to-expected ratio =
0.70; 95% confidence interval [CI] = 0.47-1.11) and overpredicts risk
slightly for those in the highest quintile (observed-to-expected ratio = 1.21; 95% CI
= 0.92-1.64). Costantino et al. (9) provide additional information
on the validity of these projections.
|
A computer program designed to help health care providers estimate the risk of IBC and discuss the use of tamoxifen for chemoprevention with their patients is available through the NCI's Cancer Information Service (telephone No. 1-800-4-CANCER or Web site http://cancerTrials.nci.nih.gov). Alternatively, one can approximate the projected 5-year risk of IBC very well from Table 2
. First one computes a
relative risk for the counselee by multiplying four relative risk factors. Then one multiplies this
relative risk by the appropriate age- and race-specific baseline 5-year risk (Table 2). For example, consider a nulliparous 42-year-old white woman who began
menstruating at age 12 years, who has no affected first-degree relatives, and who has had one
previous breast biopsy with specimens interpreted as benign and no evidence of atypical
hyperplasia. From Table 2, her relative risk is 1.10 x 1.70 x
1.55 x 0.93 = 2.70, and her projected 5-year risk of IBC is 2.70 x 0.450
= 1.2%. The computer program also yields 1.2% in this case. The algorithm
in Table 2 is an approximation because, unlike the computer program, it does not account for
competing causes of death; such competing risks usually produce small effects over 5-year
periods, however.
|
2) Projecting In Situ Breast Cancer Because the proportion of in situ cancers depends on the intensity of screening and because women who take tamoxifen are likely to be in a program of annual screening with mammography, we recommend estimating the ratio of incidence rates of in situ breast cancer to IBC from populations receiving annual screening, such as the placebo arm (excluding those with LCIS) of the BCPT. These incidence ratios were estimated from BCPT data as 0.53 and 0.31, respectively, for the age groups 49 years old or younger and 50 years old or older. We recommend projecting the 5-year risk for in situ breast cancer by multiplying a woman's projected 5-year risk of IBC by the ratio appropriate to her age.
3) Other Strong Risk Factors and Protective Factors The model of Gail et al. (4) and its modification for the BCPT were based on women in the BCDDP with no history of breast cancer or evidence of IBC or DCIS on their initial screening evaluation. Women who have had a previous breast cancer have a risk of developing contralateral disease five times higher than that of the general population (12). Women with LCIS in the placebo arm of the BCPT had a 5-year risk of IBC of 6.47%, which is 3.9 times the risk of an average 60-year-old woman and about twice the risk of other participants on the placebo arm of the BCPT. Risks for women with DCIS are even higher (see section V, part A). Thus, women with a previous breast cancer and women with LCIS or DCIS are at high risk, even compared with the population of high-risk women who entered the BCPT. Women known to carry cancer-causing mutations of the genes BRCA1 or BRCA2 are thought to have a cumulative breast cancer risk to age 70 years in the range 37%-85% (13-15). Such women constitute only about 0.7% of the general U.S. population, however (16,17), although perhaps 2% of Ashkenazi Jewish women are carriers (13). Other rare inherited disorders, such as Cowden's syndrome (18) and the Li-Fraumeni syndrome (19), also carry high risks of breast cancer. A woman from rural Japan or China has only one fifth of the risk of a woman of the same age from the United States (20). Available risk models do not explicitly take these strong risk factors or protective factors into account. Thus, when using available models, a counselor should look for these factors and modify risk estimates accordingly.
4) Alternative Modeling Approaches and Areas of Research Claus et al. (21) used data from the Cancer and Steroid Hormone (CASH) Study to develop a risk projection model for IBC and in situ breast cancer based solely on the counselee's age and on detailed family history, including the ages at onset in affected relatives. The theory underlying this model is that all familial risk is conferred by a single autosomal dominant gene. This model nicely explains why the proportion of mutation carriers is higher in women with breast cancer incident at younger ages than at older ages. Claus et al. (22) have shown, however, that the full extent of familial aggregation cannot be explained by current autosomal dominant models.
There are several opportunities for improving projections of breast cancer risk. One approach is to find and use additional strong risk factors. A promising candidate is the percentage of dense tissue observed in a baseline mammogram. Indeed, studies based on data in the Canadian National Breast Screening Study (23) and in the BCDDP (24) indicate that this factor is an even stronger predictor than family history. It might be possible to make more efficient use of family history information by using it to estimate the probability that a woman is a carrier of a BRCA1 or BRCA2 mutation (25). Direct measurements of cancer-causing gene mutations in BRCA1 or BRCA2 would have an important bearing on risk projections, but such abnormalities are rare, even among women with breast cancer (26).
B. Endometrial Cancer
We base predictions of endometrial cancer incidence rates on age-
and race-specific SEER incidence rates from 1991 through 1995 (Table
3). To predict risk for women with a uterus, the SEER
rates were divided by the estimated age-specific prevalence of having a
uterus, obtained from Fig. 1 in Merrill and Feuer
(27). These prevalences were 0.88, 0.80, 0.72, 0.65, 0.64,
0.63, 0.62, 0.66, and 0.65, respectively, for the 5-year age intervals
35-39, . . . , 75-79. The same prevalence factors were used for
white and black women, since we could find no data specific for black women.
|
|
In discussing the risk of endometrial cancer, counselors should be aware of risk factors that can increase or decrease risk compared with the composite average rates in Table 3
. Important factors, such as long-term use of estrogens unopposed by progestins, are
indicated in Table 4, but readers are urged to consult references (28-31) for details.
|
C. Stroke, Pulmonary Embolism, and Deep Vein Thrombosis
For white women and other nonblack women, the age-specific incidence
rates (Table 3) for stroke, pulmonary embolism, and deep vein
thrombosis are from studies of the predominantly white population of
Rochester, MN. The data for strokes were taken from the period
1975-1984 in Fig. 2 of Broderick et al. (32). Data for
pulmonary embolism and deep vein thrombosis are from the period 1986
through 1990 in Table 1 of Silverstein et al. (33). The
age-adjusted stroke [International Classification of Diseases (ICD)
code Nos. 430-438.9 (34)] mortality rate per 100 000
white women aged 35-84 years in Olmstead County, MN, was 69.8,
compared with a U.S. rate of 92.2. Thus, stroke incidence rates for
white women in Table 3 may underestimate U.S. incidence rates. These
mortality data from 1979 through 1996 were obtained from
http://wonder.cdc.gov and were adjusted to the 1990 U.S. white female
population distribution. Similarly, the age-adjusted mortality rates
for pulmonary embolism (ICD code Nos. 415-417.9) for white women aged
35-84 years were 8.1 for Minnesota, 10.2 for Olmstead County (where
counts were too small to be reliable), and 8.1 for the United States.
Because few studies provide direct information on the incidence rates of stroke, pulmonary embolism, or deep vein thrombosis in black women, we estimated the incidence rates for black women by multiplying rates for white women by a black/white risk ratio. We estimated these ratios for strokes from stroke (ICD code Nos. 430-438.9) mortality ratios computed from Tables 1-27 in Vital Statistics of the United States, 1992 (35). These ratios were 3.4, 2.8, 2.9, 2.3, and 1.2, respectively, for age groups less than 40 years, 40-49 years, 50-59 years, 60-69 years, and 70 years or older. Rates were age-adjusted within age groups to the 1990 U.S. populations for whites and blacks before these ratios were computed. The mortality ratios for pulmonary circulatory failure (ICD code Nos. 415-417.9), which is caused mainly by pulmonary emboli, were used for pulmonary embolism and deep vein thrombosis. These ratios were 2.9, 3.1, 3.0, 2.3, and 1.6 for the previous age groups, respectively.
Several studies confirm higher stroke incidence rates in black women than in white women.
Rosamond et al. (36) presented data from the Atherosclerosis Risk in
Communities (ARIC) cohort of 15 792 individuals aged 45-64 years from Jackson (MS),
Forsyth County (NC), Washington County (MD), and Minneapolis (MN). The black/white stroke
incidence ratios were 2.8 (95% CI = 1.4-5.5) for men and women under age 55
years and 2.2 (95% CI = 1.7-3.0) for older men and women. The overall stroke
incidence ratios were the same in men and women, 2.66. Unpublished combined data from ARIC
and from the Cardiovascular Health Study (CHS), which included 5873 men and women over age
64 years from Allegheny County (PA), Forsyth County (NC), Sacramento County (CA), and
Washington County (MD) (37,38), yielded a black/white incidence ratio
for women aged 65-74 years of 2.4 (95% CI = 1.5-3.9). A study of stroke incidence
in upper Manhattan (NY) (39) yielded black/white ratios of 3.0 for
women of ages 40-59 years and 2.4 for women of ages 60-79 years. All of these data support the
use of the stroke mortality ratios used in Table 3.
It is not surprising that black women in the general population have higher stroke rates than white women because they have a higher prevalence of risk factors such as hypertension (38). Black/white stroke incidence ratios may be smaller in healthier populations, such as women who volunteer for prevention trials like the BCPT or the Women's Health Initiative (WHI) (40). It is noteworthy, however, that, even after adjustment for age, gender, hypertension, diabetes, location, education, smoking, and coronary heart disease (36), the black/white incidence ratio in ARIC was 1.4 (95% CI = 1.0-1.8).
The rates for stroke and pulmonary embolism in Table 3 overestimate
the rates found in the placebo arm of the BCPT, possibly because healthy women tended to
volunteer. Calculating the expected numbers of events from rates in Table 3, we found observed-to-expected ratios of 24/46.1 = 0.52, 6/14.3 =
0.42, and 22/18.5 = 1.19 for stroke, pulmonary embolism, and deep vein thrombosis,
respectively. Combining stroke and pulmonary embolism, we found an observed-to-expected ratio
of 30/60.2 = 0.50. In supplemental analyses to evaluate the sensitivity of our methods and
to advise women who might volunteer for prevention trials, we multiplied the stroke and
pulmonary embolism rates for white women in Table 3 by 0.50.
Table 4 lists factors other than age and race that increase the risk of
stroke, pulmonary embolism, and deep vein thrombosis. See references (41-43) for details.
Pulmonary embolism and deep vein thrombosis can be considered together as venous
thromboembolism because they frequently occur simultaneously and have the same risk factors (43,44). These references provide details on the risk factors shown in
Table 4.
D. Fractures
Estimated rates of fractures for white women (Table 3) of the
proximal femur (hip), vertebra (spine), and distal forearm were
obtained from studies in Rochester (MN) [see Table 1 in
reference (45)]. Rates for Colles' fracture (Table 3) were
obtained by multiplying rates for distal forearm fractures by 0.23,
which was the fraction of distal forearm fractures classified as
Colles' fractures in the BCPT. Silverman and Madison (46)
calculated hip fracture incidence rates for non-Hispanic white women,
Hispanic women, black women, and Asian women from hospital discharge
data in California. The black/white ratios of incidence rates were
0.855, 0.539, and 0.381 for age groups less than 50 years, 50-59
years, and 60 years or more, respectively. We obtained rates for black
women in Table 3 by multiplying rates for white women by these ratios.
Compared with white women, Hispanic and Asian women had ratios 0.419,
0.240, and 0.334 and 0.435, 0.266, and 0.546, respectively, for these
age groups (46). Jacobsen et al. (47) reported
rates
of hip fractures in black and white women over age 64 years not very
different from those in Table 3. Baron et al. (48)
also
reported similar rates of hip fractures to those in Table 3 in a 5%
sample of the U.S. Medicare population aged 65-89 years, both for
white and for black women.
In addition to age and race, other factors influence the risk of fractures in women. Women who have lost 20% of their weight since age 25 years have a 67% increase in hip fracture risk, adjusted for age and other factors [Table 2 in Cummings et al. (49)]. A number of other factors contribute to increased risk, including a history of maternal hip fracture, previous hyperthyroidism, current consumption of long-acting benzodiazapines, anticonvulsants or caffeine, lack of exercise (walking), inability to rise from a chair, previous fracture, and decreased bone density. Cummings et al. show how risk increases with decreasing bone density and with the number of other such risk factors.
E. Cataracts
Baseline estimates of cataract incidence (Table 3) were calculated
from data in the placebo arm of the BCPT because this cohort of women
reflects current ophthalmologic practice and is the largest cohort with
reports on cataracts in women. Cataract incidence was based on
self-reports of cataract diagnoses. Unfortunately, the BCPT yielded
little data on cataracts for black women, and we are unaware of other
data on cataract incidence in black women. Data on race as a risk
factor for lens opacity are inconsistent (50). Therefore, we
used the same rates for black as for white women in Table 3. The rates
in Table 3 are intermediate between the higher rates in studies in
which cataracts are diagnosed solely on the basis of changes detected
by slit-lamp examinations, without any requirement for a decrease in
visual acuity (51-53) and the lower rates in studies that
require, in addition to evidence of lens opacity, a decrease in visual
acuity below 20/30 (54,55).
Certain medical conditions and medications, such as diabetes, oral steroids, and medicines for gout, are associated with a modestly increased risk of lens opacities, as are current smoking and low educational attainment (50,56). Current users of vitamin supplements have a reduced risk of lens opacities (56). Hodge et al. (57) reviewed studies of these and other risk factors.
|
III. EFFECTS OF TAMOXIFEN |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
Fisher et al. (1) described the effects of tamoxifen in the BCPT. Overall relative risks (tamoxifen to placebo) were 0.51 for IBC and 0.50 for in situ breast cancer (Table 5).
There was no statistically significant evidence
of heterogeneity of relative risks of invasive cancer across groups
defined by age, number of affected first-degree relatives, projected
5-year risk of IBC, or LCIS status. We therefore assume that the
relative risk reductions from tamoxifen for IBC and for in situbreast cancer are uniform across all subgroups. The preventive
benefit of tamoxifen appeared to be greater in women with atypical
hyperplasia, but the numbers of events were small among such women, and
this subgroup was only one of five examined (1).
|
BCPT investigators (1) identified eight non-breast cancer conditions whose rates were affected by tamoxifen (Table 5
). We have grouped all of
the outcomes into the categories "life-threatening events," "severe
events," and "other events" and labeled the individual conditions with
numbers 1, 2, . . . 10 (Table 5). Because there was no statistically
significant evidence of heterogeneity of relative risks across age groups for the non-breast cancer
outcomes in Table 5, we used overall relative risks for each of the
non-breast cancer outcomes except for endometrial cancer, for which it can be argued that
menopausal status may influence the effect of tamoxifen. We, therefore, used the relative risk 4.01
found for women older than age 49 years for that age group. Even though there was no
statistically significant evidence of increased risk with tamoxifen for women under age 50 years
(relative risk = 1.21), the numbers were small; therefore, we used the overall relative risk,
2.53, for women under age 50 years. Data from other NSABP trials support this choice (see section VIII).
There is substantial uncertainty about the magnitude of the effects of tamoxifen for some of
the outcomes, indicated by the 95% CIs in Table 5. The relative
risks in Table 5 are based only on BCPT data. These relative risks are
quite consistent, however, with the relative risks associated with adjuvant tamoxifen therapy in
several analyses of the risk of contralateral breast cancer among women with resected breast
cancer (58) and with the summary relative risk, obtained by a
meta-analysis, of 0.53 for recurrent cancer or contralateral primary cancer among women on
adjuvant tamoxifen therapy for 5 years (59). European studies with 41
women who developed breast cancers (3) and with 70 women who
developed breast cancers (2) did not demonstrate a beneficial preventive
effect of tamoxifen. These trials may be sufficiently different from the BCPT in design and
execution that it is not reasonable to combine their data with those of the BCPT (1). Nonetheless, we can obtain a combined estimate of relative risk by taking the ratio
of the number of breast cancers in the tamoxifen arms of the three studies to the number of breast
cancers in the placebo groups. That ratio is (89 + 35 + 19 + 34)/(175 + 69 + 22 + 36) =
0.59 (95% CI = 0.49-0.71). Methods of combining results that account for possible
differences in follow-up time cannot be used with the data presented in the European studies.
Other methods that allow for heterogeneity of treatment effects across the three studies would
place greater weight on the smaller studies (60). In section VIII, we
discuss the impact of using the relative risk estimate 0.59 instead of the relative risks in Table 5.
These effects of tamoxifen were found even though many women discontinued its use during the course of these studies. In the BCPT, 23.7% discontinued treatment with tamoxifen for reasons not specified in the protocol, compared with 19.7% in the placebo group (1), and 40% of women in the study by Powles et al. (2) stopped taking tamoxifen prematurely. It is not known whether the preventive effects of tamoxifen would have been greater had compliance been better.
|
IV. RISK/BENEFIT COMPARISONS FOR WOMEN ELIGIBLE FOR BCPT |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
A. Statistical Methods
Relative risks, such as those in Table 5, do not convey the actual
chance ("absolute risk") that tamoxifen will prevent or cause a
woman to develop an adverse health outcome. We have, therefore,
developed several additional tables to assist in weighing the risks and
benefits from prophylactic tamoxifen use.
One approach is to describe fully the expected numbers of various adverse outcomes in a
population of 10 000 untreated women followed for 5 years (Tables 6 and 7) and the corresponding numbers of events expected to
be prevented (or caused) by tamoxifen (Tables 6 and 8). This full description allows a woman who is mainly concerned about particular
outcomes, such as breast cancer or endometrial cancer, to focus on those absolute risks and the
effects of tamoxifen on them.
|
|
|
To compute the number of invasive breast cancers expected in a population of 10 000 women with a given risk profile over a 5-year period, we multiplied the projected 5-year risk of IBC (in percent) from the breast cancer risk program (see section II, part A.1) by 100 (Table 6
). The corresponding expected number of in situ lesions was obtained by multiplying this number by 0.53 for women under 50 years of age
and by 0.31 for women older than age 49 years (see section II, part A.2; see also Table 6).
To compute the expected number of non-breast cancer events Nx,p of
type x (x = 2, 3, 4, 5, 7, 8, 9, 10 as defined in Table 5) in untreated women (Table 7), we assumed that the
cause-specific hazard rates, Ix, and the mortality rates, Mx, from causes other than cause x are constant over the 5-year period. It follows that
 | (1) |
We used 1990 U.S. mortality rates derived from reference (61). For
non-breast cancer conditions, the rates Ix are obtained by dividing entries in
Table 3 by 1000.
Expected events for a treated population, Nx,t , are obtained in the same
way, except RxIx replaces Ix in equation 1,
where Rx is the relative risk in Table 5. To obtain Ix for this calculation for IBC or in situ lesions, we solved equation
1 with known values of Mx and with Nx,p obtained as
described above from the risk program. The entries in Tables 6 and 8 representing the numbers of events prevented by (positive number) or
caused by (negative number) tamoxifen are simply the differences Nx = Nx,p - Nx,t rounded to the nearest
integer.
To summarize the risks and benefits of tamoxifen in a single number, however, it is necessary
to define indices that assign weights to the various events. A summary index based on the severity
categories in Table 5 can be calculated from
 | (2) |
where weights W1, W2, and W3 are chosen to put varying emphasis on life-threatening, severe, and other events, respectively. We rounded indices to the nearest integer. We rely principally on the index I(1, 0.5, 0) that puts twice as much weight on life-threatening events as on severe events and that ignores other events altogether. We also investigated the robustness of our conclusions to the use of other indices, i.e., I(1, 1, 1), I(1, 1, 0), and I(1, 0.5, 0.25).
In addition to tabulating I(1, 0.5, 0) for various types of women, we analyzed the
random variability of the data to estimate the probability that the index is positive. To compute
these probabilities, we assumed that the only random elements in Nx,p, Nx,t, and equation 2 are the relative risks in Table 5 that
affect the calculation of Nx,t from equation 1 with RxIx in place of Ix. Let U and V be the observed
numbers of a particular adverse event, x, in the tamoxifen and placebo arms,
respectively, of the BCPT. We assume U and V are independent Poisson
variates with means 
U and V and that U and V have independent noninformative
exponential prior distributions with means tending to infinity. Then the posterior distributions of
2U and 2V are independent chi-squared
distributions with 2(U + 1) and 2(V + 1) degrees of freedom, respectively. It
follows that Rx is distributed as {(U + 1)/(V +
1)}{PV /PU}F2(U + 1),2(V + 1), where PU and PV
are the respective person-years of exposure in the tamoxifen and placebo groups. Thus, to obtain
an estimate of the probability that I(1, 0.5, 0) exceeds zero, we resampled Rx independently for each x and recalculated I(1, 0.5, 0). We repeated
this process 1 000 000 times and estimated the desired probability (with
precision ±0.001) as the proportion of samples in which I(1, 0.5, 0) exceeded
zero. Using a parametric bootstrap with resampled Poisson counts instead of the Bayesian
approach above, we obtained similar results.
B. Expected Events in the Absence of Tamoxifen and Numbers of Events Prevented or Caused by Tamoxifen among 10 000 Women Over a 5-Year Period
The numbers of IBCs (N1,p) and in
situ breast cancers (N6,p) expected to
develop in a population of 10 000 untreated women over a 5-year
period depends directly on the projected 5-year risk of IBC (Table 6).
For a given projected 5-year risk of IBC, the expectations for in
situ breast cancer are 0.53/0.31 = 1.7 times higher in women under
50 years of age compared with those 50 years old or older (see
section II).
The numbers of expected non-breast cancer events increase with age and vary by race (Table
7). The variations of the frequency of events reflect the variation in the
expected rates by age and race. Cataracts are by far the most common event. Hip and spine
fractures are rare, and the occurrence of a stroke is infrequent among women under age 50 years,
but these conditions are relatively frequent among those 70 years old or older. Several differences
between white and black women are important to note. In those over 50 years of age, fractures
are about two to three times more common among white women. Depending on age, the
frequency of endometrial cancer is 1.5-2.5 times higher among white women. Stroke, pulmonary
embolism, and deep vein thrombosis are two to three times higher among black women in all age
groups.
The numbers of cases of IBCs (N1) and in situ breast cancers (N6) expected to be prevented by tamoxifen among 10 000 women
over a 5-year period are shown in Table 6. Because tamoxifen reduces the
incidence of breast cancer by about one half, the numbers of breast cancers expected to be
prevented are nearly proportional to the projected 5-year risk of IBC.
Table 8 displays the expected numbers of non-breast cancer outcomes
prevented (positive number) or caused (negative number) by tamoxifen in such a population. The
patterns in Table 8 reflect variations in background rates (Table 7) and the relative risks associated with tamoxifen therapy (Table 5). Among women 50 years old or older, tamoxifen reduces the expected
numbers of fractures substantially. This benefit is counterbalanced by substantial increases in the
risks of stroke, pulmonary embolism, and deep vein thrombosis in women 60 years old or older,
especially among black women. The risk of tamoxifen-induced endometrial cancer is also
appreciable in women 50 years old or older with a uterus and is about twofold higher in white
women than in black women. Tamoxifen causes very few adverse events among black or white
women under age 50 years and has the potential to prevent IBCs and in situ breast
cancers among high-risk women in this age range.
C. Examples of Assessing Risks and Benefits From Tables 6-9
We illustrate how black and white women in the age groups 40-49
years or 50-59 years and with projected 5-year risks of IBC of 2.0%,
4.0%, or 6.0% can assess the net benefit or risk from tamoxifen
(Table 9). Table 9 was constructed by abstracting
appropriate elements from Tables 6 and 8.
|
First consider a 45-year-old white woman with a projected 5-year risk of IBC of 4%. Such a woman would have several strong risk factors for breast cancer (see section II). She can expect a net reduction of absolute risk of 151 life-threatening events per 10 000 women in 5 years or a reduction in her personal risk of 1.51% in 5 years (Table 9
). This reduction in the risk of life-threatening events is large, compared
with the general 5-year probability of death, 0.92%, in this age range. The overall 5-year
mortality can be approximated by multiplying the rates in Table 3 by
0.005 and compared with the background risk of IBC (Table 6) and
non-breast cancer life-threatening events (Table 7). Such a woman can
also anticipate a reduction of 90 severe events (0.90%) and a slight increase in the risk of
other events (22 events or 0.22%). If she is concerned only with life-threatening events, she
might use the index I(1, 0, 0) = 151. If she has no uterus, I(1, 0, 0)
= 151 + 16 = 167. If she weighs life-threatening events twice as heavily as severe
events and ignores other events, she could use the index I(1, 0.5, 0) = 196 (Table
9). This index rises to 212 if the woman has no uterus. If the woman were
55 years old instead of 45 years old, her indices would be less favorable. In particular, I(1, 0, 0) = 15 and I(1, 0.5, 0) = 38, primarily because of increasing risk of
endometrial cancer. If the 55-year-old woman has no uterus, I(1, 0, 0) = 15 +
120 =135, and I(1, 0.5, 0) = 158. A 55-year-old black woman with a projected 5-year risk of IBC of 4% would expect a net adverse effect from tamoxifen among 10 000 such women of -81 life-threatening events in 5 years (0.81% increase in absolute risk), primarily from increased risk of stroke and pulmonary embolism. The index I(1, 0.5, 0) = -74 if she has a uterus and -22 otherwise. The data therefore suggest that such a woman is unlikely to benefit from using tamoxifen. It should be remembered, however, that the data on baseline rates and effects of tamoxifen are much less well established for black women than for white women (see sections II and III); therefore these conclusions are subject to greater uncertainty.
Some generalizations can be made based on I(1, 0.5, 0) in Table 9. First, the net index increases with increasing projected 5-year risk of IBC. Second,
within any particular level of projected 5-year risk of IBC, the net index decreases with increasing
age as the result of increases in the risk of adverse effects of tamoxifen treatment. Third, the
elimination of the risk of endometrial cancer among women 50 years old or older (as in those who
have had a hysterectomy) substantially improves the net index. This is particularly notable among
white women with a low risk of breast cancer, for whom the elimination of endometrial cancer
risk improves the index from a negative value to a positive value. Fourth, because the risks of
events, such as stroke, pulmonary embolism, and deep vein thrombosis, are two to three times
higher in black women, the net indices for black women are lower than those for white women.
D. Some General Recommendations Based on the Index I(1, 0.5, 0) in Tables 10 and 11
These points can be studied further by examining the patterns of
values of I(1, 0.5, 0) for women with (Table
10) and without (Table 11) a
uterus. In addition to the value of I(1, 0.5, 0), both tables
display an asterisk when the probability that I(1, 0.5, 0)
exceeds zero, taking random variation into account (seesection IV, part A), is 0.60-0.89 and two asterisks when the
probability equals or exceeds 0.90. Thus, two asterisks indicate
"strong evidence" that tamoxifen is beneficial based on weights
W1 = 1, W2 = 0.5, and
W3 = 0, and one asterisk indicates "moderate
evidence" of a net benefit.
|
|
The patterns of values of I(1, 0.5, 0) in Tables 10
and 11 suggest certain classes of women who are likely to benefit from
prophylactic use of tamoxifen. The largest positive entry in Table 10 is
413 for a white woman aged 35-39 years with a projected 5-year risk of IBC of 7.0%. If
we think of each severe event as equivalent to half of a life-threatening event, this number
translates to a reduction in the absolute risk of a life-threatening event of 4.13% over a
5-year period, which is large, especially in comparison with the 5-year mortality rate of
0.47%. Many other positive entries in Table 10 are smaller, even
though there is strong evidence, indicated by two asterisks, that the net benefit is positive. For
example, if a 45-year-old white woman with a uterus had a projected 5-year risk of IBC of
1.5%, she would have an expected reduction in absolute 5-year risk of only 0.43%.
Thus, it is important to look at the magnitude of the effects in Tables 10
and 11 as well as whether there is strong statistical evidence that the net
effects are positive, as indicated by the asterisks.
Nonetheless, we can summarize the information in Tables 10 and 11 that identifies women for whom there is strong evidence (two asterisks)
or moderate evidence (one asterisk) that I(1, 0.5, 0) exceeds zero (Fig. 1). Among white women with a projected 5-year risk of IBC between 1.5% and
7.0%, there is strong evidence of a net tamoxifen benefit for all those under age 50 years.
For those with a uterus, strong evidence of benefit is also found for women aged 50-59 years with
a projected 5-year risk of IBC greater than or equal to 6.0%, and moderate evidence of net
benefit is found for those with a projected 5-year risk of IBC in the range of
4.0%-5.9%. For white women without a uterus, strong evidence of benefit is also
found in some high-risk women aged 60-69 years (see Fig. 1).
For black women, strong evidence of a net tamoxifen benefit is confined to younger age
groups, where the risks from stroke, pulmonary embolism, and deep vein thrombosis are smaller
(Fig. 1).
Unreported data reveal very similar patterns to those in Tables 10 and
11 and Fig. 1 for I(1, 1, 1), I(1,
0, 0), and I(1, 0.5, 0.25). Thus, these conclusions are fairly insensitive to the precise
weights used, provided the weights emphasize life-threatening and severe events.
E. Volunteers for Prevention Trials
Women in the placebo arm of the BCPT had lower mortality rates (71
deaths observed compared with 188 expected from the mortality rates in
Table 3) and lower rates of stroke and pulmonary embolism than the
general population (see section II, part C). Women who
participate in prevention trials such as the BCPT and the WHI tend to
be "healthy volunteers." To assess the sensitivity of the results
in Tables 10 and 11 to lower rates of stroke and
pulmonary embolism and
to offer information to women who are considering participating in
prevention trials and to other comparably healthy women, we multiplied
the rates of stroke and pulmonary embolism in Table 3 for white women
by the observed-to-expected ratio for stroke and pulmonary embolism in
the placebo arm of the BCPT, 0.50 (see section II, part C).
These reduced rates of stroke and pulmonary embolism were used to
calculate the index I(1, 0.5, 0) in Table 12 for white women
with and without a uterus. It is seen that, compared with indices in
Tables 10 and 11, the indices in Table 12 are larger, especially for
older ages. For example, a 65-year-old white woman with a uterus and
with a projected 5-year risk of IBC of 3% has an index equal to -175
in Table 10 and an index equal to -88 in Table 12.
Thus, the
risk/benefit trade-off is improved in "healthy volunteers."
|
Black women who participate in prevention trials such as the WHI and the BCPT are also likely to have lower rates of stroke and pulmonary embolism than in the general population (see section II, part C), and their indices may be more akin to those in Table 12
than to those in Tables 10 and 11.
Table 12 could also be used in place of Tables 10
and 11 for women who have a risk factor profile that indicates they are at
low risk for stroke and pulmonary embolism. F. Lobular Carcinoma In Situ
Women with LCIS had a projected 5-year risk of IBC of 6.47% in the
BCPT. We therefore recommend using the projected 5-year risk of IBC of
6.5% as the entry in Tables 6, 8, 10, and 11 and Fig. 1 for
such women.
|
V. APPLICABILITY OF TRIAL RESULTS TO HIGH-RISK WOMEN NOT INCLUDED IN THE BCPT |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
Although the BCPT did not include the following classes of women, many such women are seeking advice on the possible use of tamoxifen. Realizing that direct data from clinical trials would be needed to establish the role of tamoxifen definitively, we nonetheless offer some informed opinions based on workshop discussion and other sources.
A. Ductal Carcinoma In Situ
Women with a history of DCIS may be candidates for primary
prevention with tamoxifen because they are at high risk for invasive
breast cancer. Women treated with lumpectomy alone in NSABP protocol
B-17, which compared lumpectomy with lumpectomy plus radiation
(62), had a 5-year risk of IBC of 14.7%. IBC includes
contralateral breast cancer. The risk of contralateral invasive breast
cancer alone was 1.89%. The 5-year cumulative risk of IBC for women
treated with lumpectomy and radiation was 6.9% in NSABP protocol B-17
(62) and 7.2% in protocol B-24, which studied lumpectomy and
radiation with or without tamoxifen (63). The corresponding
5-year cumulative risks of contralateral invasive cancer were 2.1% and
2.3%, respectively. Thus, the risk of contralateral disease alone in
these studies is comparable to the 5-year risk of all IBCs, 3.3%, seen
in the placebo arm of the BCPT, and the 5-year risk for all IBCs for
women with DCIS is quite high compared with that for BCPT participants.
Based on these high 5-year risks, even in women treated with lumpectomy
and radiation, the data in Tables 8-12 and Fig. 1
suggest a potential benefit for some women with DCIS. Such women were
excluded from BCPT because they were eligible for a competing protocol
and not because it was thought that tamoxifen would be ineffective.
B. Recent Small, Invasive Breast Cancers
Women with a history of IBC have a risk of about 0.6% per year of developing a second, contralateral, primary breast tumor (59). This corresponds to 5-year risks of about 3%, which is near the risk of IBC of 3.3% for members of the BCPT placebo arm. Consensus opinion suggests that adjuvant therapy with tamoxifen is not indicated for women with invasive breast tumors less than 1 cm in size who have negative axillary lymph nodes (59). However, the consensus opinion was based on studies of tamoxifen as a treatment for primary cancer rather than as a preventive agent against a second new breast cancer. Because the risk of a contralateral, second invasive breast malignancy approaches 20% during the remaining years of life of a woman diagnosed with a first breast cancer at the age of 40 years and is similar to the risk for women in the BCPT, the use of tamoxifen for risk reduction may be a reasonable option, particularly for younger women. There are no data available from studies designed to examine this question, but a review of data from NSABP treatment trials and other trials showed that tamoxifen reduces the incidence of contralateral second primary breast cancers by roughly the same proportion as observed for primary breast malignancies in the BCPT (58). Thus, preventive use of tamoxifen for women with small, lymph node-negative invasive breast cancers may be justified in some cases where there is doubt about its use as adjuvant therapy.
It is not known whether some breast cancers arise without expressing estrogen receptors (ERs) at any point in their genesis or whether all invasive breast cancers pass through a developmental phase in which they produce ER protein. The data from the BCPT indicate that the breast cancers arising among women taking placebo were more likely to express ERs than were those arising in women taking tamoxifen. This suggests that tamoxifen suppressed those developing lesions that expressed ERs but had little or no effect on tumors that did not express ERs. An alternative explanation is that there are breast tumors that arise without expressing ERs at any time in their natural history. If the latter hypothesis is true and if subsequent breast cancers in women whose first cancer did not express ERs are also ER negative, tamoxifen would offer them little benefit. Alternatively, if all breast tumors pass through a phase of ER expression, then tamoxifen may offer benefit even to those women whose first primary breast cancer was ER negative. Although more basic and clinical research is necessary to resolve this question, a meta-analysis of the effects of tamoxifen (59) revealed that "the proportional reduction in contralateral breast cancer appeared to be about the same size in women with ER-poor tumours (29% [SD 15]) as in other women (30% [SD 6])."
C. Remote Diagnosis of Breast Cancer
Another group of women for whom there is no definitive answer about the use of tamoxifen for prevention are cancer-free women who were diagnosed with breast cancer 5 or more years previously ("remote diagnosis") and who were not treated with adjuvant tamoxifen. We abstracted data from several NSABP protocols (64-67) to estimate the subsequent risk of IBC in women who had survived disease free for 5 years following an initial IBC diagnosis and who had not received adjuvant tamoxifen. The subsequent 5-year cumulative risk of contralateral invasive breast cancer was 3.4%, which is close to the risk of 3.3% for IBC in the placebo arm of the BCPT, and the cumulative risk of all IBCs in such women was 14.4%. The decision to use tamoxifen for risk reduction in these patients must be informed by an assessment of the duration and quality of life remaining, the risks as well as potential benefits of tamoxifen, and the presence of competing morbidities that may weigh against the use of tamoxifen. For example, tamoxifen might be appropriate in a 50-year-old woman who is otherwise healthy, but it might be less suitable in a 68-year-old woman with a history of cataracts and deep vein thrombosis.
D. BRCA1/2 Mutation Carriers
Both prospective and retrospective genetic epidemiologic studies (13-15) have demonstrated that women who carry mutations in either the BRCA1 or BRCA2 genes are at very high risk of developing both breast and ovarian cancers. These women would appear to be ideal candidates for the use of tamoxifen in the primary prevention of breast cancer, but no data are yet available that relate directly to such women. While the mechanisms by which tamoxifen might prevent breast cancer in BRCA1/2 mutation carriers are not fully understood, there is no reason to suppose a priori that tamoxifen would necessarily be less effective in mutation carriers, other than the observation that BRCA1 mutation carriers are more likely to develop ER-negative tumors (68-72). Additional laboratory modeling of the effects of tamoxifen in vitro is necessary to address this question, as are prospective data from primary prevention trials that use tamoxifen in mutation carriers. Until these studies are completed, the use of tamoxifen in such women should be accompanied by disclosure beforehand that tamoxifen may not be effective.
|
VI. CLINICAL MONITORING OF WOMEN ON TAMOXIFEN |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
Experience with appropriate clinical management and follow-up of women taking tamoxifen for primary prevention is limited to only a few studies (1-3) and principally to the BCPT. Surprisingly little published information is available from clinical trials that used tamoxifen to treat breast cancer. Endometrial hyperplasia (unpublished data from the BCPT) and endometrial cancer (1) were more frequent among women taking tamoxifen than among women taking placebo in the BCPT, but there was no statistically significant evidence (1) of an elevated risk of endometrial cancer with tamoxifen in women under age 50 years (relative risk = 1.21; 95% CI = 0.41-3.60). There is insufficient evidence for or against the use of transvaginal ultrasound or endometrial sampling for the early detection of endometrial cancer (73), and the American College of Obstetrics and Gynecology (74) has issued the recommendation that women on tamoxifen should have annual gynecologic examinations with Pap tests and pelvic examinations. Any abnormal bleeding should be evaluated with appropriate diagnostic testing. Women should be counseled about the risk of benign and malignant conditions associated with tamoxifen. Screening procedures or diagnostic tests should be at the discretion of the treating physicians, and options should be discussed with women who are considering taking tamoxifen.
Routine screening with hematologic or chemical blood tests is not indicated because no hematologic or hepatic toxic effects attributable to tamoxifen were demonstrated in the BCPT or in clinical trials using tamoxifen as adjuvant therapy.
Because of the modest increase in the risk of cataracts (relative risk = 1.14) and cataract surgery among women on tamoxifen compared with women taking placebo, women taking tamoxifen should be questioned about symptoms of cataracts during follow-up and should discuss with their health care provider the value of periodic eye examinations.
|
VII. COUNSELING |
|---|
|
TopAbstractI. IntroductionII. Projecting Risks in...III. Effects of TamoxifenIV. Risk/Benefit Comparisons for...V. Applicability of Trial...VI. Clinical Monitoring of...VII. CounselingVIII. DiscussionAppendix: Workshop Program and...NotesReferences |
|---|
Health care providers who counsel women about tamoxifen should strive to ensure that the patient makes a fully informed decision that incorporates her personal values and preferences. The counseling process should be interactive and sensitive to the woman's educational level and cultural background. Research suggests that women who were actively involved in decision-making about hormone replacement therapy were more satisfied with their decisions and more informed (75). Because an individual's preferences and risk status can change substantially over time, it is also important that decisions about tamoxifen not be regarded by either patient or counselor as urgent or irreversible.
Any discussion of tamoxifen should occur within the context of a broader discussion of health promotion and breast cancer risk. The encounter should include a qualitative assessment of the patient's risk and, ideally, a quantitative assessment. The woman's perception of her own risk should be elicited so that it can be compared with an objective risk estimate. This discussion might include an