- Email: [email protected]
estrogen therapy in healthy postmenopausal women
PREVENTIVE
MEDICINE
Risk/Benefit
20, 79-85 (191)
Considerations of Antiestrogen/Estrogen in Healthy Postmenopausal Women’
Therapy
ANTHONY B. MILLER, M.B.,F.R.C.P. Department of Preventive Medicine and Biostatistics, Crescent W., Toronto, Ontario,
University of Toronto, 12 Queens Park M5S lA8
Canada
Using data on potential years of life lost to ages 75 and 85, the potential effect of antiestrogens and estrogens on various competing causes of death is explored. Although for antiestrogens the beneficial effect sought is a reduction in breast cancer mortality, this could be largely eliminated by a small detrimental effect on cardiovascular disease mortality. Other competing causes of death, such as endometrial, ovarian, and liver cancer, and fractured neck of femur (as an indicator of osteoporosis), are of far less import. For estrogen use at the time of the menopause, no effect on mortality is sought. However, an adverse effect on breast cancer mortality would be of far greater significance than an effect on endometrial cancer. Indeed, the present endeawor to counter an adverse effect of estrogens on endometrial cancer by adding progestogens could be counterproductive. These competing causes of mortality have to be borne in mind in designing intervention trials. 0 1991 Academic
Press. Inc.
INTRODUCTION
This article will approach the issue of risk/benefit from a theoretical perspective, considering the relative importance of the various disease states that could be influenced by antiestrogens or estrogens. Issues related to side effects or immediate toxicity will be omitted, partly because they are addressed by others in this monograph, but largely because they are more immediate and easily recognized, at least in their acute form, and therefore tend to be related to issues of compliance with the intervention, issues which are outside the purview of this article. Even so, it has to be recognized that long-term risks are factors which tend to influence potential recruitment into a program, in the sense that if an investigator is unable to provide sufficient reassurance as to the absence of long-term complications, recruitment will suffer (1). This creates a dilemma, in that unless a trial is carried out, the long-term complications in healthy subjects will not be identified. Hence the need to consider carefully what they may be, so that at least a rough order of risk versus benefit can be appreciated in advance of the trial. DISEASES TO BE CONSIDERED, AND THE RELEVANT ENDPOINTS
The prime purpose of an antiestrogen intervention is reduction in breast cancer incidence and mortality. The diseases that could be increased by the intervention are cancers of the endometrium, ovary, and liver, and osteoporosis. Increase in i Presented at the Workshop on Antiestrogen Prevention of Breast Cancer, October 2-3, 1989, Madison, WI. Proceedings cosponsored by the National Cancer Institute (Grant 1 R13 CA49561-01) and the American Cancer Society (Grant RD 291). 79 0091-7435191 $3.00 Copyright All rights
0 1991 by Academic Press, Inc. of reproduction in any form reserved.
80
ANTHONY
B.
MILLER
ischemic heart disease is also possible, as a reversal of the beneficial effect normally anticipated for estrogens, even though the preliminary studies of Love (2) suggest that the antiestrogen tamoxifen could reduce the incidence of ischemic heart disease. Incidence data for ischemic heart disease and osteoporosis are not available; therefore mortality will be used for all these conditions. This, however, creates a problem for osteoporosis, as the numbers of deaths certified to this cause relative to those from fracture of the neck of the femur, the most important cause of death secondary to osteoporosis, are minimal. Therefore death attributed to fracture of the femoral neck will be used as an indicator of death from osteoporosis. As far as the cancers are concerned, there seems to be good evidence that tamoxifen may increase the incidence of endometrial cancer (3). For liver cancer, tamoxifen appears to act as a promoter in rodents (4). It would therefore be appropriate to assume that mortality from both endometrial and liver cancer may be increased. For ovarian cancer, the evidence is less definitive. However, by analogy with endometrial cancer, and because of some work in animals, it would be prudent to at least evaluate a possible effect of antiestrogens in increasing incidence and therefore mortality from ovarian cancer. For estrogen therapy the main benefit sought is symptomatic relief of menopausal symptoms, for which there is no appropriate measure to use in a riskbenefit analysis. However, two unsought benefits may be obtained, especially if progesterone is not added, reduction in cardiovascular disease and osteoporosis. Once again, mortality has to be used to measure the benefit, also using death from femoral neck fractures as an index of osteoporosis. The risks now become increase in breast and endometrial cancer incidence, but as before in order to make a comparison, mortality will have to be used; even though there is some evidence that mortality from endometrial cancer may not be increased by estrogens, though incidence clearly is. As there is currently no evidence that estrogens (other than as a constituent of oral contraceptives) influence either liver cancer or ovarian cancer incidence, they will not be considered in this context. The measure of mortality to be used cannot be counts of deaths, as everyone will die in time. Other possible indices include age-specific rates, age-standardized rates over an appropriate age range, or cumulative rates over a similar range (in practice another form of age standardization with equal weights applied at all ages). A desire for simplicity suggests the advantage of one standardized rate, but those usually used do not provide sufftcient emphasis on premature mortality, a disadvantage when heart disease is considered as it is an increasing cause of death at older ages. The easiest approach is to use reduction (or increase) in potential years of life lost (PYLL) (5). As the effects of both antiestrogens and estrogens will only be considered when given to postmenopausal women, and as a lag before an effect is likely to be seen will almost certainly occur, PYLL are counted from age 60. The question remaining is to what age the PYLL will be computed. Were it not for uncertainty in cause of death at older ages, and the problem of increasing competing causes of death, PYLL to age 90 would seem to be the obvious choice. However, in reported statistics deaths over age 85 are usually combined, so PYLL to age 85 will be used instead. PYLL have also been computed to age 75, but, as we shall see, this does not lead to different conclusions. Computation of
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF
BREAST
CANCER
81
PYLL requires multiplying the numbers of deaths at each age in a defined population by a weight representing the number of years from the age of death to the selected upper age, in this article 75 or 85, which it seems reasonable to use as an upper limit of the “potential” years of life lost. For example for deaths at ages 60-64, the weight to age 75 is 12.5 and to age 85, 22.5. The PYLL computed in this article are based on deaths reported in Canada in 1985. These numbers should be regarded as an index of premature mortality over the age range considered. Increases or decreases in PYLL for the relevant causes of death under certain assumptions as subsequently described are summed, to provide an indication of the overall change in total (premature) mortality, accepted as the appropriate measure of the overall effect of an intervention, but not always determinable in specific studies because of sample size problems (1). The effect of changing the various assumptions can be estimated by sensitivity analyses. POTENTIAL
EFFECTS
OF ANTIESTROGENS
For breast cancer an assumption was made that the antiestrogen reduced the PYLL by 50% [relative risk (RR) = 0.51, but that it doubled the PYLL from endometrial, ovarian, and liver cancer, though variations in these latter assumptions were also tested. It was also assumed that PYLL from fractured neck of femur was doubled, with a minimal impact to age 75, and relative to the other causes, not a great impact to age 85 either. Ischemic heart disease (IHD) has a much greater PYLL than the other individual causes, approximately double that from breast cancer to age 75 in the general population, and more than double that to age 85. Two different scenarios for effects on IHD mortality were evaluated; no change in PYLL from IHD and the relative increase in PYLL from IHD required in order to abolish the reduction in PYLL from breast cancer, taking the effects on PYLL from other cancers into account. The latter computation gives a “feel” for the relative latitude over cardiovascular disease effects that may be available. No computation was made as to the effect of a reduction of IHD, though it is obvious that if an effect similar to the reduction in cholesterol observed by Love (2) were to be seen, the beneficial effect could be very substantial. Tables 1 and 2 summarize the effect of the various assumptions. Table 1 shows the effect on PYLL to age 75, assuming no effect on IHD mortality. Column 2 indicates the relative risk assumptions made. Column 3 shows the PYLL from age 60 to 75 under normal conditions. If women with normal risk of breast cancer are treated with an antiestrogen (column 4), a small effect on reduction in overall PYLL to age 75 can be expected, the effect of an increase in PYLL from other cancers almost abolishing the beneficial effect from a reduction in PYLL from breast cancer. It would take only an increase of the RR for IHD from 1.00 to 1.03 to abolish this benefit. However, it is generally anticipated that women with increased risk of developing breast cancer would be treated with antiestrogens, largely to reduce the risk/benefit ratio, and column 5 shows the effect of an assumption that women with twice the normal risk of breast cancer are treated. The overall beneficial effect is now much greater, and it would require a RR for IHD of 1.22 to abolish the overall benefit. If selecting women at
82
ANTHONY
B.
TABLE POTENTIAL
EFFECTS
Relative risk
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur
0.5 2.0 2.0 2.0 1.0 2.0
MILLER
1
OF ANTIESTROGEN:
PYLL TO AGE 15
Potential years of life lost
Change”
Changeb
Change’
12,550 1,400 3,590 310 31,808 140
- 6,215 1,400 3,590 310 0 140
- 12,550 1,400 3,590 310 0 140
- 12,550 2,100 5,380 310 0 140
- 835
-7,110
- 4,620
Total
a Among women at normal risk for breast cancer. b Among women at twice normal risk for breast cancer. c Among women at twice normal risk for breast cancer and 50% increased risk for endometrial and ovarian cancer.
increased risk for breast cancer also resulted in them having a 50% increased risk for endometrial and ovarian cancer, the net effect would be to reduce the overall PYLL benefit (column 6) and the required RR for IHD to abolish the effect would also fall to 1.15. If there was no effect of the antiestrogen in increasing PYLL from the other cancers, and women at twice normal risk of breast cancer were treated, the RR for IHD would have to be increased to 1.39 to abolish the reduction in PYLL from breast cancer and fractured neck of femur. The computations for PYLL to age 85 are set out in Table 2. Because of the greater mortality from IHD at older ages, even when PYLL is considered, the increase in RR from IHD required to abolish the overall beneficial effect is less: to 1.14 or 1.09 if other cancers are considered (columns 5 and 6, respectively) and to 1.27 if they are not (findings not tabulated here). TABLE POTENTIAL
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur Total LIChange b Change c Change endometrial
EFFECTS
Relative risk 0.5 2.0 2.0 2.0 1.0 2.0
2
OF ANTIESTROGEN:
PYLL
TO AGE
85
Potential years of life lost
Change”
Changeb
Change’
33,008 4,080 9,640 920 118,600 1,250
- 16,500 4,080 9,640 920 0 1,250
-33,000 4,080 9,640 920 0 1,250
-33,000 6,070 14,450 920 0 1,250
- 17,110
- 10,310
-610
in PYLL among women at normal risk for breast cancer. in PYLL among women at twice normal risk for breast cancer. in PYLL among women at twice normal risk for breast cancer and 50% increased risk of and ovarian cancer.
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF
BREAST
CANCER
83
It is clear, even from these relatively simplistic calculations, that the effect of the antiestrogen on mortality from IHD dominates the possibility that the predicted beneficial effect on PYLL from a reduction in breast cancer will be counteracted by an adverse effect on PYLL from other conditions, while the possibility of an adverse effect on ovarian cancer overshadows the effect on endometrial cancer. The effects on mortality for both liver cancer and fractured neck of femur are both very small, the latter because the maximal risk is not reached until after age 85. Although it would be important to try to obtain more understanding of the potential effect on other hormone-associated cancers (and one possibility not so far considered is that colorectal cancer may behave that way) because of the influence of age at first birth (6), the effect on IHD is critical. Relatively small trials of antiestrogens have been proposed, such as of 16,000 women followed for 5 years. Unfortunately, such trials could not exclude small but important increases in the RR from IHD. For example, to demonstrate an increase in the RR from IHD from 1.O to 1.5 with women ages 50-54 enrolled and followed for 5 years, 40,000 women would be required in each group for an alpha of 0.05 and a power of 95%. Ifan increase to 1.2 was to be demonstrated, the numbers required in each group would exceed 220,000. It is clear that it is very important to obtain as much information as possible from other sources on the effect of antiestrogens on IHD. POTENTIAL
EFFECTS
OF ESTROGENS
To turn now to the risk/benefit of estrogen use, here we are dealing with the objective of the intervention to relieve postmenopausal symptoms, all the disease consequences are subsidiary. However, there are several. That most often considered relates to the risk of increasing the incidence of endometrial cancer, often after a relatively short interval. Less well recognized, but increasingly so, is the possibility of an increase in the incidence of breast cancer, after a IO- to 15-year delay, especially if progesterone is also used (7). There are no other recognized cancer complications. However, beneficial effects on both IHD and osteoporosis can be anticipated, unless progesterone is used with the estrogen, when the effect on IHD might be nullified. Table 3 illustrates the evaluation of these consequences, restricted to PYLL to age 85, in view of the lag to effect on breast cancer, which I have assumed does not occur until age 65. For this table, women at normal risk of developing breast cancer are considered, as there is no selection for postmenopausal estrogen use according to such risk. Column 2 in Table 3 repeats the PYLL first tabulated in column 3 of Table 2. Column 3 of Table 3 shows the assumed RR for these conditions, for the effects of estrogen alone assuming no effect on ovarian or liver cancer or on IHD. A RR of 2.0 for both breast and endometrial cancer is assumed. The latter may be too high, as there was little evidence that mortality from endometrial cancer increased during the period of rising incidence of the disease from estrogen use in the population. The dominant effect seen in column 4 is the increase in PYLL from breast cancer. A reduction in IHD sufficient to counteract this effect would require a RR of 0.80, which is probably of the order to be anticipated. Columns 5 and 6 of Table 3 evaluate the impact of a RR for breast cancer of 4.0 and of
ANTHONY
B. MILLER TABLE
3
POTENTIAL EFFECTS OF ESTROGEN: PYLL TO AGE 85
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur
Potential years of life lost 33,000 4,080 9,640 920
118,600 1,250
Relative risk
Change”
2.0 2.0
19,950 4,080
1.0 1.0 1.0 0.8
Total
4.0
0 0 0 - 250 23,780
Relative risk
Change” 39,900
1.0 1.0 1.0 1.0 0.8
0 0 0 0 - 250 39,650
0 Change in PYLL for breast cancer from age 65.
endometrial cancer and IHD of 1.O, as might occur from estrogen plus progesterone (7). The PYLL from breast cancer is now very much greater requiring an increase in the protective effect on cardiovascular disease to a RR of 0.67 to neutralize the effect, a benefit which seems unlikely. This underlies the importance of avoiding any action that would neutralize a beneficial effect of postmenopausal estrogen use on IHD mortality. It raises the possibility that the current approach of gynecologists of adding progesterone to estrogen prescriptions for postmenopausal symptoms may have precisely the opposite effect on total health than that intended, especially as the potential beneficial effect on endometrial cancer incidence may have little or no impact on mortality. DISCUSSION
There are caveats that have to be applied to the present analysis. First, it is only a superficial look at postulated effects on PYLL, and the numbers cannot necessarily be taken at their face value, nor the more reliable relative relationships between them. Nevertheless they give an indication of the importance of individual causes of death, underlining the dominant effect of cardiovascular disease mortality at these ages, as distinct from mortality from individual cancers. Second, a large number of different assumptions have been made, very few of which are verifiable from currently available data, and some potentially unverifiable, unless large-scale intervention trials were to be run. These assumptions include not only the extent to which RRs may deviate from 1.0 but also the lag time to an effect; this may be different for IHD and osteoporosis than for different cancers. Third, as the analyses have had to be restricted to a measure of mortality, there have to be some reservations over applying them to incidence, let alone overall morbidity, which, preferably, should be our concern. Morbidity from both endometrial cancer and osteoporosis has been underestimated. Nevertheless, many will consider mortality to be the important measure and will suggest that morbidity, from endometrial cancer at least, is not particularly important. There is also an important ethical issue not so far considered. This relates to the question whether it is ethical to place some at risk, even though others benefit. Can we really balance a small increase in risk for IHD with a reduction in breast
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF BREAST CANCER
85
cancer risk, for example? Society accepts such balances in other circumstances, occupational health, for example, but the situation may be different when a potential public health intervention is being considered. The situation is in many respects analogous to that raised by screening. In each instance we are taking an intervention to the public, raising ethical issues completely different from those raised when the public come to us for help (8). In practice, when we cannot promise an overall benefit we will have difficulty in persuading the general public to accept our intervention. This reluctance to participate may even extend to the intervention trial, raising the issue whether such trials should be mounted at all, a question raised by Magnus and Miller over a decade ago (9). We should be striving for interventions that reduce total mortality, even if their effects on specific conditions are readily demonstrable. We must be careful not to allow practices to develop which may have a favorable effect on one condition, but a distinctly unfavorable one on others. Such may be the situation for progesterone added to postmenopausal estrogens. These situations will have to be monitored carefully, for those who observe the benefits may not be those who observe the detrimental effects. Our intervention trials must be large enough, and the period of their follow-up long enough, to ensure that the adverse as well as the favorable effects are fully documented. CONCLUSION
I conclude that in the age group we are considering, over the age of 50, any prophylactic that increases the risk of ischemic heart disease cannot be used, however great its effect on a single cancer. However, prophylactics that appear to have no effect on, or that may reduce the risk of, ischemic heart disease can be evaluated, but their effects on other cancers (and other conditions) must be considered. To do so may require larger trials than we might believe to be necessary for evaluating the specific effect we are interested in, but to perform such trials will be important for the overall public health. REFERENCES 1. Miller AB, Prentice R, Tibshirani R, Boyd N, Corey P. Terry Fox cancer research workshop on prevention trials. Int .I Cancer 1989; 44~767-769. 2. Love R. Antiestrogen chemoprevention of breast cancer: Critical issues and research. Prev Med 1991; 2064-78. 3. Jordan VC. Chemosuppression of breast cancer with long-term tamoxifen therapy. Prey Med 1991; 20:3-14. 4. Dragan YP, Xu Y-D, Pitot, HC. Tumor promotion as a target for estrogen/antiestrogen effects in rat hepatocarcinogenesis. Prev Med 1991; 20:15-26. 5. Romeder J, McWhinnie J. Potential years of life lost between ages 1 and 70: An indicator of premature mortality for health planning. Int J Epidemiol 1977; 6143-151. 6. Howe GR, Craib KJP, Miller AB. Age at first pregnancy and risk of colorectal cancer: A case control study. JNCI 1985; 74~1155-1159. 7. Bergkvist L, Adami HO, Persson I, Hoover R, Schairer C. The risk of breast cancer after estrogen and estrogen-progestin replacement. N Engl J Med 1989; 321:293-297. 8. Miller AB. The ethics, the risks and the benefits of screening. Biomed P/mrmacother 1988; 42:439+42. 9. Magnus K, Miller AB. Controlled prophylactic trials in cancer. JNCZ 1980; M693-699. Received Accepted
May July
21, 1990 19, 1990
MEDICINE
Risk/Benefit
20, 79-85 (191)
Considerations of Antiestrogen/Estrogen in Healthy Postmenopausal Women’
Therapy
ANTHONY B. MILLER, M.B.,F.R.C.P. Department of Preventive Medicine and Biostatistics, Crescent W., Toronto, Ontario,
University of Toronto, 12 Queens Park M5S lA8
Canada
Using data on potential years of life lost to ages 75 and 85, the potential effect of antiestrogens and estrogens on various competing causes of death is explored. Although for antiestrogens the beneficial effect sought is a reduction in breast cancer mortality, this could be largely eliminated by a small detrimental effect on cardiovascular disease mortality. Other competing causes of death, such as endometrial, ovarian, and liver cancer, and fractured neck of femur (as an indicator of osteoporosis), are of far less import. For estrogen use at the time of the menopause, no effect on mortality is sought. However, an adverse effect on breast cancer mortality would be of far greater significance than an effect on endometrial cancer. Indeed, the present endeawor to counter an adverse effect of estrogens on endometrial cancer by adding progestogens could be counterproductive. These competing causes of mortality have to be borne in mind in designing intervention trials. 0 1991 Academic
Press. Inc.
INTRODUCTION
This article will approach the issue of risk/benefit from a theoretical perspective, considering the relative importance of the various disease states that could be influenced by antiestrogens or estrogens. Issues related to side effects or immediate toxicity will be omitted, partly because they are addressed by others in this monograph, but largely because they are more immediate and easily recognized, at least in their acute form, and therefore tend to be related to issues of compliance with the intervention, issues which are outside the purview of this article. Even so, it has to be recognized that long-term risks are factors which tend to influence potential recruitment into a program, in the sense that if an investigator is unable to provide sufficient reassurance as to the absence of long-term complications, recruitment will suffer (1). This creates a dilemma, in that unless a trial is carried out, the long-term complications in healthy subjects will not be identified. Hence the need to consider carefully what they may be, so that at least a rough order of risk versus benefit can be appreciated in advance of the trial. DISEASES TO BE CONSIDERED, AND THE RELEVANT ENDPOINTS
The prime purpose of an antiestrogen intervention is reduction in breast cancer incidence and mortality. The diseases that could be increased by the intervention are cancers of the endometrium, ovary, and liver, and osteoporosis. Increase in i Presented at the Workshop on Antiestrogen Prevention of Breast Cancer, October 2-3, 1989, Madison, WI. Proceedings cosponsored by the National Cancer Institute (Grant 1 R13 CA49561-01) and the American Cancer Society (Grant RD 291). 79 0091-7435191 $3.00 Copyright All rights
0 1991 by Academic Press, Inc. of reproduction in any form reserved.
80
ANTHONY
B.
MILLER
ischemic heart disease is also possible, as a reversal of the beneficial effect normally anticipated for estrogens, even though the preliminary studies of Love (2) suggest that the antiestrogen tamoxifen could reduce the incidence of ischemic heart disease. Incidence data for ischemic heart disease and osteoporosis are not available; therefore mortality will be used for all these conditions. This, however, creates a problem for osteoporosis, as the numbers of deaths certified to this cause relative to those from fracture of the neck of the femur, the most important cause of death secondary to osteoporosis, are minimal. Therefore death attributed to fracture of the femoral neck will be used as an indicator of death from osteoporosis. As far as the cancers are concerned, there seems to be good evidence that tamoxifen may increase the incidence of endometrial cancer (3). For liver cancer, tamoxifen appears to act as a promoter in rodents (4). It would therefore be appropriate to assume that mortality from both endometrial and liver cancer may be increased. For ovarian cancer, the evidence is less definitive. However, by analogy with endometrial cancer, and because of some work in animals, it would be prudent to at least evaluate a possible effect of antiestrogens in increasing incidence and therefore mortality from ovarian cancer. For estrogen therapy the main benefit sought is symptomatic relief of menopausal symptoms, for which there is no appropriate measure to use in a riskbenefit analysis. However, two unsought benefits may be obtained, especially if progesterone is not added, reduction in cardiovascular disease and osteoporosis. Once again, mortality has to be used to measure the benefit, also using death from femoral neck fractures as an index of osteoporosis. The risks now become increase in breast and endometrial cancer incidence, but as before in order to make a comparison, mortality will have to be used; even though there is some evidence that mortality from endometrial cancer may not be increased by estrogens, though incidence clearly is. As there is currently no evidence that estrogens (other than as a constituent of oral contraceptives) influence either liver cancer or ovarian cancer incidence, they will not be considered in this context. The measure of mortality to be used cannot be counts of deaths, as everyone will die in time. Other possible indices include age-specific rates, age-standardized rates over an appropriate age range, or cumulative rates over a similar range (in practice another form of age standardization with equal weights applied at all ages). A desire for simplicity suggests the advantage of one standardized rate, but those usually used do not provide sufftcient emphasis on premature mortality, a disadvantage when heart disease is considered as it is an increasing cause of death at older ages. The easiest approach is to use reduction (or increase) in potential years of life lost (PYLL) (5). As the effects of both antiestrogens and estrogens will only be considered when given to postmenopausal women, and as a lag before an effect is likely to be seen will almost certainly occur, PYLL are counted from age 60. The question remaining is to what age the PYLL will be computed. Were it not for uncertainty in cause of death at older ages, and the problem of increasing competing causes of death, PYLL to age 90 would seem to be the obvious choice. However, in reported statistics deaths over age 85 are usually combined, so PYLL to age 85 will be used instead. PYLL have also been computed to age 75, but, as we shall see, this does not lead to different conclusions. Computation of
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF
BREAST
CANCER
81
PYLL requires multiplying the numbers of deaths at each age in a defined population by a weight representing the number of years from the age of death to the selected upper age, in this article 75 or 85, which it seems reasonable to use as an upper limit of the “potential” years of life lost. For example for deaths at ages 60-64, the weight to age 75 is 12.5 and to age 85, 22.5. The PYLL computed in this article are based on deaths reported in Canada in 1985. These numbers should be regarded as an index of premature mortality over the age range considered. Increases or decreases in PYLL for the relevant causes of death under certain assumptions as subsequently described are summed, to provide an indication of the overall change in total (premature) mortality, accepted as the appropriate measure of the overall effect of an intervention, but not always determinable in specific studies because of sample size problems (1). The effect of changing the various assumptions can be estimated by sensitivity analyses. POTENTIAL
EFFECTS
OF ANTIESTROGENS
For breast cancer an assumption was made that the antiestrogen reduced the PYLL by 50% [relative risk (RR) = 0.51, but that it doubled the PYLL from endometrial, ovarian, and liver cancer, though variations in these latter assumptions were also tested. It was also assumed that PYLL from fractured neck of femur was doubled, with a minimal impact to age 75, and relative to the other causes, not a great impact to age 85 either. Ischemic heart disease (IHD) has a much greater PYLL than the other individual causes, approximately double that from breast cancer to age 75 in the general population, and more than double that to age 85. Two different scenarios for effects on IHD mortality were evaluated; no change in PYLL from IHD and the relative increase in PYLL from IHD required in order to abolish the reduction in PYLL from breast cancer, taking the effects on PYLL from other cancers into account. The latter computation gives a “feel” for the relative latitude over cardiovascular disease effects that may be available. No computation was made as to the effect of a reduction of IHD, though it is obvious that if an effect similar to the reduction in cholesterol observed by Love (2) were to be seen, the beneficial effect could be very substantial. Tables 1 and 2 summarize the effect of the various assumptions. Table 1 shows the effect on PYLL to age 75, assuming no effect on IHD mortality. Column 2 indicates the relative risk assumptions made. Column 3 shows the PYLL from age 60 to 75 under normal conditions. If women with normal risk of breast cancer are treated with an antiestrogen (column 4), a small effect on reduction in overall PYLL to age 75 can be expected, the effect of an increase in PYLL from other cancers almost abolishing the beneficial effect from a reduction in PYLL from breast cancer. It would take only an increase of the RR for IHD from 1.00 to 1.03 to abolish this benefit. However, it is generally anticipated that women with increased risk of developing breast cancer would be treated with antiestrogens, largely to reduce the risk/benefit ratio, and column 5 shows the effect of an assumption that women with twice the normal risk of breast cancer are treated. The overall beneficial effect is now much greater, and it would require a RR for IHD of 1.22 to abolish the overall benefit. If selecting women at
82
ANTHONY
B.
TABLE POTENTIAL
EFFECTS
Relative risk
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur
0.5 2.0 2.0 2.0 1.0 2.0
MILLER
1
OF ANTIESTROGEN:
PYLL TO AGE 15
Potential years of life lost
Change”
Changeb
Change’
12,550 1,400 3,590 310 31,808 140
- 6,215 1,400 3,590 310 0 140
- 12,550 1,400 3,590 310 0 140
- 12,550 2,100 5,380 310 0 140
- 835
-7,110
- 4,620
Total
a Among women at normal risk for breast cancer. b Among women at twice normal risk for breast cancer. c Among women at twice normal risk for breast cancer and 50% increased risk for endometrial and ovarian cancer.
increased risk for breast cancer also resulted in them having a 50% increased risk for endometrial and ovarian cancer, the net effect would be to reduce the overall PYLL benefit (column 6) and the required RR for IHD to abolish the effect would also fall to 1.15. If there was no effect of the antiestrogen in increasing PYLL from the other cancers, and women at twice normal risk of breast cancer were treated, the RR for IHD would have to be increased to 1.39 to abolish the reduction in PYLL from breast cancer and fractured neck of femur. The computations for PYLL to age 85 are set out in Table 2. Because of the greater mortality from IHD at older ages, even when PYLL is considered, the increase in RR from IHD required to abolish the overall beneficial effect is less: to 1.14 or 1.09 if other cancers are considered (columns 5 and 6, respectively) and to 1.27 if they are not (findings not tabulated here). TABLE POTENTIAL
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur Total LIChange b Change c Change endometrial
EFFECTS
Relative risk 0.5 2.0 2.0 2.0 1.0 2.0
2
OF ANTIESTROGEN:
PYLL
TO AGE
85
Potential years of life lost
Change”
Changeb
Change’
33,008 4,080 9,640 920 118,600 1,250
- 16,500 4,080 9,640 920 0 1,250
-33,000 4,080 9,640 920 0 1,250
-33,000 6,070 14,450 920 0 1,250
- 17,110
- 10,310
-610
in PYLL among women at normal risk for breast cancer. in PYLL among women at twice normal risk for breast cancer. in PYLL among women at twice normal risk for breast cancer and 50% increased risk of and ovarian cancer.
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF
BREAST
CANCER
83
It is clear, even from these relatively simplistic calculations, that the effect of the antiestrogen on mortality from IHD dominates the possibility that the predicted beneficial effect on PYLL from a reduction in breast cancer will be counteracted by an adverse effect on PYLL from other conditions, while the possibility of an adverse effect on ovarian cancer overshadows the effect on endometrial cancer. The effects on mortality for both liver cancer and fractured neck of femur are both very small, the latter because the maximal risk is not reached until after age 85. Although it would be important to try to obtain more understanding of the potential effect on other hormone-associated cancers (and one possibility not so far considered is that colorectal cancer may behave that way) because of the influence of age at first birth (6), the effect on IHD is critical. Relatively small trials of antiestrogens have been proposed, such as of 16,000 women followed for 5 years. Unfortunately, such trials could not exclude small but important increases in the RR from IHD. For example, to demonstrate an increase in the RR from IHD from 1.O to 1.5 with women ages 50-54 enrolled and followed for 5 years, 40,000 women would be required in each group for an alpha of 0.05 and a power of 95%. Ifan increase to 1.2 was to be demonstrated, the numbers required in each group would exceed 220,000. It is clear that it is very important to obtain as much information as possible from other sources on the effect of antiestrogens on IHD. POTENTIAL
EFFECTS
OF ESTROGENS
To turn now to the risk/benefit of estrogen use, here we are dealing with the objective of the intervention to relieve postmenopausal symptoms, all the disease consequences are subsidiary. However, there are several. That most often considered relates to the risk of increasing the incidence of endometrial cancer, often after a relatively short interval. Less well recognized, but increasingly so, is the possibility of an increase in the incidence of breast cancer, after a IO- to 15-year delay, especially if progesterone is also used (7). There are no other recognized cancer complications. However, beneficial effects on both IHD and osteoporosis can be anticipated, unless progesterone is used with the estrogen, when the effect on IHD might be nullified. Table 3 illustrates the evaluation of these consequences, restricted to PYLL to age 85, in view of the lag to effect on breast cancer, which I have assumed does not occur until age 65. For this table, women at normal risk of developing breast cancer are considered, as there is no selection for postmenopausal estrogen use according to such risk. Column 2 in Table 3 repeats the PYLL first tabulated in column 3 of Table 2. Column 3 of Table 3 shows the assumed RR for these conditions, for the effects of estrogen alone assuming no effect on ovarian or liver cancer or on IHD. A RR of 2.0 for both breast and endometrial cancer is assumed. The latter may be too high, as there was little evidence that mortality from endometrial cancer increased during the period of rising incidence of the disease from estrogen use in the population. The dominant effect seen in column 4 is the increase in PYLL from breast cancer. A reduction in IHD sufficient to counteract this effect would require a RR of 0.80, which is probably of the order to be anticipated. Columns 5 and 6 of Table 3 evaluate the impact of a RR for breast cancer of 4.0 and of
ANTHONY
B. MILLER TABLE
3
POTENTIAL EFFECTS OF ESTROGEN: PYLL TO AGE 85
Condition Breast cancer Endometrial cancer Ovarian cancer Liver cancer Ischemic heart disease Fracture neck femur
Potential years of life lost 33,000 4,080 9,640 920
118,600 1,250
Relative risk
Change”
2.0 2.0
19,950 4,080
1.0 1.0 1.0 0.8
Total
4.0
0 0 0 - 250 23,780
Relative risk
Change” 39,900
1.0 1.0 1.0 1.0 0.8
0 0 0 0 - 250 39,650
0 Change in PYLL for breast cancer from age 65.
endometrial cancer and IHD of 1.O, as might occur from estrogen plus progesterone (7). The PYLL from breast cancer is now very much greater requiring an increase in the protective effect on cardiovascular disease to a RR of 0.67 to neutralize the effect, a benefit which seems unlikely. This underlies the importance of avoiding any action that would neutralize a beneficial effect of postmenopausal estrogen use on IHD mortality. It raises the possibility that the current approach of gynecologists of adding progesterone to estrogen prescriptions for postmenopausal symptoms may have precisely the opposite effect on total health than that intended, especially as the potential beneficial effect on endometrial cancer incidence may have little or no impact on mortality. DISCUSSION
There are caveats that have to be applied to the present analysis. First, it is only a superficial look at postulated effects on PYLL, and the numbers cannot necessarily be taken at their face value, nor the more reliable relative relationships between them. Nevertheless they give an indication of the importance of individual causes of death, underlining the dominant effect of cardiovascular disease mortality at these ages, as distinct from mortality from individual cancers. Second, a large number of different assumptions have been made, very few of which are verifiable from currently available data, and some potentially unverifiable, unless large-scale intervention trials were to be run. These assumptions include not only the extent to which RRs may deviate from 1.0 but also the lag time to an effect; this may be different for IHD and osteoporosis than for different cancers. Third, as the analyses have had to be restricted to a measure of mortality, there have to be some reservations over applying them to incidence, let alone overall morbidity, which, preferably, should be our concern. Morbidity from both endometrial cancer and osteoporosis has been underestimated. Nevertheless, many will consider mortality to be the important measure and will suggest that morbidity, from endometrial cancer at least, is not particularly important. There is also an important ethical issue not so far considered. This relates to the question whether it is ethical to place some at risk, even though others benefit. Can we really balance a small increase in risk for IHD with a reduction in breast
WORKSHOP:
ANTIESTROGEN
PREVENTION
OF BREAST CANCER
85
cancer risk, for example? Society accepts such balances in other circumstances, occupational health, for example, but the situation may be different when a potential public health intervention is being considered. The situation is in many respects analogous to that raised by screening. In each instance we are taking an intervention to the public, raising ethical issues completely different from those raised when the public come to us for help (8). In practice, when we cannot promise an overall benefit we will have difficulty in persuading the general public to accept our intervention. This reluctance to participate may even extend to the intervention trial, raising the issue whether such trials should be mounted at all, a question raised by Magnus and Miller over a decade ago (9). We should be striving for interventions that reduce total mortality, even if their effects on specific conditions are readily demonstrable. We must be careful not to allow practices to develop which may have a favorable effect on one condition, but a distinctly unfavorable one on others. Such may be the situation for progesterone added to postmenopausal estrogens. These situations will have to be monitored carefully, for those who observe the benefits may not be those who observe the detrimental effects. Our intervention trials must be large enough, and the period of their follow-up long enough, to ensure that the adverse as well as the favorable effects are fully documented. CONCLUSION
I conclude that in the age group we are considering, over the age of 50, any prophylactic that increases the risk of ischemic heart disease cannot be used, however great its effect on a single cancer. However, prophylactics that appear to have no effect on, or that may reduce the risk of, ischemic heart disease can be evaluated, but their effects on other cancers (and other conditions) must be considered. To do so may require larger trials than we might believe to be necessary for evaluating the specific effect we are interested in, but to perform such trials will be important for the overall public health. REFERENCES 1. Miller AB, Prentice R, Tibshirani R, Boyd N, Corey P. Terry Fox cancer research workshop on prevention trials. Int .I Cancer 1989; 44~767-769. 2. Love R. Antiestrogen chemoprevention of breast cancer: Critical issues and research. Prev Med 1991; 2064-78. 3. Jordan VC. Chemosuppression of breast cancer with long-term tamoxifen therapy. Prey Med 1991; 20:3-14. 4. Dragan YP, Xu Y-D, Pitot, HC. Tumor promotion as a target for estrogen/antiestrogen effects in rat hepatocarcinogenesis. Prev Med 1991; 20:15-26. 5. Romeder J, McWhinnie J. Potential years of life lost between ages 1 and 70: An indicator of premature mortality for health planning. Int J Epidemiol 1977; 6143-151. 6. Howe GR, Craib KJP, Miller AB. Age at first pregnancy and risk of colorectal cancer: A case control study. JNCI 1985; 74~1155-1159. 7. Bergkvist L, Adami HO, Persson I, Hoover R, Schairer C. The risk of breast cancer after estrogen and estrogen-progestin replacement. N Engl J Med 1989; 321:293-297. 8. Miller AB. The ethics, the risks and the benefits of screening. Biomed P/mrmacother 1988; 42:439+42. 9. Magnus K, Miller AB. Controlled prophylactic trials in cancer. JNCZ 1980; M693-699. Received Accepted
May July
21, 1990 19, 1990