Individualizing Therapy to Prevent Long-term Consequences of Estrogen Deficiency in Postmenopausal Women FREE

MANY postmenopausal women are concerned about the long-term consequences of estrogen deficiency. Hormone replacement therapy (HRT) prevents the accelerated bone loss that occurs in women after menopause^1- 3 and may decrease the risk of developing coronary heart disease (CHD),^1,4 but also appears to increase the risk of breast and endometrial cancer.^5- 6 With the recent approval by the Food and Drug Administration of alendronate sodium and raloxifene hydrochloride therapy, postmenopausal women now have new treatment options to consider. Though not as thoroughly studied as HRT, both alendronate therapy^7- 8 and raloxifene therapy (B. Ettinger, MD, unpublished data, 1998) decrease the risk of vertebral fractures without the cancer risks associated with HRT. However, they also lack some of HRT's benefits. Raloxifene, a nonsteroidal compound that has mixed estrogenic and antiestrogenic properties,^9- 12 lowers serum total cholesterol levels,¹³ does not increase the risk of endometrial cancer,¹⁴ and may decrease the risk of breast cancer.¹⁵ Alendronate, a bisphosphonate, inhibits bone resorption but has no known impact on the incidence of CHD and breast or endometrial cancer.

In studies with follow-up exceeding 30 years,¹⁶ HRT has been shown to decrease fracture rates and the incidence of CHD and to prolong survival.^1,17- 22 In contrast, primary prevention trials of alendronate therapy and raloxifene therapy had more limited follow-up and relied on surrogate measures for clinical efficacy, such as bone mineral density (BMD) and lipid profiles.^13,23 Primary prevention trials will likely require decades of follow-up to demonstrate any impact on end points such as hip fracture and CHD because of their slow evolution.^16,24

To help women and their physicians choose the most beneficial therapy based on the best data currently available, we developed a decision model to examine the effects of HRT, alendronate therapy, and raloxifene therapy on hip fracture, CHD, breast cancer, and life expectancy in postmenopausal women with different risk factor profiles.

We developed a Markov state transition model to compare the following 4 clinical strategies: (1) conservative care (ie, no HRT, alendronate therapy, or raloxifene therapy), (2) HRT, (3) alendronate therapy, and (4) raloxifene therapy. Hormone replacement therapy consists of estrogen (0.625 mg of oral conjugated estrogen or the equivalent once per day) combined with progestin for women with an intact uterus. Alendronate therapy is administered at 5 mg/d and raloxifene hydrochloride therapy at 60 mg/d.

The decision model simulates the natural history of healthy, white 50-year-old postmenopausal women receiving each treatment strategy. All women begin the simulation without evidence of osteoporosis, CHD, or breast cancer. With each subsequent simulated year, cohort members could sustain a hip fracture, develop CHD, develop breast cancer, develop any combination of these diseases, or die from other causes at rates determined by the woman's attained age (additional details of this model are described elsewhere²⁵). We used changes in BMD to predict hip fracture rates^26- 35 by applying a mathematical model³⁶ to characterize the relationship between BMD and subsequent age-specific hip fracture rates. We modeled changes in BMD measurements at the trochanter and femoral neck to predict hip fractures because of the high predictive values of measurements taken at these anatomical sites.³⁷ Fracture risk (based on BMD) was modified according to the presence or absence of risk factors other than BMD.³⁸

The chances of developing CHD^39- 40 and breast cancer⁴¹ were adjusted according to individual risk profiles using coefficients from logistic regression equations that link risk factors to disease incidence.²⁵ Mortality estimates following the development of hip fracture, CHD, breast cancer, or endometrial cancer were calculated from survival data using methods previously described.^25,42

We report the impact of treatment on life expectancy and the number needed to treat (NNT) to prevent (or cause) each target outcome for 10-, 20-, and 30-year periods according to individual risk factors for developing disease. Low risk for CHD or breast cancer corresponds to having no CHD risk factors or no first-degree relatives with breast cancer; moderate risk corresponds to having 1 risk factor for CHD or 1 first-degree relative with breast cancer; high risk corresponds to having 2 CHD risk factors or 2 first-degree relatives with breast cancer.⁴³

In randomized trials conducted among healthy postmenopausal women, HRT,²³ alendronate therapy,^23,44 and raloxifene therapy^13,45 increased BMD compared with either the baseline BMD or BMD with placebo (Table 1).^13,23 There are no data describing the effects of alendronate or raloxifene therapy beyond 3 years, although the effects of HRT appear to continue for as long as treatment is provided.⁴⁶ We assumed that all 3 therapies induced normal bone mineralization^47- 51 and maintained BMD at an elevated level compared with baseline for 10 years (assuming ongoing treatment), with BMD subsequently declining at the rate observed in an age-matched general population.²⁴

Observational studies suggest that HRT lowers the relative risk of developing CHD by 40%,⁴ increases the risk of breast cancer by 47% (if taken for longer than 5 years),⁵ and has no impact on the risk of endometrial cancer (if used with a progestin).⁶ Most well-designed studies and meta-analyses examining HRT confirm these findings.^52- 55 An observational study found that the addition of progestin to estrogen therapy did not attenuate the cardioprotective effect of unopposed estrogen therapy among women without a history of CHD (primary prevention).⁴ In contrast, a recent randomized trial found that estrogen and progestin therapy in combination did not significantly decrease the rate of secondary cardiac events in women with established CHD (relative hazard, 0.99; 95% confidence interval [CI], 0.80-1.22). However, when analyses were restricted to women who were compliant with study medications, the relative hazard was decreased (0.87), but not significantly so.⁵⁶ Because of uncertainties in the impact of HRT on women with established CHD, we assumed that HRT and raloxifene therapy had no effect on the secondary prevention of cardiac events. Because of conflicting evidence concerning HRT's impact on Alzheimer disease^57- 58 and colon cancer,^59- 61 these outcomes were not included in the model. Recent data suggest that HRT increases the risk of deep venous thrombosis (DVT)^56,62- 63 and pulmonary embolism (PE)⁵⁶ (relative risk [RR], 2.4⁶⁴ and 2.1,⁶⁵ respectively). We applied an age-adjusted model to predict baseline risk of DVT and PE,⁶⁶ assuming a 0.25% mortality rate after DVT (resulting from anticoagulation therapy and pulmonary embolism)⁶⁷ and a 31% mortality rate after PE.^68- 74

There are no data on the effects of raloxifene therapy on CHD end points, such as angina, myocardial infarction, or cardiac death. We used data describing its effects on serum lipid levels to predict its effect on CHD, using regression models to link lipid levels to the subsequent development of CHD.^39- 40,75 Just as HRT's effects on serum lipid levels do not explain all of its cardioprotective effects,^76- 81 raloxifene therapy could conceivably have cardioprotective benefits beyond its impact on lipid metabolism.^14,82- 88 We assumed in our baseline analyses that raloxifene therapy yielded no cardioprotective effects beyond lipid reduction.^89- 90 In sensitivity analyses, we explored the possibility that raloxifene therapy also had non–lipid-mediated effects proportional to those of HRT. For those sensitivity analyses, we estimated that 34% of HRT's overall cardioprotective effects were attributable to lipid reduction by comparing the predicted changes in CHD incidence resulting solely from HRT's lipid effects⁹¹ with those actually observed in longitudinal studies.

The long-term impact of raloxifene therapy on breast cancer cannot be established with any certainty at this time.¹⁴ Randomized trials evaluating 10,385 postmenopausal women for up to 3 years found an RR for breast cancer of 0.38 (95% CI, 0.22-0.68) among women treated with raloxifene (compared with placebo).¹⁴ A smaller subset of 7704 women (mean age, 67 years) were found to have an RR of 0.26 (95% CI, 0.13-0.52),¹⁵ but median follow-up was only 29 months, with only 32 cases of breast cancer. In our baseline analyses, we used the most conservative estimate of the effect of raloxifene therapy on breast cancer by applying the upper limit of the 95% CI (RR, 0.68); in sensitivity analyses we explored other possibilities. Raloxifene therapy appears to carry the same risks of DVT and PE as does HRT.¹⁴

Our model predicted that an average woman's lifetime risk for developing hip fracture (in the absence of treatment) was 16%, similar to a published estimate.⁹² To test the validity of using BMD to predict fracture rates, we compared our model's predictions with findings from the Fracture Intervention Trial,⁷ matching our starting cohort to reflect the baseline characteristics of trial participants. Our model predicted an RR of hip fracture of 0.42 for alendronate therapy after 3 years (observed RR, 0.49; 95% CI, 0.23-0.99). Simulated fracture rates among the placebo and treatment groups were 0.02 and 0.008, respectively, compared with observed rates of 0.022 and 0.011.

For women at average risk for CHD, breast cancer, and hip fracture, HRT increased life expectancy more than either raloxifene or alendronate therapy (gain in life expectancy of 8 months vs 5 months and 1 month, respectively). Raloxifene therapy would be preferred over HRT only if it reduced the risk of breast cancer by at least 66% (compared with HRT's increased risk of 47%). No single therapeutic choice was consistently better or worse than other choices for all women; gains in life expectancy depended on the woman's individual risk profile (Table 2). Gains in life expectancy from any of the treatments were modest among women at low risk for osteoporosis, CHD, and breast cancer. Women at high risk for CHD should accrue substantial gains from HRT, while women at high risk for breast cancer should gain from raloxifene therapy.

Gains in life expectancy from alendronate therapy were less than 3 months for most women. Women with 5 or more risk factors for hip fracture (including women with known osteopenia) could gain at most 6 months of increased life expectancy from alendronate therapy.

Using BMD to predict hip fracture risk, we found that HRT, alendronate therapy, and raloxifene therapy should have similar efficacies in preventing hip fractures (simulated RR, 0.57, 0.54, and 0.58, respectively). Table 3 describes the NNT to prevent a single hip fracture. Compared with HRT's RR of 0.60 to 0.65 for developing CHD, we estimated raloxifene therapy's RR to be 0.96 to 0.98. The NNT to prevent a single case of CHD was approximately 10 times higher for raloxifene therapy than for HRT (Table 4). Among women at average risk for breast cancer, 1 case of breast cancer would be induced for every 34 women treated with lifelong HRT. In contrast, 1 case of breast cancer would be prevented for every 48 women treated with lifelong raloxifene therapy (Table 5). These estimates vary according to duration of use and baseline risk for breast cancer.

If we assume that women would prefer HRT, raloxifene therapy, or alendronate therapy over conservative care only if treatment would be expected to provide a gain in life expectancy of at least 6 months, then conservative care was preferred among women at low risk for CHD, breast cancer, and hip fracture. Among women at high risk for CHD, HRT was preferred over alendronate or raloxifene therapy. Among women at high risk for breast cancer but not CHD, raloxifene therapy was preferred. Figure 1 identifies the preferred treatment for a 50-year-old woman according to her risks for CHD, breast cancer, and hip fracture. Table 6 can be used to calculate a woman's risks for CHD, breast cancer, and hip fracture, which can then be represented graphically as a point on this figure. To simplify decision making, we constructed a partition diagram that identifies the optimal treatment according to a woman's risk factor profile, using a gain in life expectancy of 6 months as a threshold (Figure 2).

Because estimates of the long-term impact of HRT, alendronate therapy, and raloxifene therapy on BMD, CHD, and breast cancer are uncertain at this time, we performed sensitivity analyses exploring a range of different efficacies. Alternative assumptions concerning the duration of impact on BMD affected the predicted efficacy of these agents in preventing hip fractures but had little impact on the rank order of choices as long as all treatments shared the same duration of effect.

Our results were sensitive to assumptions concerning the efficacy of HRT in lowering the incidence of CHD. For women at average risk for CHD and breast cancer, HRT was preferred over alendronate or raloxifene therapy as long as the RR for CHD of patients receiving HRT fell below 0.87 or 0.71, respectively (baseline, 0.60).

Our results were also sensitive to our assumptions concerning raloxifene therapy's impact on breast cancer (Table 7). The NNT to prevent 1 case of breast cancer ranged from 18 to 77 as the RR ranged from 0.2 to 0.8. Because the increased risk of breast cancer associated with HRT does not become apparent before 5 years of treatment,^5,53,55 we explored the possibility that raloxifene therapy delays, rather than prevents, breast cancer. In these analyses, we assumed that raloxifene therapy lowers the risk of breast cancer during the first 5 years of treatment (RR, 0.68) but increases the risk thereafter (RR, 1.47). Under this scenario, the cumulative incidence of breast cancer was lower among women receiving raloxifene therapy for less than 8 years but higher thereafter (compared with women receiving no treatment), resulting in a gain of 1 to 3 months in life expectancy.

If raloxifene therapy has non–lipid-mediated cardioprotective effects proportional to those of HRT, its predicted RR for CHD would be from 0.88 to 0.93, and the NNT to prevent a single case of CHD among women at average CHD risk would be 56 for lifelong treatment. An average woman would gain just under 7 months of life expectancy from raloxifene therapy.

Alendronate therapy was preferred over conservative care (but never over HRT or raloxifene therapy) for all levels of risk for CHD and breast cancer as long as the rate of fatal complications induced by alendronate therapy fell below 0.02% per year.

The use of HRT, alendronate therapy, or raloxifene therapy should substantially lower a woman's future chances of hip fracture. Using the common scale of life expectancy, however, the relative benefits of preventing osteoporosis are small compared with the treatment effects on breast cancer and CHD. Assuming that treatment with HRT, raloxifene therapy, or alendronate therapy would be preferred over conservative care only if it provided a gain in life expectancy of at least 6 months, we found that no single therapy was consistently preferred. Women at lowest risk for hip fracture and CHD would not benefit substantially from any of these treatments. Women with multiple risk factors for CHD would benefit most from HRT, with gains in life expectancy exceeding 3 years. Women at high risk for breast cancer but not CHD would benefit most from raloxifene therapy, if raloxifene therapy decreases the risk of breast cancer. If it does not, conservative care would be preferred for these women. Among women at lower risk for CHD, for whom gains in life expectancy from any treatment were modest, personal preferences take on heightened importance in choosing therapy.

Although alendronate therapy was never preferred over HRT or raloxifene therapy, some women may refuse or not tolerate the latter drugs. The relatively small gains in life expectancy from alendronate therapy may underestimate its clinical benefits, as these gains do not consider the substantial morbidity associated with osteoporosis, such as vertebral fractures. Because hip fractures occur late in life, their prevention has a smaller impact on life expectancy than does the prevention of events that occur at a younger age. Although esophagitis has been reported in less than 2% of patients using dosages of 10 mg/d or higher,^93- 95 this complication is rarely fatal and has not been reported at the 5-mg/d dosage.^23,44

The gains in life expectancy from each of the therapies examined are quite high when compared with other preventive strategies targeted at healthy populations. Gains in life expectancy from mammography are estimated at less than 1 month, gains from fecal occult blood testing are 2 months, and gains from Papanicolaou smears are approximately 3 months.⁹⁶

Our findings are not necessarily generalizable to women who are not white, older, or have known osteoporosis, breast cancer, established CHD, or a previous DVT or PE. These drugs have not been adequately tested among these populations and there presently are no predictive models linking BMD and risk factors to fracture rates among nonwhite populations. We did not include tamoxifen citrate in these analyses because it is indicated only among women at high risk for breast cancer at this time and because of its uncertain effects on both fracture rates and CHD.⁹⁰

We could not completely explore the impact of quality of life and personal preferences in our analyses. Individuals vary in their perceptions and valuations of outcomes affected by these therapies, and we lack accurate methods to measure these preferences on an individual basis. Because the risks of HRT include breast cancer, which many women fear more than CHD or hip fracture,^97- 100 any gains in life expectancy from HRT must be balanced against an individual's desire to avoid breast cancer. Similarly, the presence and severity of any symptoms of estrogen deficiency should be considered on an individual basis, as these may be relieved by HRT and worsened by raloxifene therapy.¹⁴ Potential iatrogenic complications that have no impact on mortality were excluded (ie, fibroids or vaginal bleeding associated with HRT, hot flashes associated with raloxifene therapy), although the development of these complications may adversely affect the quality of life and may limit the ability to treat patients with the chosen agent. Our results should assist in the initial choice of therapy, but we did not attempt to predict how well patients will tolerate or comply with these medications. Other considerations include the financial costs and inconvenience of lifelong therapy. These analyses do not attempt to describe who should or should not receive therapy, but rather to provide information on the expected effects of therapy on life expectancy and specific disease outcomes.

Raloxifene and alendronate were approved for the primary prevention of osteoporosis based on less than 3 years of clinical trial experience. For some, this may be an insufficient duration to be confident of the long-term impact of these therapies on fracture risk, breast cancer, or other clinical end points. As the Food and Drug Administration faces increasing pressures to expedite the drug approval process, it is likely that more drugs will be approved with less supporting evidence, potentially leading to a situation wherein patients and their physicians may have more choices but less information to guide their decisions. While waiting for ongoing randomized trials to report on clinical outcomes and the magnitude of adverse or null effects, we believe that decision analytic models can help women and physicians better understand the currently available evidence and integrate it with individual risk factors and personal preferences to make better-informed choices.

Accepted for publication March 4, 1999.

This study was supported in part by grants from the Breast Cancer Research Program of the Massachusetts Department of Public Health, Boston, Mass; the New England Medical Center Research Fund, New England Medical Center, Boston; the Robert Wood Johnson Foundation, Princeton, NJ; and the Pharmaceutical Research and Manufacturers of America Foundation, Washington, DC.

Corresponding author: Nananda Col, MD, Division of Clinical Decision Making, Informatics and Telemedicine, New England Medical Center, Box 302, 750 Washington St, Boston, MA 02111.