Invasive breast cancer

Tamoxifen vs. raloxifene. Raloxifene and tamoxifen had similar effects on invasive breast cancer in the STAR head-to-head trial (RR for raloxifene vs. tamoxifen 1.02; 0.82, 1.28),¹² and there were also no differences for estrogen receptor positive and negative subtypes.

Tamoxifen vs. placebo. Tamoxifen reduced invasive breast cancer in all four prevention trials using long-term follow-up data.^20,23,26,29 Reductions ranged from 20% to 43% with the biggest effect from the largest trial, the NSABP P-1 trial (RR 0.57; 0.46, 0.70).²³ Combining results in meta-analysis indicates a summary RR of 0.72 (0.61, 0.86; 4 trials) for all breast cancer (Figure 3) and 0.70 (0.59, 0.82; 4 trials) for invasive breast cancer specifically (Figure 4). Tamoxifen reduced risks for estrogen receptor positive (RR 0.58; 0.42, 0.79; 4 trials), but not estrogen receptor negative breast cancer (RR 1.19; 0.92, 1.55; 4 trials) (Figure 5).^20,23,26,29

Figure 3

Meta-analysis results for all breast cancer outcomes. * per 1.000 women-years # Veronesi, 2007 and Barrett-Connor, 2006 reported mean or median duration of the actual treatment period.

Figure 4

Meta-analysis results for invasive breast cancer. * per 1.000 women-years # Veronesi, 2007 and Barrett-Connor, 2006 reported mean or median duration of the actual treatment period.

Figure 5

Meta-analysis results for estrogen receptor positive and negative breast cancer. * per 1,000 women-years # Veronesi, 2007 and Barrett-Connor, 2006 reported mean or median duration of the actual treatment period.

The IBIS²⁰ and Royal Marsden²⁶ trials provided results for invasive and estrogen receptor positive breast cancer for both active treatment (mean duration 5 and 8 years, respectively) and post treatment periods (mean duration 3 and 5.2 years, respectively). These results indicate continued risk reduction after discontinuation of tamoxifen, providing point estimates of even larger reductions in breast cancer during the post treatment period (Figure 6). However, differences between periods were not statistically significant by subgroup comparison analysis.

Figure 6

Meta-analysis results for invasive and estrogen receptor positive breast cancer—active and post treatment. * per 1,000 women-years

Raloxifene vs. placebo. Raloxifene reduced invasive breast cancer by 44% and 66% in the MORE⁵¹ and RUTH⁴⁶ trials. Combining results in meta-analysis indicated a summary RR of 0.53 (0.34, 0.84; 2 trials) for all breast cancer (Figure 3) and 0.44 (0.27, 0.71; 2 trials) for invasive breast cancer specifically (Figure 4). Raloxifene reduced risk for estrogen receptor positive (RR 0.33; 0.18, 0.61; 2 trials), but not estrogen receptor negative breast cancer (RR 1.25; 0.67, 2.31; 2 trials) (Figure 5).

Tibolone vs. placebo. Tibolone reduced invasive cancer by 68% in the LIFT trial (RR 0.32; 0.13, 0.80; 1 trial).¹⁰ The LIFT trial did not report specific results for estrogen receptor types or noninvasive breast cancer.

Indirect comparisons. Where we lacked data from direct head-to-head trials, we used meta-regression to compare differences in risk ratios derived from placebo-controlled trials. As described above, invasive cancer outcomes for raloxifene vs. tamoxifen were not significantly different when directly compared in the STAR trial.¹² Indirect comparison of raloxifene vs. tibolone also indicated no significant differences (raloxifene vs. tibolone, ratio of risk ratios [RRR] 1.37; 0.49, 3.84). Tibolone and tamoxifen were not compared indirectly because of important differences in patient populations.

Noninvasive breast cancer including ductal carcinoma in situ (DCIS)

Tamoxifen vs. raloxifene. STAR reported nonsignificantly increased risks for noninvasive cancer (RR 1.40; 0.98, 2.00) and DCIS (RR 1.46; 0.90, 2.41) among women using raloxifene vs. tamoxifen.¹²

Tamoxifen vs. placebo. All four tamoxifen trials reported noninvasive cancer outcomes, although specific diagnoses varied between trials. Risks were reduced in the NSABP P-1²³ and IBIS²⁰ trials, and increased in the Royal Marsden²⁶ and Italian²⁹ trials, although results were significant only in the NSABP P-1 trial (RR 0.63; 0.45, 0.89). When combined in meta-analysis, the risk of noninvasive breast cancer was not significantly reduced (RR 0.85; 0.54, 1.35; 4 trials) (Figure 7).

Figure 7

Meta-analysis results for noninvasive breast cancer. * per 1,000 women-years # Veronesi, 2007 and Barrett-Connor, 2006 reported mean or median duration of the actual treatment period.

Raloxifene vs. placebo. Both the MORE⁵¹ and RUTH⁴⁶ trials indicated increased risks for noninvasive breast cancer, although results were not statistically significant. Combining estimates in meta-analysis indicated a nonsignificant elevation in risk (RR 1.47; 0.75, 2.91; 2 trials) (Figure 7). For DCIS specifically, MORE reported 9 cases for raloxifene and 5 for placebo.

Tibolone vs. placebo. One case of DCIS was noted in the tibolone group and one in the placebo group.¹⁰

Breast cancer mortality

Tamoxifen vs. raloxifene. Not reported.

Tamoxifen vs. placebo. All four tamoxifen trials reported breast cancer specific death rates using long-term follow-up data.^20,23,26,29 None of these results were significantly different for tamoxifen vs. placebo (RR 1.07; 0.66, 1.74; 4 trials) (Figure 8).

Figure 8

Meta-analysis results for all-cause and breast cancer death. * per 1,000 women-years # Veronesi, 2007 and Barrett-Connor, 2006 reported mean or median duration of the actual treatment period.

Raloxifene vs. placebo. Very few breast cancer deaths occurred in the MORE/CORE trial and no relative risks were reported.⁵¹

Tibolone vs. placebo. Not reported.

All-cause mortality

Tamoxifen vs. raloxifene. Total death rates am ong women in the STAR trial were similar for women treated with tamoxifen or raloxifene (RR 0.94; 0.71, 1.26).¹²

Tamoxifen vs. placebo. All four tamoxifen trials reported all-cause death rates using long-term follow-up data, and none were significantly different for tamoxifen vs. placebo (RR 1.07; 0.90, 1.27; 4 trials) (Figure 8).^20,23,26,29

Raloxifene vs. placebo. The RUTH and MORE trials reported all-cause death rates that were nonsignificantly reduced compared to placebo (RR 0.91; 0.81, 1.02; 2 trials) (Figure 8).^46,51

Tibolone vs. placebo. The LIFT trial reported 26 deaths among women using tibolone and 28 among those using placebo (p=0.89).¹⁰

Osteoporotic fractures

Tamoxifen vs. raloxifene. Results of the STAR trial indicated no differences between tamoxifen and raloxifene for clinical vertebral, hip, wrist, or total fractures, although all rates were slightly less for raloxifene.¹²

Tamoxifen vs. placebo. The NSABP P-1,²³ IBIS,²⁰ and Royal Marsden²⁶ trials reported fractures as secondary outcomes. The tamoxifen trials enrolled subjects 15 to 20 years younger and with much lower fracture rates than subjects in trials of raloxifene.

In the NSABP P-1 trial, tamoxifen reduced risk of combined clinical vertebral, wrist, and hip fractures with tamoxifen compared to placebo (RR 0.68; 0.51, 0.92).²³ Point estimates of risk ratios were also reduced for these fractures in the IBIS²⁰ and Royal Marsden trials,²⁶ however, results were not statistically significant. Meta-analysis of trials indicates nonsignificant reductions in total (RR 0.84; 0.67, 1.05; 2 trials) and osteoporotic site fractures (i.e., hip, spine, wrist) (0.81; 0.55, 1.18; 2 trials) (Figure 9). Clinical vertebral fractures specifically were not significantly reduced in the NSABP P-1 trial (RR 0.75; 0.48, 1.15) (Figure 10), although hip and wrist fractures combined were (RR 0.66; 0.45, 0.98) (Figure 11).²³

Figure 9

Meta-analysis results for all fractures and osteoporotic site fractures. * per 1,000 women-years. Results are from the active treatment period except for Fisher, 2005 that includes data from the total length of follow-up.

Figure 10

Meta-analysis results for vertebral fractures. * per 1,000 women-years # Barrett-Connor, 2006 reported median duration of the actual treatment period.

Figure 11

Meta-analysis results for nonvertebral fractures. * per 1,000 women-years # Only hip and radius fractures were included.

Raloxifene vs. placebo. The MORE trial recruited women with low BMD (T-score ≤−2.5) and/or prior vertebral fractures.^35,45 At baseline, 37% of women had prior vertebral fractures. In MORE, raloxifene reduced vertebral fractures (RR 0.60; 0.53, 0.69),³⁵ but not nonvertebral or hip fractures compared to placebo.⁴⁵ Results were similar for women with and without prior vertebral fractures and for women using two different doses of raloxifene (60 or 120 mg/day). The RUTH trial measured fractures as secondary outcomes.^46,54 RUTH reported reduced clinical vertebral fractures (RR 0.65; 0.47, 0.89), but not nonvertebral fractures (RR 0.96; 0.84, 1.10) among raloxifene users compared to placebo, consistent with results of MORE.⁴⁶ Combining the results of MORE and RUTH in a meta-analysis indicates a vertebral fracture RR 0.61 (0.54, 0.69) (Figure 10) and a nonvertebral fracture RR 0.97 (0.87, 1.09) (Figure 11).

Tibolone vs. placebo. The LIFT trial¹⁰ recruited women with low BMD (T-score ≤−-2.5) and/or prior vertebral fractures, similar to the MORE trial. At baseline, 22% of women had prior nonvertebral fractures and 26% had prior vertebral fractures. Tibolone reduced vertebral (RR 0.55; 0.41, 0.74), nonvertebral (RR 0.74; 0.58, 0.93), and wrist (RR 0.54; 0.35, 0.82), but not hip fractures (RR 0.72; 0.32, 1.63). Tibolone appeared to reduce more fractures for women with prior vertebral fractures (vertebral RR 0.39; 0.24, 0.63; nonvertebral 0.53; 0.35, 0.81) than for women without prior vertebral fractures (vertebral RR 0.69; 0.48, 1.00; nonvertebral RR 0.86; 0.65, 1.14).

Description of tamoxifen studies

For tamoxifen, information on adverse effects was confined to the four large placebo controlled primary prevention trials,^{19–27,29,30,50,55–69} and the STAR head-to-head trial.^12,18,70,71 We identified no other randomized controlled trials or observational studies that evaluated adverse effects in women without breast cancer. We considered all adverse outcomes at all reported follow-up times to capture potential short and long-term adverse effects. However, because the NSABP P-1 trial was unblinded after reporting initial results in 1998, we focused on data from the earlier 1998 publication,²⁴ and then compared these results with data from the subsequent 2005 publication.²³

Trials reported adverse effects in different ways depending on the outcome. Most evaluated adverse effects at clinic visits using either self or staff administered questionnaires and checklists. The NSABP P-1 trial documented them by using a global index modeled after the Women’s Health Initiative.^{23,24,55,57,59,64} Patients were administered a baseline Health Related Quality of Life examination that was repeated at 36 months. Follow-up visits occurred at 3 and 6 months, and then every 6 months thereafter.⁵⁵ Endometrial cancer and thromboembolic events were considered secondary end points in this trial. Gynecologic symptoms of hot flashes, vaginal discharge, vaginal dryness, and abnormal vaginal bleeding were monitored, and clinical sites reported additional uterine and ovarian disorders and gynecologic procedures.⁵⁷ Medical records for subjects with suspected cardiovascular disease events were collected by the clinical sites and adjudicated by investigators blinded to treatment assignment.⁶⁴ Although trial results were initially reported in 1998 and the study was unblinded at that time, most subjects were followed 7 years.²³ During follow-up, nearly 1/3 of women in the placebo group elected to either enter the STAR trial or begin a SERM for breast cancer prevention.²³ Long-term results of the NSABP P-1 trial are limited by fewer years of follow-up in the placebo group, substantial contamination, and unblinded ascertainment of outcomes.

In the IBIS trial, adverse effects were assessed differently during the active and follow-up phases of the study in Europe and the U.K.; in Australia and New Zealand, the same procedures were used during the entire study.^19,20 During active treatment and post treatment follow-up phases, a checklist of predefined adverse effects with a free text field was used. Predefined adverse outcomes included myocardial infarction, cardiovascular disease events, thromboembolic events, osteoporotic fractures, any non-breast cancer, nausea, vomiting, hot flushes, headaches, vaginal discharge, vaginal dryness, and vaginal bleeding. During the active treatment phase, these questions were asked directly to subjects. During the follow-up phase, a less detailed version of the checklist was mailed to subjects. For postal replies, adverse outcomes were confirmed by medical record review. Approximately 85% of women returned at least one questionnaire during follow-up.

In the Royal Marsden trial, follow-up visits occurred every 6 months during the course of the trial.^25,26 Acute toxicity and other conditions were assessed at each visit and mammograms were performed annually. Further details of the follow-up procedures for adverse effects were not reported.

Subjects underwent a physical examination every 6 months, and blood testing and mammography every 12 months in the Italian trial.^27,29,50,56 After completion of treatment, or in the case of dropouts, women were followed on an annual basis. Information about major endpoints, such as death, serious adverse events, or cancer, was collected continuously and submitted to the data center. Secondary endpoints included cardiovascular disease, psychological measures, and cognitive function. Surveillance for onset of acute or chronic liver injury based on blood levels of transaminases was also included. Only adverse events that occurred during study treatment were reported.

Subjects in the STAR trial were followed every 6 months for 5 years and annually thereafter.¹² Gynecologic examinations, complete blood counts, and routine serum chemistry tests were obtained annually. Information about the occurrence of all protocol-defined endpoints (endometrial cancer, cardiovascular disease, stroke, pulmonary embolism, deep vein thrombosis, transient ischemic attack, osteoporotic fracture, cataracts, death, quality of life, other cancers) was ascertained at each follow-up visit and verified by reviewing relevant records. Self reported symptoms were collected at each contact. In-depth quality of life assessments were also obtained.¹⁸

Description of raloxifene studies

For raloxifene, we obtained adverse effect data from the two large placebo-controlled prevention trials, MORE and RUTH,^{31–35,37–41,46,47,72} the STAR head-to-head trial,^12,18,70,71 12 smaller trials evaluating either bone density, biochemical profiles, or fractures (Appendixes D-1 and D-2),^73–85 and one observational study.⁸⁶ No other observational studies met inclusion criteria. In general, the smaller trials of raloxifene and the observational study contribute little to the evaluation of harms because they involve so few women relative to the large primary prevention trials.

Details of the ascertainment of adverse outcomes were described in the MORE and RUTH trials. Subjects were followed every 6 months in the MORE trial and were queried about potential adverse effects at every visit.^32–34,39 Fasting plasma glucose levels were evaluated annually. Endometrial changes were monitored with transvaginal ultrasound at 17 clinic centers; some centers only performed transvaginal ultrasound on a subset of women. All cases of endometrial cancer were confirmed by a panel blinded to treatment assignment. Medical records and reports were reviewed for subjects reporting possible thromboembolic events by three physician adjudicators blinded to treatment assignment. In RUTH, subjects were followed every 6 months by either a visit or telephone call, and adverse events were ascertained at each evaluation through unsolicited reporting by subjects.⁴⁶ Electrocardiograms were performed at baseline, years 2 and 4, and the final visit. Serum lipids were measured at baseline, years 1 and 5, and the final visit. Committees of experts blinded to treatment assignment adjudicated coronary events, breast cancer, stroke, thromboembolism, and death outcomes.

The 12 smaller trials ranged in size from 129 to 1,145 postmenopausal women. Women had osteoporosis in 5 trials.^{74,79–81,83} The dose of raloxifene ranged from 30 to 150 mg per day, although all trials evaluated a 60 mg per day dose. The duration of the studies ranged from 6 months to 5 years. Several of the smaller trials adequately collected and reported data for selected adverse outcomes, but reported others inadequately or not at all (Appendix D-1), and none evaluated more than 1 to 3 adverse outcomes. Of the 12 smaller raloxifene trials,^{73,74,76–85} only 6 reported thromboembolic events^{77–79,81,82,84} and none reported cardiovascular events. Four trials evaluated uterine outcomes,^73,74,79,80 one urinary outcomes,⁷⁶ and one cognitive function.⁸³ The most commonly reported adverse events were hot flashes and vasomotor symptoms reported in eight trials.^{74,77,78,80–84} The one included observational study evaluated the effect of raloxifene on vaginal bleeding and endometrial thickness.⁸⁶ No other observational studies met inclusion criteria. In general, the smaller trials of raloxifene and the observational study contribute little to the evaluation of harms because they involve so few women relative to the large primary prevention trials.

Description of tibolone studies

The LIFT trial, ^10,87 seven additional randomized placebo-controlled trials (Tables 5 and 6 and Appendixes D1 and D-2),^88–96 and one large cohort study, the Million Women Study (Appendixes C-5, D-1 and D-2),^97,98 met inclusion criteria. Trials ranged in size from 106 to 4,538 subjects, daily tibolone treatment doses ranged from 0.3 to 5 mg, and duration of treatment from 3 months to 3 years. In the large Million Women Study, the dose and duration of tibolone use varied, and the average lengths of follow-up were 2.6 years for incidence of outcomes, and 4.1 years for mortality.⁹⁸ Primary outcomes in these studies included fracture,¹⁰ cardiovascular disease,¹⁰ breast cancer,^10,98 endometrial cancer,⁹⁷ menopausal symptoms,^91,93,94 breast density,⁹⁵ depression,⁹⁶ bone density,^88,92 carotid intima-media thickness,⁸⁸ and lipids,^94,96 although all reported additional secondary outcomes and adverse effects.

Other trials of tibolone were excluded because they enrolled less than 100 subjects, lacked a placebo or nonuse comparison group, or included subjects with a history of breast cancer (Appendix B). For example, the Tibolone Histology of the Endometrium and Breast Endpoints Study (THEBES)⁹⁹ did not contain a placebo group, and the Livial Intervention Following Breast Cancer; Efficacy, Recurrence and Tolerability Endpoints (LIBERATE) trial¹⁰⁰ enrolled women with a history of breast cancer. Other observational studies were reviewed and excluded^101–104 due to the lack of non-use comparison groups, small numbers of tibolone users within a larger pool of menopausal hormone therapy users, and/or lack of reported adverse effects.

Overall, the LIFT trial was well powered for several adverse event outcomes, providing data on cancer, stroke, gastrointestinal, and gynecological outcomes for older postmenopausal women with osteoporosis.^10,87 Although most of the remaining tibolone trials reported some data on various adverse events, most were underpowered to determine statistically significant differences for major outcomes such as death, stroke, and cancer. Other less serious adverse effects were reported with varying degrees of detail.

The large 3-year Osteoporosis Prevention and Arterial effects of tiboLone (OPAL) trial compared tibolone to other types of menopausal hormone therapy or placebo in Europe and the U.S.^89,90 A total of 866 predominantly Caucasian, healthy postmenopausal women ages 45 to 79 years were randomized to tibolone (2.5 mg/daily), conjugated equine estrogen (CEE) with medroxyprogesterone acetate (MPA) (0.625 mg/2.5mg respectively), or placebo for 36 months. Primary outcomes included bone mineral density (BMD) and carotid artery intima-medial thickness; adverse effects on the endometrium and vaginal bleeding were secondary outcomes. Approximately 30% of subjects were lost to follow-up compromising results.

A trial to determine bone density effects of tibolone enrolled 770 healthy postmenopausal women over age 45 years from over 47 sites in the U.S.⁹² Subjects were randomized to either placebo or one of four daily doses of tibolone (from 0.3 to 2.5 mg) for 24 months. Adverse effects were well documented and included deep vein thrombosis, pulmonary embolus, vaginal symptoms, hot flashes, and others.⁹² Loss to follow-up was 34 % in treatment and 29% in comparison groups.

A trial evaluating tibolone’s effect on menopausal vasomotor symptoms enrolled 775 Scandinavian women experiencing severe hot flashes and sweating to either daily placebo or one of four doses of tibolone ranging from 0.625 to 5 mg for 3 years.⁹¹ The placebo group had a higher drop-out rate compared to the tibolone group (20% vs. 11%, respectively) largely due to the lack of a therapeutic effect on vasomotor symptoms.

Four smaller trials conducted in various countries randomized between 106 to 396 healthy postmenopausal women to either 2.5 mg tibolone daily or placebo;^95,96 two trials included a 1.25 mg tibolone daily dose.^93,94 The U.S.⁹³ and Romanian⁹⁴ studies measured vasomotor and sexual function outcomes, the Turkish trial lipids and depression,⁹⁶ and the Swedish trial breast density.⁹⁵ Multiple adverse effects data were well documented in two trials,^93,94 while the other two provided limited data.^95,96 These trials had several methodological limitations, including no description of an intention-to-treat analysis,^94–96 differences between comparison groups for baseline patient characteristics,⁹³ and inadequate information on randomization procedures.⁹⁴ Applicability of the results was also limited because of the enrollment of small, selected populations including women seeking treatment for vasomotor symptoms.

The Million Women Study, a large, population-based prospective cohort study, compared breast and endometrial cancer outcomes of women using various hormone therapy regimens for symptomatic relief of menopausal symptoms with nonusers.^97,98 This study enrolled women age 50 to 64 years who were invited for routine breast cancer screening in the U.K. (N=1,084,110; mean age 56 years). Approximately 6% of the active hormone therapy users in this study were using tibolone. Data included self-reported information on sociodemographic and other personal factors and menopausal status, and cancer incidence and death rates from the National Health Service Central Registers.⁹⁸ This study is limited by the biases introduced by its observational design and subjects’ self-selection of various regimens for symptomatic relief of menopausal symptoms. Some research indicates possible preferential prescribing of tibolone to women at higher risk for breast or endometrial cancer,¹⁰⁵ confounding associations with these outcomes.

Thromboembolic events

Tamoxifen vs. raloxifene. In the STAR trial, raloxifene caused fewer thromboembolic events compared to tamoxifen, including composite measures of thromboembolic events (RR 0.70; 0.54, 0.91), pulmonary embolism (RR 0.64; 0.41, 1.00), and deep vein thrombosis (RR 0.74; 0.53, 1.03).¹²

Tamoxifen vs. placebo. The four tamoxifen prevention trials identified thromboembolic complications as an adverse effect of active treatment, although the evaluation of this outcome varied by trial.^20,24,26,27 None of the trials indicated if thromboembolic events were adjudicated. All trials measured pulmonary embolus and deep venous thrombosis outcomes, the IBIS trial also measured superficial thrombophlebitis and retinal vein thrombosis,²⁰ and the Italian trial measured visceral, retinal, and superficial thrombophlebitis.²⁷ All of these trials excluded women with either a history of prior thromboembolic events or one within 10 years prior to study enrollment.

Active treatment with tamoxifen increased composite measures of thromboembolic events in all four prevention trials resulting in a summary risk ratio of 1.93 (1.41, 2.64; 4 trials) (Figure 12).^20,24,26,27 The IBIS²⁰ and Royal Marsden²⁶ trials provided results for both active and post treatment periods indicating no increased risk after discontinuation of active treatment (RR 1.02; 0.53, 2.97; 2 trials) (Figure 12).

Figure 12

Meta-analysis results for venous thromboembolism. * per 1,000 women-years # For tamoxifen trials, venous thromboembolic events include deep-vein thrombosis (DVT) and pulmonary embolism (PE) only. For other trials, other thrombosis such as retinal vein (more...)

Only the NSABP P-1²⁴ and Italian trials²⁷ evaluated outcomes by type of thromboembolic event. In the NSABP P-1 trial, tamoxifen increased risks for pulmonary embolism (RR 3.01; 1.15, 9.27); but risk was not statistically significantly increased for deep vein thrombosis (RR 1.60; 0. 91, 2.86).²⁴ In the Italian trial, risks were not elevated.²⁷ Summary risk ratios are 2.69 (1.12, 6.47; 2 trials) for pulmonary embolism and 1.45 (0.89, 2.37; 2 trials) for deep vein thrombosis (Figure 13).

Figure 13

Meta-analysis results for deep vein thrombosis and pulmonary embolism. * pet 1,000 women-years

Tamoxifen caused superficial thrombophlebitis in the Italian (RR 1.96; 1.10, 3.51)²⁷ and IBIS trials (RR 2.84; 1.07, 8.78),²⁰ with a summary risk ratio of 2.14 (1.29, 3.56; 2 trials) (Figure 13). The Italian trial also reported one retinal vein thrombosis in each arm of the trial and one visceral thrombosis in the placebo group.²⁷

Raloxifene vs. placebo. Raloxifene increased thromboembolic events in both the MORE (RR 2.10; 1.20, 3.80)³⁹ and RUTH (RR 1.44; 1.06, 1.95)⁴⁶ trials, with similar event rates for women in control groups for both trials (3.50 and 3.67 per 1000 women years, respectively). Further analysis of the MORE trial by year of treatment indicated the highest risks during the first two years of therapy (RR ≥6 in years 1 and 2 vs. 0.9 in year 4).³⁹ Combining results of both trials in a meta-analysis results in a summary estimate of 1.60 (1.15, 2.23; 2 trials) (Figure 12). Both trials also reported nonstatistically significantly elevated risks for pulmonary embolus (combined RR 2.19; 0.97, 4.97; 2 trials) and deep vein thrombosis specifically (combined RR 1.91; 0.87, 4.23; 2 trials) (Figure 13). Although six other smaller trials reported information on thromboembolic events,^{77–79,81,82,84} only two events occurred among women randomized to raloxifene and one among women randomized to placebo in these trials and they were not included in the meta-analyses.

Tibolone vs. placebo. Tibolone did not increase the risk of thromboembolic events,¹⁰ deep vein thrombosis,^91,92 or pulmonary embolism^91,92 in the few trials reporting these outcomes. Rates of thromboembolism in the LIFT trial were 0.8 per 1000 women years in the tibolone group vs. 1.3 in the placebo group.¹⁰

Cardiovascular events

Tamoxifen vs. raloxifene. The STAR trial reported no differences between raloxifene and tamoxifen for a composite measure of ischemic coronary heart disease events (RR 1.10; 0.85, 1.43).¹² Specific events, such as myocardial infarction, severe angina, and acute ischemic syndrome, were also not significantly different between medications.¹² Stroke and transient ischemic attacks were also similar for raloxifene and tamoxifen in STAR (RR 0.96; 0.92, 1.32 and 1.21; 0.79, 1.88, respectively).¹²

Tamoxifen vs. placebo. Although the four prevention trials evaluated cardiovascular events,^20,24,26,27 definitions of outcomes, and the quality and detail of reporting varied across trials. Only the Italian trial indicated that they excluded women with a history of cardiovascular disease other than stable angina.²⁷

The NSABP P-1 trial provided the most detailed information on cardiovascular outcomes, although it did not explicitly describe how these events were defined or adjudicated.²⁴ In this trial, rates of a composite measure of coronary heart disease, myocardial infarction, acute coronary syndrome, and severe angina were similar for tamoxifen and placebo.²⁴ The IBIS trial reported no increase in a composite measure of “all cardiac problems,” including myocardial infarction, angina and other cardiac problems, as well as myocardial infarction specifically for both active treatment and post treatment periods.²⁰ Definitions for these outcomes were not provided. The Italian trial indicated no increase in myocardial infarction but identified an elevated rate of atrial fibrillation (RR 1.73; 1.02, 2.98) among women randomized to tamoxifen,²⁹ however, this is the only trial reporting atrial fibrillation. The Royal Marsden trial reported no differences in “cardiovascular problems.”²⁶

Since tamoxifen showed no differential effects on multiple specific coronary heart disease outcomes, we combined results of composite measures of coronary heart disease in meta-analysis, resulting in a summary risk ratio of 1.00 (0.79, 1.27; 4 trials) (Figure 14).^20,24,26,29 The risk ratio for myocardial infarction specifically is 1.01 (0.63, 1.64; 2 trials) (Figure 15).^20,24,29

Figure 14

Meta-analysis results for coronary heart disease events. * per 1,000 women-years # CHD events includes any reported coronary heart disearse, such as myocardial infarction, angina, acute ischemic syndrome and other CHD events.

Figure 15

Meta-analysis results for myocardial infarction. * per 1,000 women-years

All four prevention trials evaluated stroke outcomes, and stroke was a predefined outcome in the IBIS trial. None of the trials indicated how stroke was defined or whether it was adjudicated. Tamoxifen did not increase stroke in either the active or post treatment periods of the Royal Marsden²⁶ and IBIS²⁰ trials. The Italian²⁹ and NSABP P-1²⁴ trials reported elevated risk ratios for stroke during active treatment that did not reach statistical significance (Italian RR 3.11; 0.63, 15.4; NSABP P-1 RR 1.59; 0.93, 2.77). The summary risk ratio for stroke is 1.36 (0.89, 2.08; 4 trials) (Figure 16). After discontinuation of treatment in the IBIS²⁰ and Royal Marsden²⁶ trials, tamoxifen had no effect on stroke (RR 0.83; 0.20, 3.42; 2 trials) (Figure 16).

Figure 16

Meta-analysis results for stroke. * per 1,000 women-years # Events were reported from at least 3 months after treatment was stopped until the end of follow-up.

Tamoxifen did not increase risk for transient ischemic attack in the trials evaluating this outcome (RR 0.77; 0.46, 1.30; 3 trials) (Figure 17).^20,24,29

Figure 17

Meta-analysis results for transient ischemic attack. * per 1,000 women-years

Raloxifene vs. placebo. Cardiovascular outcomes were extensively evaluated in the MORE and RUTH trials.^32,46 In the MORE trial, raloxifene did not increase risk for a composite measure of coronary heart disease, including myocardial infarction, coronary death, silent myocardial infarction, sudden death, unstable angina, coronary ischemia, and acute coronary syndrome (RR 0.92; 0.66, 1.27).³² Results using a more narrow definition of coronary heart disease events, including coronary death, myocardial infarction, and unstable angina, were similar. Follow-up in the CORE trial also showed no relationship between the use of raloxifene for 8 years and major cardiovascular events (HR 1.16; 0.86, 1.56) or coronary events (RR 1.22; 0.82, 1.83).⁷²

The RUTH trial was designed to identify whether raloxifene prevented coronary heart disease among women at high risk for heart disease or with existing heart disease. In RUTH, raloxifene showed no benefit in reducing composite coronary heart disease outcomes including coronary heart disease death, non-fatal myocardial infarction, and acute coronary syndrome (RR 0.95; CI 0.84, 1.07).⁴⁶ Combining coronary heart disease composite measures from MORE and RUTH provides a summary risk ratio of 0.95 (0.84, 1.06; 2 trials) (Figure 14).

Raloxifene did not increase risk of stroke in the MORE³² or RUTH⁴⁶ trials (RR 0.96; 0.67, 1.38; 2 trials) (Figure 16). In CORE, raloxifene did not increase risk of stroke after eight years of treatment and follow up.⁷² None of the trials evaluated transient ischemic attacks.

Tibolone vs. placebo. Tibolone did not increase risk for coronary heart disease in the LIFT trial¹⁰ or in another smaller trial.⁹³ Reports of sinus bradycardia were higher with tibolone in the LIFT trial.¹⁰

The LIFT trial ended early because of increased ischemic and hemorrhagic strokes in tibolone users (RR 2.19; 1.14, 4.23).¹⁰ In LIFT, transient ischemic attacks were reported as rare in both tibolone group and placebo groups (0.3 % vs. 0.2 %, respectively).¹⁰

Genitourinary outcomes

Tamoxifen vs. raloxifene. Raloxifene users had lower rates of endometrial cancer than tamoxifen users in STAR (1.25 vs. 2.0 per 1000 women years, respectively),¹² but differences were not statistically significant (RR 0.62; 0.35, 1.08).¹² Raloxifene users had fewer hysterectomies than tamoxifen users (RR 0.44; 0.35, 0.56),¹² with rates of 6.04 vs. 13.37 per 1000 women years, respectively; and fewer cases of endometrial hyperplasia (RR 0.16; 0.09, 0.29).¹² The STAR trial found no differences in other genitourinary cancers.¹²

Tamoxifen vs. placebo. Three prevention trials reported data on endometrial cancer;^20,24,26 the Italian trial included only women with prior hysterectomies.⁵⁰ Trials evaluated endometrial changes in different ways. The Royal Marsden trial evaluated endometrial thickness with ultrasound, although the protocol was not reported.⁶² The IBIS trial included endometrial cancer as a predefined outcome. The NSABP P-1 trial monitored gynecologic conditions and procedures during the course of the trial.⁵⁷ In the NSABP P-1 trial, women randomized after July 1994 underwent endometrial sampling prior to randomization, suggesting that women with abnormal sampling were excluded from the trial creating a cohort at lower risk for endometrial cancer.²⁴

All three trials reported increased risks for endometrial cancer with tamoxifen, although only results from the active treatment period of the NSABP P-1 trial reached statistical significance (RR 2.53; 1.35, 4.97).²⁴ Combining these results from the three trials provides a summary risk ratio of 2.13 (1.36, 3.32; 3 trials) (Figure 18). As noted above, the NSABP P-1 trial was unblinded in 1998, however, women continued to be followed for both breast cancer and other outcomes. Nearly one-third of women in the placebo arm of this trial went on to either participate in the STAR trial or electively begin tamoxifen. With these limitations in mind, the risk of endometrial cancer reported after 7 years of follow-up in this trial was even higher (RR 3.28; 1.87, 6.03).²³ When this estimate is included in the meta-analysis, the summary risk ratio is 2.43 (1.50, 4.00; 3 trials).

Figure 18

Meta-analysis results for endometrial cancer. * per 1,000 women-years, based on number of women with an intact uterus # The rate and RR were recalculated based on the number of women at risk (non-hysterectomized). The values reported in the paper were (more...)

In the NSABP P-1 trial, tamoxifen increased rates of endometrial hyperplasia without atypia (RR 2.06; 1.64, 2.60)⁵⁷ and other benign gynecologic conditions for both pre and postmenopausal women. For premenopausal women, these included endometrial polyps (RR 1.9; 1.55, 2.41), leiomyomas (RR 1.3; 1.14, 1.55), endometriosis (RR 1.9; 1.35, 2.70), and ovarian cysts (RR 1.5; 1.2, 1.78), as well as gynecologic surgical procedures including hysterectomy (RR 1.6; 1.88, 11.29).⁵⁷ For postmenopausal women, these included endometrial polyps (RR 2.4; 1.65, 3.24), leiomyomas (RR 1.4; 1.04, 1.80), endometriosis (RR 1.9; 1.29, 5.58), and gynecologic procedures (RR 2.2; 1.6, 3.13).⁵⁷ Tamoxifen had similar effects in the IBIS trial increasing rates of gynecologic procedures including hysterectomy, abnormal bleeding, endometrial polyps, and ovarian cysts.¹⁹ Tamoxifen was associated with higher rates of hysterectomy in the Royal Marsden trial than placebo (177 vs. 96 per 1000 women years, respectively; p<0.001).²⁶ None of the tamoxifen trials reported rates of ovarian cancer.

Tamoxifen increased vaginal symptoms, including dryness, discharge, and other types, in all of the prevention trials.^20,24,26,29 Over twice as many women using tamoxifen vs. placebo reported vaginal discharge (p<0.001) or vaginal symptoms (p=0.008) in the Royal Marsden trial.²⁶ In the NSABP P-1 trial, 13% of women taking placebo and 29% taking tamoxifen reported vaginal discharge that was at least moderately bothersome.²⁴ Tamoxifen increased risks for vaginal dryness (RR 1.14; 0.97, 1.34) and discharge (RR 3.44; 2.9, 4.09) in the Italian trial.²⁹

Tamoxifen increased symptoms of cystitis and incontinence in the Italian trial (RR 1.52; 1.23, 1.89),²⁹ but not similar symptoms during and after active treatment in the Royal Marsden trial.²⁶

Raloxifene vs. placebo. The raloxifene trials differed in their methods of ascertaining endometrial cancer outcomes. In the MORE trial, 17 clinical centers performed annual transvaginal ultrasonography in all subjects with a uterus, carefully monitoring uterine pathology.³⁹ In the RUTH trial, endometrial cancer was determined on the basis of unsolicited reporting by the participant.⁴⁶ In neither trial were the risks of endometrial cancer elevated (combined RR 1.14; 0.65, 1.98; 2 trials) (Figure 18).^39,46

Raloxifene did not cause uterine bleeding in several trials^{33,46,73,74,77–80,82,84} and the one observational study⁸⁶ reporting this outcome. Raloxifene increased rates of endometrial cavity fluid, as determined by periodic transvaginal ultrasound in the MORE trial (p<0.009).³³ Raloxifene did not increase rates of ovarian cancer in RUTH, the only trial reporting this outcome.⁴⁶ Raloxifene increased urinary symptoms in the CORE trial (2.1% raloxifene vs. 1.2% placebo; p=0.041).⁵¹

Tibolone vs.. placebo. Three studies provide conflicting data on tibolone and endometrial cancer. The OPAL trial⁹⁰ reported only one case of endometrial cancer in each of the placebo and treatment groups, while women with an intact uterus in the LIFT trial⁸⁷ had a trend toward increased risk with tibolone (0 vs. 4 cases, respectively, p=0.06). The mean age of women in the LIFT trial was 10 years older than the age of women in the OPAL trial (68 vs. 58.7, respectively). In contrast, tibolone users with a mean age of 58 years and no prior cancer or hysterectomy in the U.K. Million Women’s cohort study showed an increased risk for endometrial cancer (RR1.79; 1.43, 2.25). ⁹⁷ In the Million Women’s Study, endometrial cancer risk was increased for woman age ≥60 and with >3 years use of tibolone compared with younger women and shorter durations of use.⁹⁷ Tibolone did not increase risk for cervical cancer¹⁰ or uterine cancer⁸⁹ in the two trials reporting these outcomes.

Tibolone did not increase risk for endometrial hyperplasia and moderate or severe dysplasia;¹⁰ however, tibolone was associated with increased rates of procedures for endometrial thickness, hyperplastic polyps,⁸⁷ and endometrial biopsy.¹⁰ Tibolone did not increase endometrial thickness in two other trials, the large OPAL trial in the U.S. and Europe and another small Romanian study.⁹⁴

Tibolone increased vaginal bleeding and spotting in the LIFT and OPAL trials.^87,90 A large Scandanavian trial in younger women reported a dose effect for bleeding and spotting with highest rates with 5 mg/day.⁹¹ Tibolone did not increase vaginal bleeding rates at 6 month follow-up in a trial that reported 12% to 15% bleeding rates.⁹⁴ Other trials report bleeding or spotting as tolerable with no differences between tibolone and placebo.^92–94

Tibolone increased pelvic pain, vaginal infection, and vaginal discharge in LIFT.¹⁰ Tibolone did not increase rates of uterine spasm,⁹³ enlarged abdomen,⁹³ genital pruritus,⁹³ or abdominal pain.⁹¹ Tibolone improved vaginal maturation measures,⁹³ vaginal dryness, and sexual function.⁹⁴

Non-cancer breast outcomes

Tamoxifen vs. raloxifene. No results.

Tamoxifen vs.. placebo. Tamoxifen is associated with reductions in breast density in both the IBIS and NSABP P-1 trials. In a subsample of 69 women in the IBIS trial, at 18 months, women on tamoxifen had a 7.9% greater decrease in breast density than women on placebo; at 54 months, the difference was 13.7% (p<0.001).²⁰ In the NSABP P-1 trial, between 1 to 3.4 years, 38.5% of tamoxifen users had decreased breast density compared with 6.7% of placebo (p<0.069),⁵⁵ and between 3.5 and 5 years, the difference was 48% compared with 22% (p<0.114).⁵⁵ Tamoxifen did not cause breast symptoms in the IBIS and Royal Marsden trials.^20,26

Raloxifene vs. placebo. Raloxifene did not decrease breast density in a small trial of postmenopausal women with osteoporosis (1.3% reduction for placebo, 1.5 % for raloxifene 60 mg/day, 1.7% raloxifene 120 mg/day).⁷⁵

Tibolone vs. placebo. Tibolone did not reduce breast density^94,95 or cause breast pain.^92,93,95 Breast pain ranged from approximately 5%⁹³ to 10%⁹⁴ in tibolone users. Tibolone users without prior hysterectomies in the LIFT trial had more breast discomfort.¹⁰

Ophthalmologic disorders

Tamoxifen vs. raloxifene. In the STAR trial, women on raloxifene had fewer cataracts (RR 0.79; 0.68, 0.92) and cataract surgery (RR 0.82; 0.68, 0.99) than women on tamoxifen.¹²

Tamoxifen vs. placebo. All four prevention trials evaluated ocular outcomes,^20,24,26,29 although the Italian trial reported data on the composite category of “ophthalmologic diseases.”²⁹ None of the trials described how women were evaluated for ophthalmologic outcomes. The NSABP P-1,²⁴ Royal Marsden,²⁶ and IBIS²⁰ trials reported increased cataracts with tamoxifen, although results for the IBIS trial did not reach statistical significance. Combining results in meta-analysis indicates a summary risk ratio of 1.13 (0.70, 1.83; 3 trials) (Figure 19). A sensitivity analysis including 7-year follow-up data from the NSABP P-1 trial²³ (see limitations discussed above) rather than short-term follow-up, indicates a summary risk ratio of 1.27 (1.00, 1.62).^20,23,26 Cataract surgery was also evaluated in the NSABP-1 trial and risk estimates were elevated in the initial (RR 1.57; 1.16, 2.14)²⁴ and follow-up (RR 1.21; 1.10, 1.34)²³ reports.

Figure 19

Meta-analysis results for cataracts. * per 1,000 women-years

Raloxifene vs. placebo. Raloxifene did not cause more cataracts than placebo in the MORE and RUTH trials.^39,46

Tibolone vs. placebo. Tibolone did not increase rates of retinal detachment in one trial.⁹¹

Gastrointestinal and hepatobiliary disorders

Tamoxifen vs. raloxifene. No results.

Tamoxifen vs. placebo. Tamoxifen did not cause gastrointestinal symptoms in the Italian and Royal Marsden trials.^26,29

Raloxifene vs. placebo. In RUTH, raloxifene caused more cholelithiasis and dyspepsia (230 compared with 186; p=0.03), although rates of cholecystectomy were similar.⁴⁶

Tibolone vs. placebo. Tibolone did not cause cholecystitis,⁹¹ but increased liver function tests;¹⁰ gastroenteritis was more common with placebo.¹⁰ In LIFT, tibolone reduced risk for colon cancer (RR 0.31; 0.10, 0.96).¹⁰

Other outcomes impacting quality of life

Tamoxifen vs. raloxifene. In STAR, mean scores on quality of life instruments (health survey, depression scale, sexual questionnaire) did not differ between women using tamoxifen vs. raloxifene, except sexual function was slightly better for tamoxifen (odds ratio, 1.22%; 1.01, 1.46).¹⁸ Women using raloxifene reported more musculoskeletal problems, dyspareunia, and weight gain, while those using tamoxifen had more gynecological problems, vasomotor symptoms, leg cramps, and bladder control symptoms.¹⁸

Tamoxifen vs. placebo. Tamoxifen increased vasomotor symptoms in the four prevention trials,^20,24,26,29 although vasomotor and gynecologic symptoms were combined in the IBIS trial.²⁰ In the Royal Marsden trial, 32% of women taking placebo reported hot flashes vs. 48% of women taking tamoxifen (p<0.001).²⁶ The NSABP P-1 trial had similar findings; hot flashes in 29% of placebo and 46% of tamoxifen groups.²⁴ In the Italian trial, the risk ratio for hot flashes with tamoxifen was increased at 1.78 (1.57, 2.0).²⁹

Two studies from the NSABP P-1 trial evaluated outcomes of depression and quality of life and identified no increased depression with tamoxifen.^21,22,59 Women randomized to tamoxifen reported 4% more sexual side effects than women randomized to placebo, although women on tamoxifen were slightly more sexually active (p=0.031).⁵⁹ Tamoxifen caused weight gain in the Royal Marsden trial,²⁶ but not in the Italian trial.²⁹ Tamoxifen did not increase headaches in the IBIS or Royal Marsden trials.^20,26

Raloxifene vs. placebo. Raloxifene increased vasomotor symptoms in both the MORE and RUTH trials.^33,46 In MORE, 7% of women using placebo, 11% using raloxifene 60 mg, and 12% using raloxifene 120 mg reported vasomotor symptoms (p<0.05).³³ In the RUTH trial, comprised of older women, the rates of vasomotor symptoms were lower in general than in MORE, but higher for women taking raloxifene compared with placebo (4.8% placebo vs. 8.0% raloxifene; p<0.001).⁴⁶ Raloxifene also caused hot flashes and other vasomotor symptoms in three^77,78,80 of eight smaller trials that evaluated vasomotor effects.^{74,77,78,80–84}

Raloxifene caused leg cramps in three^33,46,80 of six trials.^{33,46,77,78,80,82} Raloxifene caused peripheral edema in the MORE (6.1% placebo, 7.1% raloxifene 60 mg, 7.9% raloxifene 120 mg; p=0.026)³³ and RUTH trials (12.1% placebo, 14.4% raloxifene; (p<0.001).⁴⁶

Influenza syndrome symptoms occurred at a higher rate among women taking raloxifene in MORE (14% placebo, 16.2% raloxifene 60 mg, 16.7% raloxifene 120 mg),³³ but not in two other studies.^46,84 Raloxifene caused joint pain in two trials,^46,79 but not in a third.⁸⁴ Raloxifene had no effect on mood, depression, and anxiety symptoms in three trials.^46,83,84

Tibolone vs. placebo. Unlike tamoxifen and raloxifene, tibolone reduces vasomotor symptoms, such as the number and severity of hot flashes.^91,93,94 One study showed reduction in hot flashes for the 2.5 mg/day tibolone dose, but not in the 0.3–1.25 mg/day doses.⁹² Tibolone did not increase weight in two trials.^92,93 Measures on the Beck Depression Inventory were improved with tibolone after one year of treatment in one trial.⁹⁶

Tibolone did not cause several other symptoms that impact quality of life in trials measuring these outcomes, such as musculoskeletal disorders,⁸⁹ headache,^91–93 back or abdominal pain,⁹² upper respiratory⁹³ or respiratory tract infection,⁹² allergy,⁹², sinusitis,⁹² accidental injury,⁹² anxiety and nervousness,⁹² nausea,^93,94 fluid retention,⁹⁴ and concussion.⁹¹ Tibolone did not cause moniliasis in the 0.3–1.25 mg/day doses, however, was greater in the 2.5 mg/day dose compared to placebo.⁹²

Age

The STAR,¹² IBIS,²⁰ Italian,²⁹ NSABP P-1,²³ RUTH,⁴⁷ and MORE.⁴² trials evaluated breast cancer outcomes by age categories, although categories varied by trial. In STAR, invasive cancer outcomes did not differ significantly for women using raloxifene vs. tamoxifen in the three age categories evaluated (≤49, 50 to 59; ≥60 years), and results were similar across categories.¹² In the three tamoxifen vs. placebo trials, summary risk estimates for invasive or all cancer outcomes were significantly reduced and similar for women ≤50 and >50 years (Figure 20).^20,23,29 The raloxifene trials stratified results for invasive cancer using different age categories (MORE <65 years; RUTH <60 years) and we did not combine them in a meta-analysis. MORE reported a reduced risk ratio for women ≥65 vs. <65 years,⁴² and RUTH an increased risk ratio point estimate for women ≥60 vs. <60 years,⁴⁷ although confidence intervals overlap (Figure 20).

Figure 20

Subgroup analysis by age. * per 1,000 women-years & The total follow-up time is averaged over both MORE and CORE for the 7705 participants.

The NSABP P-1 trial suggested higher risks for deep vein thrombosis, pulmonary embolism, and stroke for women >50 vs. ≤50 years; rates and risk ratios are higher, but results are not statistically significant.²⁴ Age >60 years was also an important risk factor for venous thrombosis in the Italian trial.²⁷ The NSABP P-1 trial also found that endometrial cancer was more common among women >50 vs. ≤50 years (RR 4.01; 1.70, 10.90 vs. 1.21; 0.41, 3.60; respectively).²⁴ In LIFT, rates of stroke were highest among tibolone users age >70 vs. 60 to 70 years (6.6 vs. 2.8 per 1000 women years).¹⁰

Menopausal status

The IBIS²⁰ and Royal Marsden²⁶ trials evaluated breast cancer outcomes by menopausal status (pre vs. post). Point estimates indicate similar risk reduction with tamoxifen vs. placebo for both pre and postmenopausal women, although results were not statistically significant for postmenopausal women in both trials (Figure 21). We detected no significant differences between pre and postmenopausal women by subgroup comparison analysis.

Figure 21

Subgroup analysis by menopausal status. * per 1.000 women-years

Hysterectomy status

In RUTH, raloxifene did not significantly reduce risk for invasive cancer for women with prior hysterectomies or oophorectomies, while risk reduction was significant in women without these prior surgeries.⁴⁷ However, these differences could reflect the smaller numbers of women in the surgical subgroups.

Use of exogenous estrogen

The IBIS,²⁰ Italian,²⁹ NSABP P-1,²³ RUTH,⁴⁷ and MORE⁴² trials evaluated breast cancer outcomes by use of menopausal hormone therapy (estrogen with or without progestin). In the tamoxifen trials, women were allowed to use hormones during the trial, and use rates varied from <10% in NSABP P-1²⁴ to 40% in IBIS.¹⁹ Women in the raloxifene trials were not allowed to use hormones during the trial and hormone use status represented prior use. For both tamoxifen and raloxifene trials, point estimates improved and results became statistically significant for hormone nonusers compared to users, although summary estimates were not significantly different (Figure 22). These findings may reflect the smaller numbers of hormone users in the trials.

Figure 22

Subgroup analysis by estrogen use. * per 1,000 women-years # For tamoxifen trials, hormone replacement therapy (HRT) use refers to HRT use during the trial period only. For raloxifene trials, HRT use refers to prior HRT use.

Risk of breast cancer

Family history. The STAR,¹² Italian,²⁹ NSABP P-1,²³ RUTH,⁴⁷ and MORE⁴² trials evaluated breast cancer outcomes by family history of breast cancer, most commonly referring to the number of first-degree relatives with breast cancer. In STAR, invasive cancer did not differ significantly for women using raloxifene vs. tamoxifen in the three family history categories evaluated (0, 1; >2), and results were similar across categories.¹² Tamoxifen reduced invasive and all breast cancer for women without a family history in the two tamoxifen vs. placebo trials, but had dissimilar results for women with a family history (Figure 23). In the NSABP P-1 trial, risk was similar for women in both family history groups; in the Italian trial, risks were reduced for women with no family history and increased for women with family history, although results were not statistically significant (Figure 23). The raloxifene trials indicate similar significantly reduced risk estimates for women without family history and dissimilar results for women with family history (Figure 23). These results may reflect the smaller numbers of women with positive family history for breast cancer in these trials rather than true medication effects. We did not combine results for women with family history for tamoxifen or raloxifene trials in a meta-analysis.

Figure 23

Subgroup analysis by family history of breast cancer. * per 1,000 women-years # With family history (FH) is defined as having at least one first-degree relative with breast cancer, and otherwise it is without FH.

Body mass index. A nested case-control analysis of data from the NSABP P-1 trial indicates that elevated body mass index is associated with higher risk of thrombembolic events among women in both the placebo and control groups (RR 3.69; 2.09, 6.65).¹⁰⁶ Additional analysis of the prothrombin gene mutation and Factor V Leiden deficiency indicated no interaction with tamoxifen and risk of thromboembolic events. This analysis also indicated that the risk of thromboembolic events was elevated only during the first 3 years of use of tamoxifen. The RUTH and MORE trials evaluated invasive breast cancer by body mass index (BMI ≤25 vs. >25).^42,47 While MORE indicated similar significantly reduced risk estimates for women with low and high BMI, RUTH reported lower risk estimates for women with high BMI (Figure 24), although estimates were not significantly different between women with low or high BMI.

Figure 24

Subgroup analysis by body mass index. * per 1,000 women-years # For Grady 2008, total n = 2416 for BMI <=25, and 7655 for BMI > 25.

History of breast abnormalities. In STAR, tamoxifen and raloxifene had similar effects on invasive breast cancer regardless of history of LCIS or atypical hyperplasia.¹² In NSABP P-1, tamoxifen reduced invasive cancer compared to placebo regardless of history of LCIS or atypical hyperplasia, although reduction was greatest among women with prior atypical hyperplasia (RR 0.25; 0.10, 0.52).²³

Predicted breast cancer risk. In STAR, tamoxifen and raloxifene had similar effects on invasive breast cancer for women in all risk categories determined by the modified Gail model (5-year predicted risk ≤3.00; 3.01 to 5.00;≥5.01).¹² In NSABP P-1, tamoxifen reduced risk for invasive cancer compared to placebo for women in all modified Gail model risk categories (5-year predicted risk ≤2.00, 2.01 to 3.00; 3.01 to 5.00, ≥5.01).²³ Cancer rates were highest and risk reduction greatest among women in the highest risk group in this trial. In RUTH, raloxifene reduced risk for invasive cancer compared to placebo for women in all modified Gail model risk categories (5-year predicted risk ≤2.00, 2.01 to 3.00; 3.01 to 5.00), although results were statistically significant only for the large number of women in the lowest risk group.⁴⁷

Estradiol levels. Raloxifene had less effect on invasive cancer outcomes among women with estradiol levels <5 pmol/L (RR 0.52; 0.26, 1.06) than women with higher levels (5 to 10 pmol/L, RR 0.33 ; 0.13, 0.84; >10 pmol/L, RR 0.25; 0.14, 0.47) in MORE/CORE.⁴²

Reproductive factors. Raloxifene reduced risk for invasive cancer regardless of age at menarche (<11, ≥11 years), parity (0, 1 to 2, ≤3), or age at first live birth (<20, ≥20 years) in the RUTH trial.⁴⁷

Rates of adherence and persistence

Adherence is the extent to which a patient acts in accordance with the prescribed interval and dose of a dosing regimen.¹¹⁶ Persistence is the duration of time from initiation to discontinuation of therapy.¹¹⁶

Adherence was reported by one tamoxifen trial,²⁶ four raloxifene trials,^34,46,76,84 and one tibolone trial,¹⁰ and was lacking for several trials including STAR (Table 7).¹² Of trials reporting adherence, results indicate at least 70% adherence with the planned treatment dose, however, these data do not allow direct comparisons between trials. In the Royal Marsden trial, adherence was 8% lower with tamoxifen vs. placebo (p=0.002).²⁶ In RUTH, there were no differences between raloxifene and placebo; 70% vs. 71% took at least 70% of the study medication, respectively.⁴⁶ Adherence was not reported separately in MORE; 92% of the entire study population took at least 80% of the assigned study medication.³⁴ In LIFT, 91% received at least 80% of the assigned study medication.¹⁰

Table 7

Compliance outcomes for trials of tamoxifen, raloxifene, and tibolone.

Persistence was measured as duration of treatment in the STAR trial,¹² one tamoxifen trial,²⁹ three raloxifene trials,^46,76,80 and one tibolone trial;¹⁰ and as completion of the planned course of treatment by two tamoxifen trials,^20,29 six raloxifene trials,^{46,76,79–81,84} and two tibolone trials.^90,109

In the STAR trial, treatment was ongoing at the time of publication and final persistence rates have not been published, although the mean duration of treatment was similar for raloxifene and tamoxifen (3.2 vs. 3.1 years, respectively).¹² In the Italian trial, designed for 60 months of treatment, women using tamoxifen had lower completion rates than placebo (59.8% vs. 61.8%, respectively).²⁹ The IBIS trial had similar results, although both groups had higher completion rates than the Italian trial (63.9% vs. 72%, respectively).²⁰ In RUTH, women using raloxifene had slightly higher completion rates than placebo (80% vs. 79%; p=0.02), although the median duration of treatment was 5.05 years for both groups.⁴⁶ Additional trials of raloxifene reported 60% to 91% of subjects completing the planned duration of treatment.^{76,79–81,84} In LIFT, prematurely discontinued due to preset stopping rules, the median duration of treatment with tibolone was 34 months.¹⁰ Completion rates in OPAL were 69% for tibolone and 70% for placebo,⁹⁰ and 89% overall in another tibolone trial.¹⁰⁹

Harms or benefits affecting adherence and persistence

Evidence that harms or secondary potential benefits affect adherence and persistence was sporadically reported in tamoxifen and tibolone trials as protocol specified and non-protocol specified events. Protocol specified events are outcomes explicitly stated in the protocol requiring that a participant discontinue the study medication.

Tamoxifen vs. placebo. Two trials reported treatment discontinuation due to non-protocol specified events.^24,29 In the Italian trial, 7.6% of tamoxifen vs. 6.9% of placebo groups withdrew from treatment due to protocol specified events, and 26.7% vs. 25.3% due to non-protocol specified events.²⁹ In the NSABP P-1 trial, 23.7% of tamoxifen vs. 19.7% of placebo groups discontinued due to non-protocol specified events.²⁴

Raloxifene vs. placebo. Eight raloxifene trials provided information on discontinuation rates due to adverse events.^{34,46,73,76,79–81,84} In RUTH, 22% of raloxifene and 20% of placebo groups discontinued study medications due to adverse events (p=0.01); specific adverse events were not described.⁴⁶ In the MORE trial, significantly more women receiving raloxifene than placebo withdrew from treatment due to hot flashes.³⁴ In another trial to evaluate the effect of raloxifene on hot flashes in postmenopausal women, vasomotor symptoms caused discontinuation in two women using raloxifene and four using placebo, and 14 other patients discontinued due to other adverse events that were not described.⁸⁴ In the OPAL trial, discontinuation rates for hot flashes (5%) and leg cramps (1%) were higher for raloxifene than placebo (1% vs. 0%).⁸⁰ In a trial to assess the uterine effects of raloxifene in healthy postmenopausal women, discontinuation due to gynecologic adverse events were not significantly different between groups (3 placebo, 1 raloxifene 60 mg/day, 2 raloxifene 120 mg/day).⁷³ Three other trials reported discontinuation rates due to adverse events that were not further described.^76,79,81

Tibolone vs. placebo. The LIFT trial reported higher rates of discontinuation due adverse events for tibolone, but did not provide data.¹⁰ A trial designed to evaluate the effects of 1.25 and 2.5 mg/day doses on early postmenopausal bone loss reported discontinuation rates due to adverse events as 7% for tibolone vs. 17.4% for placebo.¹⁰⁹

Surveys of treatment choice and concordance

Concordance occurs when a health care provider and patient reach a shared agreement about therapeutic goals. In concordance, the patient is informed of the condition and options for treatment and is involved in the treatment decision.¹¹⁷ Seven studies described treatment choice for breast cancer risk reducing medications,^{108,110–115} and three of these also investigated the relationship between physician recommendations and patient choice (Table 8).^110,114,115 This collection of small descriptive studies suggests that women are making decisions based on their concern for side effects as well as their risk for breast cancer.^110–114 Also, women weigh their physicians’ recommendation when deciding whether to use risk reducing medications.^110,114,115 One additional survey of physicians evaluated risk reducing medication prescribing practices.¹⁰⁷ All studies considered tamoxifen use.

Table 8

Descriptive studies of treatment decisions for medications to reduce risk of breast cancer.

In an interview-based cross-sectional study, 17.6% of women were inclined to use tamoxifen following an educational session about its indications and adverse effects.¹¹² More than half of the subjects listed breast cancer (68.8%), pulmonary embolism (67.2%), endometrial carcinoma (62.7% of women without a hysterectomy), and deep vein thrombosis (58.4%) as “very important” in making their decisions.

In a study testing a new decision guide for identifying women with high risk for breast cancer and informing them about risk reduction with tamoxifen, women who were interested in taking tamoxifen were allowed to choose between accepting a prescription for tamoxifen or enrolling in STAR.¹¹¹ Results indicated that 11.8% of women selected tamoxifen, 76.5% declined, and 11.8% were undecided. Major side effects (60.7%) and small benefit from tamoxifen (32.1%) were the most common reasons for declining. However, 90% of women stated that they would take a medication with the same benefit as tamoxifen if it had no side effects. Approximately half of women also stated that if a medication were developed with the same side effects but could eliminate the chance of getting breast cancer, they would take the medication.

In a pre/post survey study, women completed a questionnaire after receiving information about tamoxifen.¹¹³ Of the 43 subjects, 4.7% selected tamoxifen, 34.8% declined, and 60.5% were undecided. Upon later follow-up, none of the 60.5% who were undecided changed to selecting tamoxifen. Of the patients who did not select tamoxifen, 75.6% reported a concern for side effects, including endometrial cancer and thromboembolic events, as a reason for not using tamoxifen. Other reasons were the feeling that not enough information was available (12.2%) and not wanting to discontinue hormone replacement therapy (4.9%).

A telephone survey of 1,287 women with Blue Cross/Blue Shield insurance was designed to determine if women would be “interested in a medication to prevent breast cancer.”¹⁰⁸ The 23% of responders interested in risk reducing medications believed themselves to be at greater risk for breast cancer and were more worried about breast cancer than women who were not interested (p<0.05).

Three studies evaluated the relationship between physician recommendations and treatment choice.^110,114,115

A study of concordance with physician recommendations included women age 35 to 80 years who were evaluated for benign breast findings in a breast clinic.¹¹⁴ They were provided with Gail model estimates of risk and the option of using tamoxifen for risk reduction, and were asked to discuss tamoxifen use with their family physicians. Of the 89 women, 48 discussed the decision with their family physician. Physicians recommended using tamoxifen for 3 women, not using tamoxifen for 37, and made no recommendation and left the decision up to the patient for 8. Only one woman in the study decided to use tamoxifen. While this study did not include raloxifene as a potential breast cancer risk reduction option, another 5 patients reported that their physicians had prescribed raloxifene for osteoporosis with the secondary benefit of breast cancer prevention. Patients identified one or more of the following factors as influencing their decision: concern for adverse effects (46%), breast cancer risk not high enough to warrant therapy (33%), family physician’s decision (31%), personal decision (25%), lack of sufficient information (10%).¹¹⁴

A study of patient/physician concordance assessed women’s decisions to use tamoxifen or raloxifene at 2 and 4 months after risk counseling.¹¹⁰ At two months follow-up, 29% of women chose to take tamoxifen, another 27% opted for enrolling in the STAR trial, 24% declined treatment, and 20% were undecided. At 4 months follow-up, 12% changed from choosing or undecided to decline, however, it was unclear whether anyone who changed from choosing tamoxifen to declining had started taking risk reduction medications in the intervening 2 months. Not all women made a decision by the 4-month follow-up, with 13.9% remaining undecided. All women in this trial were advised by a physician of their eligibility for risk reduction with tamoxifen or raloxifene, however, not all women reported receiving a recommendation from their physician to choose treatment or not. For women who received a recommendation from their physician, most recommendations were related to treatment choice (p<0.0001). Concern for side effects of tamoxifen was a significant factor in women’s treatment decision (p<0.006).¹¹⁰

A descriptive study was designed specifically to evaluate the effect of physician recommendations to women eligible for the NSABP P-1 trial.¹¹⁵ Women were surveyed after attending an informational session about the trial, and 175 of 360 attendees reported having discussed their participation with their primary care physicians and receiving a recommendation for participation or non-participation. Among the 175 women who discussed the decision with their physician, the physician recommendation was related to trial participation (p<0.001). Women whose physicians recommended enrollment were 13 times more likely to enroll than women whose physician recommended against enrollment.

A mailed survey to 350 physicians indicated that 27% prescribed tamoxifen for risk reduction for their patients within the prior 12 months.¹⁰⁷ Physicians who had prescribed tamoxifen were more likely to have a family member with breast cancer (19.8% vs. 8.7%; p=0.004). Prescribers and nonprescribers differed in their responses to several statements including: the benefits of tamoxifen outweigh the risks (62.5% vs. 39.4%; p<0.001), physicians in their community are prescribing tamoxifen for breast cancer prevention (33.3% vs. 16.6%; p<0.001), it is easy to determine who is eligible to take tamoxifen for breast cancer risk reduction (28.1% vs. 10.9%; p<0.001), and many female patients ask for information about taking tamoxifen for breast cancer risk reduction (14.6% vs. 4.8%, p=0.002). Physicians did not differ in their beliefs about the following: whether the evidence that tamoxifen significantly reduces breast cancer is controversial; it is too time consuming to discuss taking tamoxifen with women in my practice; the risk of endometrial cancer is too great to prescribe tamoxifen for breast cancer risk reduction; and, the risk of thromboembolic events is too great to prescribe tamoxifen for breast cancer risk reduction.

Risk stratification models

The Gail model was the first major breast cancer risk stratification model to be used clinically.⁴⁹ This model was derived from multivariate logistic regression analysis of identified risk factors for breast cancer.⁴⁹ In the original version of the Gail model, breast cancer incidence rates and baseline hazard rates were determined for invasive cancer, DCIS, and LCIS from a cohort of women in the Breast Cancer Detection and Demonstration Project (BCDDP). The model was subsequently modified by using U.S. national data for invasive cancer from the Surveillance, Epidemiology, and End Results (SEER) program.¹²⁴ From these data, a model was developed to allow the prediction of individualized absolute risk (probability) of developing breast cancer in women undergoing annual screening mammography.

Subsequent risk stratification models use a similar approach as the Gail model, however, they vary in their use of reference standards. Age-specific breast cancer rates and attributable risk estimates to determine baseline age-specific hazard ratios should ideally be obtained from an applicable population reference standard, such as SEER data in the U.S. Several studies of newer models do not provide information about their reference standards.^{119,120,122,125,131}

Models also vary by the variables they include (Table 10). The original Gail model included age, age at menarche, age of first birth, family history of breast cancer in first degree relatives, number of prior breast biopsies, and history of atypical hyperplasia.⁴⁹ Subsequent models include one or more of these variables in addition to other factors. These include race,^{125,126,129,130} body mass index or height,^{118,119,123,125,128,129} estrogen and progestin use,^{118,119,125,129} parity,^119,125 history of breast feeding,¹²⁵ menopause status or age,^119,123,129 smoking,¹²⁵ alcohol use,^118,119,125 physical activity,^118,125 breast density,^128–130 and diet.¹¹⁸

Table 10

Variables included in risk stratification models.

Studies of calibration

Calibration is a measure of how well predicted probabilities agree with actual observed risk. The calibration of a model refers to its ability to predict the average risk in a subset of the population. When the predicted risk matches the proportion that actually develops disease, a model is considered to be well calibrated. In a perfect prediction model, the predicted risk in a population (% expected) would equal the observed number of cases (% observed) such that the % expected/% observed (E/O) equals 1.0.

Of the nine models reviewed, calibration was calculated for all except the Chen, Chlebowski and Boyle model.^118,125,128 For most models, the expected numbers of cases of breast cancer closely match the observed numbers (Figure 25).^{118,119,121,123–126,128–130} Six studies evaluated the Gail model,^{118,121,122,124,125,132} demonstrating E/O ratios ranging from 0.69 (0.54, 0.90)¹³² to 1.03 (0.88, 1.21).¹²⁴ Two studies reported values <0.90, indicating under prediction of breast cancer cases.^125,132 In one study, under prediction could be attributed to dissimilarities of the study population; women were included who were undergoing assessment at a family history clinic, rather than a primary care setting, were younger than women in other studies, and were not all undergoing routine mammography screening.¹³² The Gail model demonstrated good calibration for estrogen receptor positive cancers (E/O 1.06), but inferior calibration when estrogen receptor negative cancers were included (E/O 0.79).¹²⁵

Figure 25

Calibration of breast cancer risk models. *Chen and Chlebowski Models did not report Calibration Ratio for the models developed in the studies. † Gail model used to determine inclusion for P-1, P-2 trials; measured in RUTH, MORE.

The Gail model was modified to evaluate its utility in a population of Italian women. The Italian Gail Model (IT-GM) differed from the Gail model by one ordinal value for one variable, and the Italian-1 Gail Model (IT1-GM) differed by using categorical rather than ordinal variables.¹²¹ Both versions demonstrated good calibration in two studies.^118,121 In one study, E/O ratio for the IT-GM was 0.96 (0.84, 1.17) and 1.00 (0.88, 1.16) for the IT1-GM.¹²¹ A second study demonstrated good calibration for the IT-GM (E/O 1.04).¹¹⁸ The Gail model itself also demonstrated good calibration in this population (E/O 0.96; 0.84, 1.17;¹²¹ E/O 1.12).¹¹⁸

All of the other models demonstrated good calibration across the studies (E/O 1.00 to 1.09),^{126,129–132} except for the Tyrer-Cuzick model assessing risk in a population with biannual mammography screening (E/O 0.81; 0.62, 1.08).¹³² Categories based on age demonstrated good calibration in the Gail ^121,122,124 and BCSC-Tice models,¹³⁰ except for women <50 years in an Italian population (E/O 0.61; 0.49, 0.80)(Figure 25). When age alone was used to calculate risk of developing breast cancer in an Italian population, breast cancer was under predicted (E/O 0.73; 0.64, 0.86).¹²¹ Two models that include race also demonstrated good calibration, the Gail-AA model for use in the U.S. African American population¹²⁶ and the BCSC-Tice model.¹³⁰

Studies of discriminatory accuracy

Discriminatory accuracy is a measure of how well the model can separate those who do and do not have the disease of interest. In diagnostic testing, it is the ability to identify individuals with or without the disease of interest. In prognostic modeling, it is the ability to correctly classify individuals at higher risk from those at lower risk, and is measured by the model’s concordance statistic or c-stat. The c-stat is determined by the area under the receiver operator curve, a plot of sensitivity (true positive rate) versus 1-specificity (false-positive rate). Perfect discrimination is a c-stat of 1.0 and occurs when all cases attain higher risk scores than all non-cases. A c-stat of 0.5 would result from chance alone. An acceptable level of discrimination is considered as ≥0.70 and <0.80, excellent ≥0.80 and <0.90, and outstanding ≥0.90.¹³³

Thirteen studies of nine models indicate that discriminatory accuracy for most models is <0.70 (Figure 26).^{118,120–122,124–132} Only one study reported levels >0.70 for both the Gail-2 and the Tyrer-Cuzick models, with c-stats of 0.74 (0.67, 0.80) and 0.76 (0.70, 0.82), respectfully.¹³² However, this study was small (<100 cases) and did not include a primary care population, limiting its clinical applicability. The BCSC-Tice model, drawing from large U.S. national populations, provided the next highest discriminatory accuracy, with a c-stat of 0.66 (0.65, 0.66).¹³⁰ The model with the lowest level of discrimination was the Gail-AA, with a c-stat of 0.56.^126,127 The discriminatory accuracies of age^129,131 or breast density alone¹²⁹ as a predictor of breast cancer risk ranged from 0.55 to 0.57 and 0.55 to 0.56, respectfully.

Figure 26

Discriminatory accuracy of breast cancer risk models. *Gail model used to determine inclusion for P-1, P-2 trials; measured in RUTH, MORE. †Models including breast density as risk factor.

Studies of risk quintiles

In some of the breast cancer primary prevention trials, women were assessed for their individual risks for developing breast cancer, and only those meeting established risk thresholds were eligible to participate.^{12,19,23–25} Three studies evaluated this approach to risk stratification by determining calibration and/or discriminatory accuracy based on risk quintiles,^121,125,130 and one study determined these values based on a low (<1.67%) vs. high (≥1.67%) risk threshold (Table 11).¹³⁰ This threshold was used as inclusion criteria in the NSABP P-1 and STAR trials, and is included in the FDA’s approval of the use of SERMS for risk reduction. The BCSC-Tice model demonstrated high calibration (E/O 0.99 to 1.03), and consistent, although low, discriminatory accuracy across the quintiles (c-stat 0.61 to 0.64).¹³⁰ The Gail and the Italian Gail Model demonstrated high calibration in the higher risk quintiles, but variable results in the lower quintiles (Table 11).^121,125

Table 11

Calibration (expected/observed ratio) and discriminatory accuracy of Gail Model quintiles.