Survival Outcomes After Contralateral Prophylactic Mastectomy
A Markov model is a recursive decision tree that guides a hypothetical cohort between mutually exclusive health states depending on transition probabilities obtained from published data. We developed a Markov state-transition model to simulate survival outcomes after CPM and no CPM for women with stage I and II breast cancer without BRCA mutations (Figure 1). The model simulates the long-term prognosis of hypothetical cohorts of women with newly diagnosed unilateral breast cancer under two scenarios: 1) CPM (ie, double mastectomy) and 2) no CPM (assuming that women undergo either lumpectomy with radiation therapy or unilateral mastectomy). We projected the benefit of CPM for cohorts of women defined by age at breast cancer diagnosis (40, 50, or 60 years), stage of primary breast cancer (I, II), and estrogen receptor (ER) status (positive, negative).
(Enlarge Image)
Figure 1.
Markov model. Women in each Markov state (no contralateral breast cancer [CBC], CBC stage I, CBC stage II, CBC stage III, CBC stage IV) can die from other causes or die from breast cancer. Each cycle of the Markov model is 1 year. The model ran the lifetime of the cohort. For example, the model predicting life expectancy in the 40-year-old cohort ran 61 cycles to obtain data through age 100 (lifetime). The model predicting life expectancy for the 50-year-old cohort ran 51 cycles, and the model for the 60-year-old cohort ran 41 cycles. To generate 20-year survival curves we only used model output for the first 20 years. CPM = contralateral prophylactic mastectomy; ER = estrogen receptor.
The model tracks each cohort of women through health states over time. Each year after treatment of the ipsilateral cancer, women may die from their primary breast cancer, develop CBC, or experience no adverse event. After development of CBC, women are at an increased risk of dying from breast cancer (ie, the risk associated with their ipsilateral and contralateral cancers). Data from the 2008 life tables for US women were used to incorporate the age-specific annual risk of dying from other causes. Model output for each strategy consisted of LE, overall survival, and disease-free survival. The model was programmed using TreeAge Pro 2012 (TreeAge Software, Williamstown, MA).
The probabilities used in baseline analyses and the ranges evaluated in sensitivity analyses are listed in Table 1.
Primary Breast Cancer. We derived stage-specific breast cancer mortality rates from the relative survival curves reported in the Surveillance, Epidemiology, and End Results (SEER) data. We used SEER stat to obtain 20-year relative survival curves for patients with stage I or II breast cancer, where stage was defined by the American Joint Committee on Cancer SEER modified staging system. SEER reports a breast cancer–specific mortality risk (ie, 1 − relative survival percentage) for women with stage I breast cancer of 1.8% at 10 years and 10.0% at 20 years. For women with stage II breast cancer, cancer-specific mortality was 23.1% at 10 years and 42.2% at 20 years.
Contralateral Breast Cancer. We assumed the stage-specific mortality associated with CBC was the same as reported by SEER. For patients who developed CBC, we added the stage-specific cancer mortality rate of their ipsilateral cancer to the stage-specific cancer mortality rate of their contralateral cancer.
Several studies have evaluated the risk of developing CBC. For our base-case values, we used the recent meta-analysis from the Early Breast Cancer Trialists' Collaborative Group (EBCTCG) that reported an annual probability of invasive CBC of approximately 0.4% for patients with ER-positive breast cancer treated with tamoxifen and approximately 0.5% for patients with ER-negative breast cancer. All age, tumor, and treatment subgroups had probabilities less than 0.7% per year. We assumed that every woman in our cohort with ER-positive breast cancer was treated with endocrine therapy for our base-case analysis. Therefore, in our model at baseline we used an annual probability of developing CBC of 0.4% in ER-positive patients and 0.5% in ER-negative patients, varying from 0.2% to 0.7% in our sensitivity analysis to capture uncertainty and differences to treatment adherence to endocrine therapy.
CBC Stage. Using the Oregon State Cancer Registry database, Quan et al. reported that more than 90% of CBCs were either ductal carcinoma in situ or early-stage breast cancer. To capture the maximum potential benefit of CPM, we modeled invasive breast cancer only as this would impact survival and used CBC probabilities reported by Quan et al. after excluding ductal carcinoma in situ. We estimated that the probability of developing stage I CBC was 67%, the probability of developing stage II CBC was 24%, the probability of developing stage III CBC was 5%, and the probability of developing stage IV CBC was 4%. We used the stage distribution reported by SEER for primary breast cancer presentation in a sensitivity analysis (Table 1).
Effectiveness of Contralateral Prophylactic Mastectomy. Several studies have demonstrated that CPM is effective in reducing the risk of CBC (relative risk reduction: 83% to 97%). We assumed that CPM reduced the annual risk of CBC by 90% in our base-case analysis. Because breast cancer surgery is associated with a very small risk of mortality, we did not incorporate surgical mortality into our model.
We assumed that the survival rates were the same after mastectomy as compared with lumpectomy and radiation for treatment of the affected breast. Thus, all of the survival benefit of bilateral mastectomies (ie, CPM) is obtained from removing the unaffected contralateral breast.
We performed sensitivity analyses to assess the stability of results to variation in the base-case parameter estimates. The variables analyzed in the sensitivity analysis included probability of CBC, stage of CBC, and effectiveness of CPM. When several published point estimates were available for a particular parameter, we evaluated the full range of published estimates. In instances in which there were limited available published data and uncertainty for a variable estimate (eg, stage of CBC), we varied our base-case estimate over the broadest range that seemed plausible. The probabilities used in baseline analyses and the ranges evaluated in sensitivity analyses are listed in Table 1.
Methods
Model Design
A Markov model is a recursive decision tree that guides a hypothetical cohort between mutually exclusive health states depending on transition probabilities obtained from published data. We developed a Markov state-transition model to simulate survival outcomes after CPM and no CPM for women with stage I and II breast cancer without BRCA mutations (Figure 1). The model simulates the long-term prognosis of hypothetical cohorts of women with newly diagnosed unilateral breast cancer under two scenarios: 1) CPM (ie, double mastectomy) and 2) no CPM (assuming that women undergo either lumpectomy with radiation therapy or unilateral mastectomy). We projected the benefit of CPM for cohorts of women defined by age at breast cancer diagnosis (40, 50, or 60 years), stage of primary breast cancer (I, II), and estrogen receptor (ER) status (positive, negative).
(Enlarge Image)
Figure 1.
Markov model. Women in each Markov state (no contralateral breast cancer [CBC], CBC stage I, CBC stage II, CBC stage III, CBC stage IV) can die from other causes or die from breast cancer. Each cycle of the Markov model is 1 year. The model ran the lifetime of the cohort. For example, the model predicting life expectancy in the 40-year-old cohort ran 61 cycles to obtain data through age 100 (lifetime). The model predicting life expectancy for the 50-year-old cohort ran 51 cycles, and the model for the 60-year-old cohort ran 41 cycles. To generate 20-year survival curves we only used model output for the first 20 years. CPM = contralateral prophylactic mastectomy; ER = estrogen receptor.
The model tracks each cohort of women through health states over time. Each year after treatment of the ipsilateral cancer, women may die from their primary breast cancer, develop CBC, or experience no adverse event. After development of CBC, women are at an increased risk of dying from breast cancer (ie, the risk associated with their ipsilateral and contralateral cancers). Data from the 2008 life tables for US women were used to incorporate the age-specific annual risk of dying from other causes. Model output for each strategy consisted of LE, overall survival, and disease-free survival. The model was programmed using TreeAge Pro 2012 (TreeAge Software, Williamstown, MA).
Data Sources
The probabilities used in baseline analyses and the ranges evaluated in sensitivity analyses are listed in Table 1.
Cancer Incidence and Prognosis
Primary Breast Cancer. We derived stage-specific breast cancer mortality rates from the relative survival curves reported in the Surveillance, Epidemiology, and End Results (SEER) data. We used SEER stat to obtain 20-year relative survival curves for patients with stage I or II breast cancer, where stage was defined by the American Joint Committee on Cancer SEER modified staging system. SEER reports a breast cancer–specific mortality risk (ie, 1 − relative survival percentage) for women with stage I breast cancer of 1.8% at 10 years and 10.0% at 20 years. For women with stage II breast cancer, cancer-specific mortality was 23.1% at 10 years and 42.2% at 20 years.
Contralateral Breast Cancer. We assumed the stage-specific mortality associated with CBC was the same as reported by SEER. For patients who developed CBC, we added the stage-specific cancer mortality rate of their ipsilateral cancer to the stage-specific cancer mortality rate of their contralateral cancer.
Several studies have evaluated the risk of developing CBC. For our base-case values, we used the recent meta-analysis from the Early Breast Cancer Trialists' Collaborative Group (EBCTCG) that reported an annual probability of invasive CBC of approximately 0.4% for patients with ER-positive breast cancer treated with tamoxifen and approximately 0.5% for patients with ER-negative breast cancer. All age, tumor, and treatment subgroups had probabilities less than 0.7% per year. We assumed that every woman in our cohort with ER-positive breast cancer was treated with endocrine therapy for our base-case analysis. Therefore, in our model at baseline we used an annual probability of developing CBC of 0.4% in ER-positive patients and 0.5% in ER-negative patients, varying from 0.2% to 0.7% in our sensitivity analysis to capture uncertainty and differences to treatment adherence to endocrine therapy.
CBC Stage. Using the Oregon State Cancer Registry database, Quan et al. reported that more than 90% of CBCs were either ductal carcinoma in situ or early-stage breast cancer. To capture the maximum potential benefit of CPM, we modeled invasive breast cancer only as this would impact survival and used CBC probabilities reported by Quan et al. after excluding ductal carcinoma in situ. We estimated that the probability of developing stage I CBC was 67%, the probability of developing stage II CBC was 24%, the probability of developing stage III CBC was 5%, and the probability of developing stage IV CBC was 4%. We used the stage distribution reported by SEER for primary breast cancer presentation in a sensitivity analysis (Table 1).
Effectiveness of Contralateral Prophylactic Mastectomy. Several studies have demonstrated that CPM is effective in reducing the risk of CBC (relative risk reduction: 83% to 97%). We assumed that CPM reduced the annual risk of CBC by 90% in our base-case analysis. Because breast cancer surgery is associated with a very small risk of mortality, we did not incorporate surgical mortality into our model.
We assumed that the survival rates were the same after mastectomy as compared with lumpectomy and radiation for treatment of the affected breast. Thus, all of the survival benefit of bilateral mastectomies (ie, CPM) is obtained from removing the unaffected contralateral breast.
Sensitivity Analysis
We performed sensitivity analyses to assess the stability of results to variation in the base-case parameter estimates. The variables analyzed in the sensitivity analysis included probability of CBC, stage of CBC, and effectiveness of CPM. When several published point estimates were available for a particular parameter, we evaluated the full range of published estimates. In instances in which there were limited available published data and uncertainty for a variable estimate (eg, stage of CBC), we varied our base-case estimate over the broadest range that seemed plausible. The probabilities used in baseline analyses and the ranges evaluated in sensitivity analyses are listed in Table 1.