A Simulation Model to Predict the Impact of Prophylactic Surgery and Screening on the Life Expectancy of BRCA1 and BRCA2 Mutation Carriers

Inherited mutations in the BRCA1 and BRCA2 (BRCA1/2) cancer susceptibility genes convey high lifetime risks of breast and ovarian cancer, in the range of 40% to 66% and 13% to 46%, respectively (1, 2). Because female BRCA1/2 mutation carriers have 5- to 40-fold higher cancer risks than average-risk women in the United States, a number of intensive risk-reducing strategies are recommended. These interventions include prophylactic mastectomy, prophylactic oophorectomy, breast cancer screening with a combination of mammography and MRI, and chemoprevention with agents such as tamoxifen and raloxifene (3). Because of the relative rarity of BRCA1/2 mutation carriers (4, 5), as well as the invasive nature of several of these risk-reducing interventions, no randomized controlled trial has yet showed the efficacy of prophylactic surgery, cancer screening, or chemoprevention in this high-risk population. Thus, there remains significant controversy about the overall health benefits of several risk-reducing strategies in BRCA1/2 mutation carriers.

Screening guidelines for this high-risk population emphasize the initiation of annual mammograms and contrast-enhanced breast MRI at the age of 25 years (6–8). Because of the poor overall sensitivity of screening mammography among BRCA1/2 mutations, MRI is often recommended in addition to mammography (8, 9) and is regarded as cost-effective, despite being several times more expensive and increasing the rate of false-positive test results (10).

Bilateral prophylactic mastectomy consists of removing both breasts to reduce the risk of developing breast cancer. Studies estimate a 90% to 95% reduction in breast cancer incidence for women with BRCA1/2 mutations who undergo prophylactic mastectomy (11, 12). Despite its benefits, however, the decision to undergo prophylactic mastectomy is complicated, given its potential effects on body image and quality of life.

Prophylactic oophorectomy consists of removing the ovaries and fallopian tubes bilaterally with the primary intent of preventing ovarian and fallopian tube cancers; observational studies in BRCA1/2 mutation carriers estimate its efficacy at 80% to 90% (13–15). Studies also report reduction in breast cancer risk with premenopausal prophylactic oophorectomy, with hazard ratios (HR) of 0.3 to 0.5 (12, 16, 17). The large, prospective PROSE study of BRCA1/2 mutation carriers recently reported an overall mortality reduction, with declines in breast and ovarian cancer–related deaths from early prophylactic oophorectomy (11). However, another study questioned whether BRCA1 versus BRCA2 mutation carriers receive equivalent benefits from prophylactic oophorectomy (14), and a large cohort study of premenopausal oophorectomy in the general population found increased mortality, with excess risk of dementia and Parkinson disease (18–20). Moreover, premature menopause yields increased risk of osteoporotic hip fracture and possibly cardiovascular disease (21, 22).

Given conflicting reports about its benefit, and the highly personal nature of the decision to undergo prophylactic surgery, a clear understanding of the impact of prophylactic oophorectomy and prophylactic mastectomy on life expectancy, in the presence of screening with mammography and MRI, has great clinical relevance. Previously, we developed a computer simulation model to estimate survival among BRCA1/2 mutation carriers undergoing different risk-reducing strategies (23) and adapted this model into an online tool to support shared decision making (24). Here, we provide a detailed technical description of the model and present model-based estimates of the impact of risk-reducing strategies on lifetime expectancy among BRCA1/2 mutations carriers.

We previously developed a Monte Carlo model to simulate the natural history of breast cancer in the average-risk U.S. population, incorporating the effect of screening and treatment interventions (25, 26). We then adapted this model to BRCA1/2 mutation carriers, including the age-specific elevated breast and ovarian cancer risks unique to this population, and used it to estimate the effectiveness and cost-effectiveness of breast screening with MRI (10). Subsequently, this model was modified to incorporate the impact of prophylactic mastectomy and prophylactic oophorectomy on breast and ovarian cancer incidence and survival (23, 24). We reported an evaluation of this model's estimated survival probabilities for BRCA1/2 mutation carriers undergoing various risk reduction strategies (23, 24). We now report on the methodology developed for the model's implementation, with emphasis on the effect of prophylactic surgery on life expectancy in BRCA1/2 mutation carriers. In particular, we present life expectancy estimates for BRCA1/2 mutation carriers undergoing risk reduction strategies at different ages, with sensitivity analysis focused on specific areas of uncertainty, including cancer incidence in women older than 70 years, and the impact of prophylactic oophorectomy on cancer-specific incidence and mortality.

On the basis of published literature, we assumed a 90% reduction in breast cancer risk after prophylactic mastectomy (12). Because of limitations of our natural history model of breast cancer (described below), we assumed that an asymptomatic patient is detected with a malignant breast tumor at time of prophylactic mastectomy if the tumor diameter is greater than 2 mm; however, if the tumor diameter is less than 2 mm, we assume the patient has a 95% probability of being cured, with only a 5% probability of disease progression after prophylactic mastectomy. Furthermore, first and second primary breast cancers are eradicated by prophylactic mastectomy independently of each other and tumors not eradicated by prophylactic mastectomy are detected at the same age, size, and stage they would have been in the absence of prophylactic mastectomy.

If a woman's first primary breast cancer would have been symptomatically detected after she undergoes prophylactic oophorectomy, we generate a new age at symptomatic detection by assuming that the annual probability of symptomatic detection of breast cancer after prophylactic oophorectomy at age i (p_{i, PO}) is |$p_{i{\rm,PO}} = p_{i{\rm,NoPO}}^\alpha$|⁠, where i is greater or equal to the age at prophylactic oophorectomy and α is the risk reduction expressed as a hazard ratio. Simulation is conducted to ensure that woman's age at first cancer diagnosis, under the prophylactic oophorectomy scenario, is greater than or equal to her age at first cancer diagnosis in the absence of prophylactic oophorectomy. On the basis of published reports in women with BRCA1/2 mutations (16), we assumed the HR (α) of 0.5 for women who undergo prophylactic oophorectomy at or after the age of 40 and 0.36 for women who undergo prophylactic oophorectomy before the age of 40. Moreover, prophylactic oophorectomy is assumed to have no effect on breast cancer incidence if the procedure is conducted at age ≥50.

In modeling the effect of prophylactic oophorectomy on the second primary breast cancer, we used a similar approach. The time to a second breast cancer after prophylactic oophorectomy was computed from the time of the first breast cancer's diagnosis in the presence of prophylactic oophorectomy, so that the chronologic order of tumors remained unchanged compared with the no prophylactic oophorectomy scenario. We also ensured that the time interval to a second breast tumor in the presence of prophylactic oophorectomy was no shorter than the same interval under the no prophylactic oophorectomy scenario.

On the basis of published literature (13), we assumed 80% reduction in ovarian cancer risk after prophylactic oophorectomy. Furthermore, as prophylactic oophorectomy removes tissue at risk of developing ovarian cancer, this reduction is treated as a probability of ovarian cancer eradication. If the cancer is not eradicated by prophylactic oophorectomy, we assumed that the age at symptomatic detection is unchanged compared with the scenario of no prophylactic oophorectomy.

We define the “first primary breast cancer” as the breast cancer that would have been diagnosed first chronologically, due to symptoms, in the absence of any intervention. To model incidence of a first primary breast cancer for BRCA1/2 mutation carriers, we leveraged a large population-based meta-analysis which estimated the age-specific incidence of breast and ovarian cancer in this population (1). Because age-specific breast and ovarian cancer incidence for women who choose not to undergo prophylactic oophorectomy is a key input to our model and this meta-analysis did not include information on prophylactic oophorectomy use (1), we derived this key model input by assuming that the meta-analysis constituted a mixture of women who chose premenopausal prophylactic oophorectomy and women who did not. In Supplementary Appendix SI, we present the algorithm that we developed to estimate age-specific breast cancer incidence in the absence of risk-reducing interventions.

BRCA1/2 mutation carriers have high risks of a second primary breast cancer, once they have been diagnosed with a first breast cancer. On the basis of a prospective cohort study for the subpopulation of BRCA1 mutation carriers who did not undergo premenopausal oophorectomy or take tamoxifen (11), the 5-year risk of a contralateral second primary breast cancer is 27.1% and the corresponding 10-year risk is 43.4%. For a similar subpopulation of BRCA2 mutation carriers, the corresponding 5- and 10-year risks are 23.5% and 34.6%, respectively. To model the time to a second primary breast cancer as a continuous variable, we assumed a 2-parameter Weibull distribution and estimated parameters by fitting to 5- and 10-year risks separately for BRCA1 and BRCA2 mutation carriers.

To predict the impact of screening on breast cancer incidence and staging, we model the natural history of the disease and the ability of the screening test to detect the disease before symptoms arise, as in our prior work (10, 26).

The assumptions underlying our natural history model of BRCA1/2-associated breast cancers are similar to those we previously used for the average-risk female U.S. population (26, 27) but incorporate modifications to reflect the histopathologic characteristics of the breast cancers in BRCA1/2 mutation carriers. We previously reported a detailed description of our assumptions related to tumor growth, tumor volume doubling time (TVDT), and hazard functions characterizing the transition times from local to regional and distant stages and the time of symptomatic detection (26, 27). We also provide an overview of these assumptions and a description of the model's parameters in the Supplementary Appendix SII.

For the average-risk population, we obtained maximum likelihood estimates of our natural history model parameters using data on tumor size and stage at diagnosis obtained from the Surveillance Epidemiology and End Results (SEER) registry in the era before mammographic screening (26). However, clinical studies have found that histopathologic characteristics of BRCA1/2-associated breast cancers, such as hormone receptor status and grade, differ from those of women with sporadic breast cancers. In particular, BRCA1 mutation carriers are significantly more likely to be diagnosed with grade III, hormone receptor–negative breast cancers than other women (28). Because higher tumor grade is associated with more aggressive clinical behavior, we modeled them as faster growing. In particular, we stratified our natural history model by breast tumor grade: low (grade I–II) versus high (grade III). Model parameter estimation by grade was based on SEER data for women diagnosed at ages 40 to 65 years between 1975 and 1981, using only the records with tumor grade available. The complete set of estimated natural history model parameters for both low- and high-grade tumors may be found in Supplementary Table A.1 of Supplementary Appendix SII.

For parameter identifiability, two constraints were imposed. First, we assumed that for a given TVDT, the distribution of tumor size at symptomatic detection is the same for low- and high-grade tumors. Second, we fixed the mean TVDT for low-grade tumors, which allowed us to compute ratio of the mean TVDT for high-grade to the mean TVDT of low-grade tumors as 0.54.

We derived the absolute values of the mean TVDT for low- and high-grade breast tumors by calibrating to a 1-year sensitivity of breast MRI for a cancer detection of 85% for BRCA1/2 mutation carriers (29–31). We obtained an overall mean TVDT for BRCA1 and BRCA2 mutation carriers by using reported proportions of low- versus high-grade tumors in each population (17, 28, 32).

Because of the absence of empiric data, we assumed that estrogen receptor (ER) status was determined by grade (17, 32). For low-grade tumors, the proportion of ER-positive breast cancers was 91% for BRCA1 and 94% for BRCA2 mutation carriers. On the other hand, for high-grade tumors, the proportion of ER-positive breast cancers was 18% for BRCA1 and 61% for BRCA2 mutation carriers. We did not model borderline cases of ER status.

From the time of symptomatic diagnosis, we applied breast cancer survival curves for the general U.S. population, using SEER data for ages 40 to 65 in the prescreening era (1975–1981). We stratified these survival curves according to tumor size (<2, 2–5, and ≥5 cm), SEER historic stage (local, regional, distant), and tumor grade (I + II vs. III). We used the same survival curves for first and second primary breast cancer. We took the same approach as described in our prior work to model survival after breast cancer is detected by screening or at the time of prophylactic mastectomy (26).

Following ovarian cancer diagnosis, we applied survival curves on the basis of ovarian cancer cases diagnosed during 1975 to 2001, as recorded in the SEER 9 registries. We stratified these curves by 5-year age groups. On the basis of prior publications which have reported on histologic subtypes of ovarian cancers associated with BRCA1/2 mutations (33–35), we used only the following ICD-O-3 histologic codes: 8000-8001, 8005, 8010, 8020, 8050, 8140, 8230, 8260, 8310, 8380, 8440-8441, 8450, 8460-8461, 8470-8471, and 8480-8481.

We derived mortality from causes other than breast and ovarian cancer from the Berkeley Mortality Tables (36), according to birth cohort and age at death. We modified these tables by subtracting deaths due to breast and ovarian cancer, using 2004 cross-sectional death rates from the Centers for Disease Control (CDC; ref. 37). We translated CDC death rates, which were reported in 5-year age intervals, into single-year rates by linear interpolation under the assumption that reported rates correspond to the middle of the age interval (i.e., ages 22, 27, 32, 37…92) and that rates for ages <20 years are equal to 0. Death rates for ages 92+ were held constant. We converted annual probabilities of death (q_i) during age intervals (i, i + 1), as reported in the Berkeley Mortality Tables, to death rates (r_i) by assuming exponential distribution: r_i = −ln (q_i). Then, the annual probability of death from causes other than breast and ovarian cancer was computed as |$q_i^{ - ({{\rm BC + OC}})} = \exp ({ -({r_i - r_i^{{\rm BC}} - r_i^{{\rm OC}}})} \right)$|⁠.

We modeled an increase in deaths from cardiovascular disease, hip fracture, and dementia after premenopausal prophylactic oophorectomy according to previous reports (18, 20, 21). We used death rates as reported by the CDC (37) in 2004 for the general U.S. population, from cardiovascular disease (⁠|$_i^{{\rm CV}}$|⁠; ICD-10: I20-I25) and dementia (⁠|$r_i^{{\rm Dem}}$|⁠; ICD-10: F01, F03, G20-G21, G30). We used estimates for death rates because of hip fractures |$r_i^{{\rm Hip}}$| after prophylactic oophorectomy based on a previously published analysis of oophorectomy outcomes (38). For ages greater than or equal to a woman's age at prophylactic oophorectomy, we computed annual other-cause mortality probabilities as:

where HR^CV, HR^Dem, HR^Hip are hazard ratios associated with cardiovascular disease (21), dementia (18, 20, 22), and hip fracture (22), respectively.

In prior work (10, 23, 24), we found that overall survival gain under the various interventions is sensitive to variations in assumptions about cancer risks, MRI sensitivity for breast cancer diagnosis, and the HR for breast cancer incidence after premenopausal prophylactic oophorectomy. Here we conducted sensitivity analysis on 3 different sets of parameters, namely, those used to calculate: (i) breast cancer incidence estimates obtained by removing the effect of prophylactic oophorectomy from the work of Antoniou and colleagues (1), (ii) cancer incidence after the age of 69, and (iii) the duration of the protective effect of prophylactic oophorectomy on breast cancer risk after surgery. To date, there is no evidence indicating that the protective effect of prophylactic oophorectomy on breast cancer lasts more than 15 years after surgery (16) and little is reported on the breast cancer incidence of BRCA1/2 mutation carriers at older ages.

The sensitivity analysis conducted on our breast cancer incidence estimates from removing the effect of prophylactic oophorectomy compare the cumulative incidence of breast cancer by the age of 70 when varying assumptions related to the efficacy of the protective effect of prophylactic oophorectomy over breast cancer (α), the age and percentage of women undergoing prophylactic oophorectomy.

The sensitivity analyses regarding cancer incidence after the age of 69 considered 2 scenarios. In scenario A, we assumed that cancer risk beyond this age is similar to that of the general population, based on the SEER registry. In scenario B, we assumed that the risk for this population is still higher than that of the general population; hence, cancer incidence was kept constant for years following the last period (age, 67–69 years) in which BRCA1/2 mutation–associated incidence was estimated from the work of Antoniou and colleagues (1).

Regarding the duration of the protective effect of prophylactic oophorectomy on risk of breast cancer, our sensitivity analysis varied this parameter from 10 to 15 years after undergoing the procedure. This particular sensitivity analysis required use of the algorithm reported in the Methods section to remove the effect of prophylactic oophorectomy and compute annual age-specific breast cancer incidences for each case.

Figure 1A and B shows the estimated annual probabilities of a first primary breast cancer in the absence of risk-reducing interventions for BRCA1 and BRCA2 mutation carriers, respectively. Compared with the breast cancer risk in the work of Antoniou and colleagues, we report a greater cumulative risk of breast cancer by the age of 70 (0.65 vs. 0.71 for BRCA1 and 0.45 vs. 0.50 for BRCA2 mutation carriers) because we removed the protective effect of premenopausal prophylactic oophorectomy on subsequent breast cancer risk.

Table 1 shows the gains in life expectancy resulting from prophylactic intervention and screening, for BRCA1 and BRCA2 mutation carriers, when simulating one million women for each model run. Because of differences in the annual incidence of breast and ovarian cancer, gains in life expectancy from prophylactic mastectomy and prophylactic oophorectomy are highly dependent on the mutated gene (BRCA1 vs. BRCA2) and age at the time of the intervention. With no intervention, the remaining life expectancy of a 30-, 40-, and 50-year-old BRCA1 mutation carrier who has never had cancer is 41.5, 32.7, and 26.1 years, respectively; the remaining life expectancies of BRCA2 mutation carriers of the same ages are 48.6, 39.4, and 30.7 years. We present the effect of a single risk-reducing strategy as well as the effect of combining different strategies. For example, the life expectancy gain for 30-year-old BRCA1 mutation carriers is 5.2, 7.4, or 10.3 years, if she chooses prophylactic mastectomy alone, prophylactic oophorectomy alone, or both, respectively.

Given that many women may delay prophylactic mastectomy and prophylactic oophorectomy due to reproductive or personal preferences, we present the gains in life expectancy when the prophylactic procedures are delayed 5 or 10 years after carrier status determination. Figure 2A and B compares the life expectancies of 30-, 40-, and 50-year-old BRCA1 and BRCA2 mutation carriers, respectively, under different strategies, when screening is initiated from the time of BRCA status determination (mammography at age 40 if age at status determination is >40) until the age of 70 and prophylactic surgery is conducted at various ages up to the age of 60.

Delaying prophylactic surgery from age 30 to 40, a comparatively low-risk decade for BRCA1/2 mutation carriers, does not affect life expectancy as much as would a longer delay. For example, if a BRCA1 mutation carrier of age 30 delays prophylactic oophorectomy for 10 years, she forfeits an expected gain of 1.9 years (7.4 vs. 5.5 years); in comparison, for a 40-year-old, a 10-year delay costs 3.3 years of life expectancy (6.1 vs. 2.8 years).

Table 2 shows the sensitivity analysis conducted by varying the parameters used to calculate breast cancer incidence in the absence of risk-reducing interventions. We varied presumed percentage of prophylactic oophorectomy and age at prophylactic oophorectomy in the meta-analysis cohort of Antoniou and colleagues (1) and reported the cumulative probability of breast cancer by the age of 70. For example, we obtain a 76% probability of developing breast cancer for BRCA1 mutation carriers when 30% of women at age 50 years underwent prophylactic oophorectomy and its protective effect had a hazard reduction of 0.8. We find that the cumulative breast cancer incidence by the age of 70 is highly sensitive to variations in the protective effect of prophylactic oophorectomy on breast cancer incidence and the percentage of women with BRCA1/2 mutations who underwent prophylactic oophorectomy. The estimates for breast cancer incidence by the age of 70 are less sensitive to the age at which prophylactic oophorectomy is conducted, in part, because the tested age range is quite narrow. We chose to vary this parameter only between ages 40 and 50 because prophylactic oophorectomy is not likely to be conducted before the age of 40 and has been reported to have a small effect on breast cancer incidence for postmenopausal women, typically older than 50 (16).

Results presented in Table 3 show sensitivity of the different risk-reducing procedures to the variation in the breast cancer incidence among the older women. It is assumed that age of BRCA1/2 mutation status determination is fixed at 30 years. The effect on life expectancy of the different assumptions on the breast cancer incidence in older women (age > 69 years) does not exceed 0.2 years as compared with the base case (0.3 years for BRCA2 screening and 10 years delayed prophylactic mastectomy and prophylactic oophorectomy). As is expected for scenario A, where incidence for older women is similar to that of average risk population, the effect of prophylactic procedures is smaller than that for the base case. In Table 4, we present sensitivity of the effectiveness of different risk-reducing strategies to varying assumptions on the duration of the prophylactic oophorectomy protective effect on the breast cancer incidence. This sensitivity analysis was conducted under the assumption that age of BRCA1/2 mutation status determination is 30 and 40. Assuming a 10-year window of breast cancer risk reduction from prophylactic oophorectomy, a 30-year-old BRCA1 mutation carrier gains 4.4 years from immediate prophylactic oophorectomy, whereas 5- and 10-year delays offer slight gains of 4.6 and 4.5 years, respectively.

We report on the gains in life expectancy from conducting prophylactic surgery and screening among women with BRCA1/2 mutations, at different ages of genetic testing and risk-reducing interventions. In general, the observed gain in life expectancy from premenopausal prophylactic oophorectomy is greater than that provided by prophylactic mastectomy, as prophylactic oophorectomy reduces the risk of both breast and ovarian cancer. However, screening with annual mammography plus breast MRI provides significant gains in life expectancy, although its benefits are less than those of prophylactic mastectomy or prophylactic oophorectomy.

The greatest life expectancy gain is provided by prophylactic mastectomy + prophylactic oophorectomy at the time of genetic testing for BRCA1/2 mutations. We found that gains in life expectancy are highly dependent on the timing of prophylactic surgery. However, delaying prophylactic surgery can still provide an important gain in life expectancy, especially in tandem with annual breast screening.

In sensitivity analysis, assuming a limited duration for prophylactic oophorectomy's reduction in breast cancer risk renders timing even more critical: prophylactic oophorectomy can provide an equivalent or greater life expectancy increase when timed to affect the years of high risk. The impact of prophylactic oophorectomy on breast cancer risk reduction was the most influential variable in our sensitivity analysis.

Several prior studies have investigated the impact of prophylactic mastectomy and prophylactic oophorectomy on survival of BRCA1/2 mutation carriers (39–43). Most used a Markov model for decision analysis, which is conceptually different from the approach of explicitly modeling cancer growth and progression that was adopted here. Our modeling approach is closest to that of Schrag and colleagues (43) who reported a comparable life expectancy gain from prophylactic mastectomy at 2.9 to 5.3 years for 30-year-old women but a lower gain from prophylactic oophorectomy (0.3–1.7 years), likely due to their assumption that prophylactic oophorectomy reduces ovarian, but not breast cancer risk; other early articles showed similar results (40, 43). More recent studies suggest that BRCA1 mutation carriers benefit more from prophylactic oophorectomy than prophylactic mastectomy, with the reverse finding for BRCA2 mutation carriers (39, 41); our results agree with the exception that we report a larger life expectancy gain from prophylactic oophorectomy. Consistent with other studies, we find a greater life expectancy from prophylactic mastectomy or prophylactic oophorectomy than from breast screening alone (41), although the life expectancy afforded by annual screening mammography plus breast MRI is significant, particularly in combination with prophylactic oophorectomy. The recent simulation model–based study (42) investigated effect of different screening modalities and strategies alone, including MRI, film and digital mammography on the life expectancy of the BRCA1/2 mutation carriers. In that study, the maximum gain of 1.86 years for BRCA1 (range, 1.3–1.86) and 1.76 years (range, 1.46–1.76) for BRCA2 mutation carriers was attained for alternating digital mammography with MRI at 6 months intervals starting at the age of 25. Results are comparable with our results for BRCA2 and somewhat lower for BRCA1 mutation carriers.

We used simulation modeling to estimate the life expectancy of women with BRCA1/2 mutations who chose to undergo prophylactic surgery for cancer risk reduction at various ages. To address limitations of our model-based approach, we conducted sensitivity analysis on deep model parameters. In prior work, we identified limitations in clinical assumptions of our model (10, 23, 24). Of note, chemoprevention is not included in our current analysis as an alternative to prophylactic surgery because its efficacy is still being evaluated among mutation carriers. Despite limitations of simulation modeling, it may offer the most practical approach to urgent questions for patient care when randomization is infeasible. Our model integrates existing data to quantify the overall effect of cancer risk-reducing strategies, readily incorporates emerging data, and enables a risk–benefit calculation customized for an individual woman's age and BRCA1/2 mutation.

No potential conflicts of interest were disclosed.

Conception and design: B.M. Sigal, A.W. Kurian, S.K. Plevritis

Development of methodology: B.M. Sigal, D.F. Munoz, S.K. Plevritis

Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): D.F. Munoz, S.K. Plevritis

Analysis and interpretation of data (e.g., statistical analysis, biostatistics, computational analysis): B.M. Sigal, D.F. Munoz, S.K. Plevritis

Writing, review, and/or revision of the manuscript: B.M. Sigal, D.F. Munoz, A.W. Kurian, S.K. Plevritis

Administrative, technical, or material support (i.e., reporting or organizing data, constructing databases): D.F. Munoz, S.K. Plevritis

Study supervision: S.K. Plevritis

This work was supported by the NIH (grant numbers U01 CA088248, U01CA152958), the Robert Wood Johnson Foundation (Physician Faculty Scholars Award, grant number 64317), and a Stanford Cancer Institute Developmental Research Award.

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.