Skip to main content
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2026 Oct 23.
Published in final edited form as: J Surg Res. 2026 Jun 23;325:336–346. doi: 10.1016/j.jss.2026.04.040

CBCRisk-Mastectomy: A Prediction Tool to Aid Contralateral Prophylactic Mastectomy Decision-Making

Ibrahim H Sajal 1, Ruth M Pfeiffer 2, Ismail Jatoi 3, Mitchell H Gail 2, Reena S Cecchini 4, Pankaj K Choudhary 1,*, Swati Biswas 1,*
PMCID: PMC13690293  NIHMSID: NIHMS2191229  PMID: 42335630

Abstract

Purpose:

Unilateral breast cancer (BC) patients scheduled for mastectomy often choose to undergo contralateral prophylactic mastectomy (CPM), despite substantial declines in contralateral breast cancer (CBC) risk in recent decades. Models predicting absolute risk of future CBC can aid informed decision-making about CPM. CBCRisk is an existing CBC absolute risk prediction model trained on unilateral BC patients regardless of whether they had mastectomy. Here we developed CBCRisk-Mastectomy, tailored specifically to BC patients scheduled for mastectomy and considering CPM.

Patients and Methods:

We used data on BC patients who underwent mastectomy to treat their first BC from two nationally representative sources: Breast Cancer Surveillance Consortium (BCSC) and Surveillance, Epidemiology, and End Results (SEER) cancer registry. We imputed missing data in the BCSC sample and used conditional logistic regression models, trained on 2,660 BC patients (665 CBC cases) from BCSC, to identify predictors and estimate relative risks (RRs). These were combined with attributable risks and CBC incidence rates to obtain absolute risk. Cross-validation was used to internally validate CBCRisk-Mastectomy and compare with CBCRisk.

Results:

CBCRisk-Mastectomy has nine predictors: first BC type, lobular carcinoma in situ status, estrogen receptor status, tumor stage, breast density, age at BC diagnosis, family history of BC, age at first birth, and body mass index. The areas under the curve and their 95% confidence intervals for 5-year predictions for CBCRisk-Mastectomy and CBCRisk were 0.62 (0.59, 0.65) and 0.58 (0.55, 0.61), respectively.

Conclusions:

CBCRisk-Mastectomy may aid clinicians in counseling BC patients scheduled for mastectomy, enabling improved decision-making regarding CPM.

Keywords: Breast Cancer Surveillance Consortium; Surveillance, Epidemiology, and End Results; CBCRisk; breast density

Introduction

Women affected with unilateral breast cancer (BC) are naturally worried about the risk of developing cancer in the opposite, unaffected breast, referred to as contralateral breast cancer (CBC). As a preventive measure, unilateral BC patients whose treatment involves mastectomy (surgical removal of the affected breast) may choose to undergo a contralateral prophylactic mastectomy (CPM), i.e., removal of the contralateral unaffected breast, at the same time as the mastectomy for the breast with the initial BC. Since the late 1990s, CPM rates have risen significantly in the United States (US), with a nearly six-fold increase between 1998 and 20111. For US women aged 20–44 years, diagnosed with invasive unilateral early-stage BC that is treated with surgery, CPM rates rose from 10.5% in 2004 to 33.3% in 2012, reaching as high as 42% in some US states 2. A recent study conducted on 55,060 unilateral BC patients in the US reported a significant increase in CPM rates from 2016 to 2019, with the highest rates among younger women3. Another recent study found tripling of the CPM rate from 21.8% (in 2000–2009) to 64.4% (in 2017–2021) over two decades in a rural population4.

Although CPM is effective in reducing CBC risk, recent advances in adjuvant systemic therapies have also contributed to a decline in CBC incidence. In the US, CBC incidence decreased by 3% annually from 1985 to 20065. A systematic review of nine studies, stratified by the median year of first BC diagnosis as 1989–1994, 1995–1999, and 2000–2003, reported declining median annual CBC incidence rates of 0.5%, 0.4%, and 0.2%, respectively6. The standardized incidence ratio of CBC among BC patients who did not undergo CPM compared to BC patients in the general population dropped from 2.49 in 1992–1997 to 1.57 in 2010–2015, suggesting that the advances in systemic therapies may be the primary driver of this decline7.

CPM is an aggressive approach for CBC prevention, yet it is not clear that it prolongs overall survival among BC patients without BRCA1/2 gene mutation8–10. Although several studies have reported improved overall survival among women with CPM, these are likely attributable to unmeasured confounders, and thus, are not causal11. Affluent non-Hispanic White women, those with private insurance, or higher socioeconomic status are more likely to opt for CPM followed by breast reconstruction12–14. While CPM may be medically justifiable for women at a high risk of CBC, there is no evidence of benefit for those with early-stage unilateral BC and low-to-average CBC risk15. The physical, psychological, and financial impacts of CPM can be substantial, including a lengthy recovery period, loss of breast sensation, and potential need for re-operations16,17. Compared to unilateral mastectomy, bilateral mastectomy carries higher risks of post-operative complications, more emergency room visits during the immediate post-operative period, and longer hospital stays17–19.

The decision to undergo CPM is often driven by a desire to reduce CBC risk, alleviate anxiety about CBC, and achieve breast symmetry. Yet risk reduction is minimal for most women because of their low initial CBC risk (notable exceptions are BRCA1/2 carriers or patients who have received mantle radiotherapy for Hodgkin lymphoma)18,20,21. Informing women about their risk of developing CBC based on personal characteristics can help them make more informed decisions about choosing the optimal surgical treatments, and, if CBC risk is low, avoid unnecessary CPMs22, and reduce anxiety of CBC23,24.

CBCRisk is an existing risk prediction model that estimates the absolute risk of developing CBC in women with unilateral BC23,25. It was trained on nationally representative data from the Breast Cancer Surveillance Consortium (BCSC) and the National Cancer Institute’s (NCI’s) Surveillance, Epidemiology, and End Results (SEER 17) database26,27. The model provides CBC risk predictions for all unilateral BC patients and was built using the same methodology as the NCI’s Breast Cancer Risk Assessment Tool (BCRAT), also known as “the Gail Model”, which predicts personalized risk of a first BC28. However, only those BC patients who undergo mastectomy for treatment of their initial BC, typically opt for CPM. Therefore, a CBC model trained solely on such patients is likely to be more useful for decision making about CPM. In this study, we developed CBCRisk-Mastectomy, which is specifically tailored for patients undergoing mastectomy and considering CPM.

Data Description

BCSC data:

BCSC is a network of US breast imaging registries26. This study used data from 5 active and 2 historical registries. Our study cohort included women who received their first BC diagnosis at ages 18–88 years between 1995 and 2009 and whose first BC was either DCIS or invasive. Upon applying the exclusion criteria described in the online supplement, the study cohort consisted of 24,962 women.

A woman was classified as a CBC case if she was diagnosed with invasive BC or DCIS in the opposite healthy breast at least six months after her initial BC diagnosis. Each CBC case was matched to three controls based on race, year of diagnosis, and equal or longer follow-up time compared to the case. Follow-up started at the time of initial BC diagnosis and ended at the time of CBC diagnosis for a case and at the time of censoring or death for controls. The final analytic dataset included 665 cases and 1995 controls, with 25 potential predictors (all categorical). Further preprocessing described in the supplement reduced the number of predictors to twenty. BRCA1/2 status or family history beyond first degree relatives is not available in the BCSC dataset. Additionally, while anti-estrogen therapy use is recorded, the specific type administered is unknown.

SEER 17 data:

We used the Incidence – SEER Research Plus Data (November 2023 submission, version 17), which provides population-level cancer information collected between 2000–2021 from 17 registries across the US27. As the month of diagnosis is not available in SEER, we classified a woman as a CBC case if she was diagnosed with BC in the opposite healthy breast at least one year after her initial BC diagnosis. Upon applying the exclusion criteria described in the online supplement, the final dataset included 445,266 women with BC of whom 15,984 developed CBC.

Statistical Methods

We trained CBCRisk-Mastectomy on unilateral BC patients in BCSC and SEER who had undergone mastectomy as part of their treatment26,27. We refer to this subset of BC patients as “mastectomy patients”. The BCSC dataset has missing values for several important risk factors. Since excluding women with missing data would greatly reduce the sample size for fitting a multivariable regression model, in the training of CBCRisk, these women were included by allowing for an “unknown” category for most risk factors. However, as the unknown category is very heterogeneous, and breast cancer patients typically do know this information, here we imputed the missing data by employing multiple imputations as implemented in the Multiple Imputation by Chained Equations (MICE) package29. By restricting the training data to mastectomy patients and imputing missing data, we aim to provide more accurate CBC risk estimates to help BC patients make informed decisions about CPM. The observed data on predictors and CBC status were used to create ten imputed datasets, on which we fitted univariable and multivariable conditional logistic regression models.

Building the relative risk (RR) model:

To conduct an overall likelihood ratio test combining results from ten regression models with the same predictors fitted to the ten imputed datasets, we used a multiple imputation likelihood ratio test (MI-LRT)30,31 described in Supplement Section S3. For building the RR model, first we fitted univariate conditional logistic regression models for each of the twenty potential predictors, retaining those that had an MI-LRT p-value less than 0.2523,24. These predictors were then considered in constructing a multivariable (full) model. A risk factor was retained if the MI-LRT, comparing the full model with a reduced model (obtained by excluding that risk factor), yielded a p-value less than 0.10. Overall estimates of the log-odds ratios β^s for association were obtained by averaging the corresponding estimates from the ten fitted models. These estimates approximate log-RRs as probability of CBC is small. Overall standard errors (SEs) were obtained by applying Rubin’s rule32. We computed 95% confidence intervals (CIs) for a regression coefficient as β^±(1.96*SE). Estimates of RR and CIs were obtained by exponentiating the corresponding quantities.

Building the absolute risk model:

Following the statistical methods used in the Gail model28, CBCRisk23, and CBCRisk-Black24, we obtained age-specific composite incidence rates from SEER and attributable risks (AR) from the BCSC data. For the AR calculation, the study population was divided into thirteen age groups with 5-year intervals, specifically, < 30, 30–34, 35–39, ..., 85–89. The incidence rates and ARs were then combined to estimate baseline age-specific hazard rates of CBC. We used SEER data from 2000–2021 to also estimate age-specific competing mortality rates from causes other than CBC. We then combined the RRs, ARs, age specific baseline CBC hazards, and competing mortality hazards as described in Gail model28 and the supplementary material of CBCRisk23 to obtain the absolute risk prediction model. Here, competing mortality events include death due to first BC as it precludes the possibility of CBC and all other deaths due to non-CBC causes. All such events of competing mortality are accounted for in the absolute risk calculation of CBC. A bootstrap method33 with 500 resamples was used to construct CIs for absolute risk estimates (see Supplement Section S3).

Model Evaluation:

To evaluate the discriminatory performance of CBCRisk-Mastectomy, we calculated the area under the receiver operating characteristic curve (AUC) and its 95% CI for 5-year absolute risk predictions obtained using leave-one-group-out cross-validation (CV), as detailed in Supplement Section S3. A “group” refers to one case and its three matched controls. For comparison, we also computed leave-one-group-out CV AUC of CBCRisk. However, the imputed datasets could not be used for this purpose because CBCRisk has “Unknown” category for most of the predictors and hence, data without unknown categories (missing values) cannot be used to fit the model. Instead, we use the original (unimputed) matched case-control data resulting in a single AUC value for CBCRisk.

The statistical software R (4.4.0) was used for all computations, including the MICE (3.18.0)34 and CBCRisk (2.0)23 packages. The package pROC (1.18.5) was used to estimate AUC and obtain its CI35. For conducting MI-LRT, we used the publicly available R code30,36.

Results

An assessment in Supplement Section S4 shows that the missing at random assumption for the data is plausible, justifying the analysis based on multiply imputed data. Table 1 presents the distributions and coding of the twenty potential risk factors among cases and controls prior to imputation. Based on univariable analyses of the twenty risk factors, we retained twelve predictors: histologic type of first BC, lobular carcinoma in situ (LCIS), estrogen receptor (ER) status, breast density (obtained using the Breast Imaging Reporting and Data System (BI-RADS) method), age at first BC diagnosis, family history of BC (only first degree available in BCSC), age at first birth, body mass index (BMI), tumor stage, tumor size, anti-estrogen therapy, and radiation therapy. These twelve variables were used to fit the base model. To assess the significance of each variable in the presence of the other variables, we constructed twelve reduced models by excluding one variable at a time and applied MI-LRT. Tumor stage, tumor size, anti-estrogen therapy, and radiation therapy had p-values greater than 0.10 and hence were removed. The remaining eight variables were retained as important risk factors for predicting CBC at this stage.

Table 1:

Distribution of potential risk factors associated with CBC among women diagnosed with unilateral breast cancer who underwent a mastectomy. The variables related to tumor characteristics and treatment are for the first BC.

# Risk Factors Categories Cases (n = 665) Controls (n = 1995)

Personal Characteristics

1 Age at first birth 40+ 8 (1.2%) 8 (0.4%)
Nulliparous/<40 203 (30.5%) 587 (29.4%)
Unknown 454 (68.3%) 1400 (70.2%)

2 Age at first BC diagnosis <30 7 (1.1%) 7 (0.4%)
30–40 51 (7.7%) 108 (5.4%)
40+ 607 (91.3%) 1880 (94.2%)

3 Bilateral oophorectomy Yes 29 (4.4%) 100 (5.0%)
No 344 (51.7%) 1021 (51.2%)
Unknown 292 (43.9%) 874 (43.8%)

4 Biopsy history No 275 (41.4%) 843 (42.3%)
Yes 133 (20.0%) 343 (17.2%)
Unknown 257 (38.6%) 809 (40.6%)

5 BMI Obese 42 (6.3%) 92 (4.6%)
Overweight 47 (7.1%) 128 (6.4%)
Normal/Underweight 69 (10.4%) 240 (12.0%)
Unknown 507 (76.2%) 1535 (76.9%)

6 Breast density Extremely dense 40 (6.0%) 85 (4.3%)
Heterogeneously dense 160 (24.1%) 386 (19.3%)
Scattered 111 (16.7%) 358 (17.9%)
Almost entirely fat 6 (0.9%) 37 (1.9%)
Unknown 348 (52.3%) 1129 (56.6%)

7 Family history of BC Yes 79 (11.9%) 169 (8.5%)
No 236 (35.5%) 761 (38.1%)
Unknown 350 (52.6%) 1065 (53.4%)

8 Hormonal use No 249 (37.4%) 726 (36.4%)
Yes 73 (11.0%) 228 (11.4%)
Unknown 343 (51.6%) 1041 (52.2%)

9 Menopausal status Post 277 (41.7%) 819 (41.1%)
Peri 6 (0.9%) 10 (0.5%)
Pre 69 (10.4%) 234 (11.7%)
Unknown 313 (47.1%) 932 (46.7%)

Tumor characteristics of first BC

10 ER status Negative 142 (21.4%) 338 (16.9%)
Positive 327 (49.2%) 1075 (53.9%)
Unknown 196 (29.5%) 582 (29.2%)

11 PR status Positive 297 (44.7%) 928 (46.5%)
Negative 173 (26.0%) 477 (23.9%)
Unknown 195 (29.3%) 590 (29.6%)

12 HER2 status Positive 21 (3.2%) 58 (2.9%)
Negative 68 (10.2%) 210 (10.5%)
Unknown 576 (86.6%) 1727 (86.6%)

13 First BC type DCIS 65 (9.8%) 132 (6.6%)
Invasive 600 (90.2%) 1863 (93.4%)

14 High grade No 251 (37.7%) 714 (35.8%)
Yes 298 (44.8%) 947 (47.5%)
Unknown 116 (17.4%) 334 (16.7%)

15 LCIS Yes 16 (2.4%) 18 (0.9%)
No 639 (96.1%) 1966 (98.5%)
Unknown 10 (1.5%) 11 (0.6%)

16 Tumor stage Advanced primary/Metastatic 111 (16.7%) 232 (11.6%)
Early 489 (73.5%) 1563 (78.3%)
Unknown 65 (9.8%) 200 (10.0%)

17 Tumor size T1 250 (37.6%) 825 (41.4%)
T0 5 (0.8%) 27 (1.4%)
T2 174 (26.2%) 580 (29.1%)
T3 56 (8.4%) 118 (5.9%)
T4 16 (2.4%) 33 (1.7%)
TIS 121 (18.2%) 294 (14.7%)
Unknown 43 (6.5%) 118 (5.9%)

Treatment of first BC

18 Anti-estrogen therapy No 474 (71.3%) 1351 (67.7%)
Yes 156 (23.5%) 550 (27.6%)
Unknown 35 (5.3%) 94 (4.7%)

19 Chemotherapy No 390 (58.6%) 1213 (60.8%)
Yes 248 (37.3%) 703 (35.2%)
Unknown 27 (4.1%) 79 (4.0%)

20 Radiation therapy No 521 (78.3%) 1655 (83.0%)
Yes 131 (19.7%) 309 (15.5%)
Unknown 13 (2.0%) 31 (1.6%)

Before finalizing the model, we explored other reduced models obtained by systematically excluding different subsets of the four non-significant variables and performing MI-LRTs to evaluate the significance of the excluded set of variables. This analysis led to the inclusion of tumor stage in the final model. Additionally, we examined interaction effects between pairs of risk factors for which sufficient data were available for each combination level. However, none of the interactions were significant.

Table 2 shows the odds ratios and 95% CIs for the nine risk factors in the final RR model. Having a DCIS diagnosis for the first BC type and the presence of LCIS were associated with a higher risk of developing CBC. CBC risk was also elevated in BC patients with negative ER status, family history of BC, higher BMI, denser breasts, younger age at first BC diagnosis, and older age at first birth. Furthermore, women whose first BC was diagnosed at an advanced/metastatic stage had an increased risk of developing CBC. We note that the confidence interval for the less than 30 years category of age at first diagnosis is very wide due to the small number of cases (n = 7) in this group. Also, for model stability, we had to merge two consecutive categories for age at first birth, BMI, and stage as one of the categories had low sample size. Supplement Table S1 shows results without the merging, which are not substantially different.

Table 2:

Risk factors associated with CBC risk in the final relative risk model among women diagnosed with a unilateral breast cancer who underwent a mastectomy.

# Risk Factors Categories Odds Ratio* 95% CI*

1 Age at first birth 40+ 2.07 1.19, 3.57
Nulli/<40 1

2 Age at first diagnosis <30 3.07 0.94, 10.00
30–40 1.39 0.95, 2.04
40+ 1 -

3 BMI Obese 2.05 1.50, 2.79
Overweight 1.51 1.30, 2.20
Normal/Underweight 1

4 Breast density Extremely dense 3.22 1.44, 7.24
Heterogeneously dense 2.97 1.35, 6.54
Scattered 1.96 0.86, 4.47
Almost entirely fat 1

5 Family history of BC Yes 1.53 1.11, 2.10
No 1

6 ER status Negative 1.35 1.03, 1.78
Positive 1

7 First BC type DCIS 1.39 1.12, 1.74
Invasive 1

8 LCIS Yes 2.51 1.22, 5.18
No 1

9 Tumor stage Advanced primary/Metastatic 1.46 1.11, 1.91
Early BC 1
*

Odds ratios and 95% CIs are pooled estimates obtained using ten imputed datasets.

Among the nine risk factors in the final model, six risk factors, namely, histologic type of first BC, ER, breast density, age at first diagnosis, family history of BC, and age at first birth are present in CBCRisk. The risk factors absent in CBCRisk are BMI, tumor stage, and LCIS. However, instead of LCIS, CBCRisk includes high-risk pre-neoplasia, which combined LCIS and atypical hyperplasia. First BC type in CBCRisk has three categories, while CBCRisk-Mastectomy combines pure DCIS and invasive DCIS into a single category, DCIS, reducing the total to two categories. Additionally, anti-estrogen therapy use is included in CBCRisk, but not in CBCRisk-Mastectomy.

Figure 1 presents the SEER age-specific composite CBC incidence and competing mortality rates (per 1,000 person-years). The CBC incidence rates ranged from 2.58 to 4.89, with the lowest rate observed in the [18, 30) age group and the highest in the [75, 80) age group. The rate generally increased up to age 80 and then declined. Non-CBC mortality rates ranged from 16.42 to 107.90, with the lowest rate in the [45, 50) age interval and the highest in the [85, 90) interval.

Figure 1:

Figure 1:

Age-specific composite incidence rates of CBC (left) and non-CBC mortality rates (right) estimated using SEER data. Age (in years) on the horizontal axis can be interpreted as age at counseling.

The average AUC (along with pooled 95% CI computed using Rubin’s rule) for 5-year absolute risk predictions from CBCRisk-Mastectomy across the ten imputations was 0.620 (0.586, 0.654). In comparison, for CBCRisk, the AUC (along with 95% CI computed using the standard asymptotic formula) based on the original data was 0.577 (0.547, 0.608). This shows that CBCRisk-Mastectomy has better discrimination ability than CBCRisk for the target population of mastectomy patients. A statistical test for difference of AUCs from the two models is not straightforward due to the difference in datasets (complete versus imputed) and methods for computation (Rubin’s rule versus asymptotic). Further, calibration measures such as the ratio of expected to observed numbers of cases cannot be meaningfully computed for matched case-control data, as the observed proportion of cases is fixed by design, which is 1/4 for our data.

Figure 2 shows the distributions of 5-year absolute risk predictions separately among cases and controls for both models. CBCRisk-Mastectomy better discriminates cases from controls as reflected in a larger separation in their risk distributions. This larger difference leads to the higher AUC of CBCRisk-Mastectomy.

Figure 2:

Figure 2:

Leave-one-group-out 5-year absolute risk predictions using CBCRisk-Mastectomy based on ten imputed datasets and CBCRisk. Here absolute risk values above 15% absolute risk are excluded to ensure a smaller scale and better readability. The full figure without this restriction is given in the online supplement as Figure S3.

In Table 3, we present 5-year absolute risk predictions of CBC, using both CBCRisk-Mastectomy and CBCRisk models, for eight women selected from those with complete risk factor profiles in BCSC data, i.e., those without any missing data on the predictors. These examples cover a broad spectrum of risk factor combinations and current ages. The table clearly illustrates that CBC risk varies substantially based on a woman’s specific risk profile. The first row represents a woman with each risk factor at the most common category, which is also the reference category for all risk factors in CBCRisk-Mastectomy except breast density. In this case, the 5-year CBC risk is 1.06%, indicating that the risk for a typical BC patient is generally low. The other examples further show that CBC risk remains low across different risk factor combinations. Only in the last example the risk exceeds 10% and there is, in fact, only one woman with that risk profile in the subset of women with complete information on all nine risk factors. Additionally, for the last example, the 5-year absolute risk of CBC predicted using CBCRisk-Mastectomy is notably higher than that by CBCRisk. This elevated risk may be attributed to differences in the risk factor profiles incorporated in the models. In particular, while CBCRisk considers anti-estrogen therapy as a risk factor, it does not include BMI or tumor stage, both of which are included in CBCRisk-Mastectomy.

Table 3:

Examples of 5-year CBC risk prediction with 95% confidence interval in parenthesis using CBCRisk-Mastectomy. The highlighted cells represent non-baseline categories.

First BC Type LCIS ER Breast density Age at first BC diagnosis Family history BMI Age at first birth Tumor stage Antie-strogen therapy* Age CBCRisk CBCRisk - Mastectomy
Invasive No Positive Heterogeneously dense 40+ No Normal/Underweight Nulliparous/<40 Early BC Yes 52 1.13 (0.95, 1.36) 1.06 (0.84, 1.30)
Invasive No Positive Almost entirely fat 40+ No Obese Nulliparous/<40 Early BC Yes 46 0.64 (0.38, 1.05) 0.68 (0.28, 1.37)
Invasive No Negative Extremely dense <30 Yes Normal/Underweight Nulliparous/<40 Early BC No 29 4.98 (2.37, 9.11) 2.97 (1.14, 15.05)
DCIS No Positive Scattered 40+ No Obese 40+ Early BC No 52 7.75 (3.02, 19.26) 4.04 (2.13, 7.42)
DCIS No Negative Heterogeneously dense 30–40 No Overweight Nulliparous/<40 Advanced primary/Metastatic No 39 3.53 (2.30, 5.36) 4.72 (2.75, 8.55)
Invasive Yes Negative Scattered 40+ No Obese Nulliparous/<40 Advanced primary/Metastatic No 41 2.25 (1.51, 3.45) 5.94 (2.75, 13.92)
DCIS Yes Positive Heterogeneously dense 40+ No Obese Nulliparous/<40 Early BC No 56 3.29 (2.18, 4.95) 7.67 (3.45, 16.39)
DCIS Yes Negative Heterogeneously dense 40+ Yes Obese Nulliparous/<40 Advanced primary/Metastatic No 52 5.41 (3.52, 8.81) 20.45 (9.52, 42.94)

Discussion

BC patients scheduled to undergo mastectomy typically decide, in consultation with their surgeon, whether to remove their other, unaffected breast as well (i.e., whether to undergo CPM) at the same time as the surgery for the original BC. To assist in making this difficult decision we have developed CBCRisk-Mastectomy, a personalized CBC risk prediction model specifically tailored to such patients. By utilizing data from two nationally representative datasets — BCSC and SEER — our model is applicable to the general US population of unilateral BC female patients (except those with genetic mutations, strong family history, or prior mantle radiation). CBCRisk-Mastectomy incorporates nine predictors: first BC type (DCIS or invasive), presence of LCIS, ER status, breast density, age at first diagnosis, family history of BC, age at first birth, BMI, and tumor stage, all well established CBC risk factors. For instance, DCIS, LCIS, and advanced-stage initial BC are associated with a higher risk of developing CBC37,38. Previous studies have also shown that negative ER status, having a family history of BC, denser breasts, and higher BMI are associated with increased CBC risk39–41, as are younger age at first BC diagnosis and older age at first childbirth42.

It is important to note that CBCRisk-Mastectomy is specifically intended for patients who are scheduled to undergo mastectomy and are considering CPM at the same time, to help them in making this decision. For women undergoing lumpectomy, CBCRisk can be used as it predicts the risk of CBC for any women with unilateral BC. It is difficult for us to provide thresholds for identifying women at high risk for CBC. In this regard, we note that Gail and Jatoi (2022) classified women with 10-year CBC risk of 0–2.49% risk as low risk, with 2.5–9.9% risk as medium risk, and with 10% risk or higher as high risk43.

As anti-estrogen therapy is known to lower CBC risk but was not selected as an important predictor during our model building steps, we explored the possibility of adding it back to the final model. However, inclusion of anti-estrogen therapy did not improve model fit (MI-LRT p-value = 0.28), likely because its effect was largely captured by ER status and first BC type. These two variables are highly correlated with anti-estrogen therapy (p-values ≈ 0) and were stronger predictors of CBC risk than anti-estrogen therapy. Thus, they got included in the final CBCRisk-Mastectomy model. For model training, we had also tried MI-LASSO technique, but it retained all the variables, which is a known limitation of the method and thus was not pursued further44. We also tried fitting the model on original data. However, there are only 61 cases with complete information on all nine risk factors of CBCRisk-Mastectomy with five or fewer observations for several categories of multiple risk factors, which prevented model fitting to the complete-case data.

Comparing CBCRisk-Mastectomy with CBCRisk demonstrated improved discrimination of the new model for mastectomy patients, as indicated by the higher AUC. Seven out of the eight risk factors in the CBCRisk model are also involved in the CBCRisk-Mastectomy model, with the only exception being anti-estrogen therapy. CBCRisk-Mastectomy also incorporates BMI and tumor stage. Additionally, while CBCRisk includes an “unknown” category for most predictors 23, CBCRisk-Mastectomy does not have “unknown” predictor categories, as we imputed the missing data before training the model. Even though having an “unknown” category allows a model to be applicable to patients with missing risk factors, it can be detrimental to model training because missing categories are inherently mixtures of other categories and thus highly heterogeneous. By using multiple imputation before training, we reduced this heterogeneity, which as our results show, increased model accuracy.

It is noteworthy that the absolute risk of CBC is generally low for most women, even among those with multiple risk factors present, as shown in Table 3. Moreover, risks of regional and distant recurrences of BC should also be taken into account in CPM decision-making, given that CPM does not reduce these risks. Although CBC is the most common second malignant tumor among women diagnosed with BC, it occurs less frequently than recurrence of the initial BC45. Younger women face a higher lifetime risk of CBC, but this risk decreases with age, while the risks of regional and distant recurrence, as well as death from other causes, increase46. Gail and Jatoi (2022) highlighted the importance of evaluating all these risks when counseling BC patients on the risks and benefits of CPM43. They emphasize that a comprehensive approach that considers personalized risks of CBC, regional recurrence, distant recurrence, and non-CBC mortality should guide the decision-making process regarding CPM15.

Even though the AUC of 0.62 for CBCRisk-Mastectomy reflects moderate discriminatory ability, it is similar to that of other established models that predict breast cancer incidence in high risk women or the general population. For example, CBCRisk reported AUCs of 0.62 and 0.61 in external validations25 and the widely used Gail model for predicting BC risk for the general US female population had a pooled AUC of 0.60 (95% CI: 0.58–0.62)47. Similarly, PredictCBC-2.0A, designed for high-risk BC patients with BRCA1/2 mutation, and PredictCBC-2.0B, designed for the general population of BC patients, had AUCs of 0.66 and 0.59, respectively48. Despite moderate discriminatory performance, these models are currently in use in clinical settings, supporting the potential utility of CBCRisk-Mastectomy for individualized CBC risk assessment and surgical decision-making.

Despite the benefits of the CBCRisk-Mastectomy model, some limitations should be acknowledged. First, the high percentage of missing data for certain variables necessitated using imputations, which induce some degree of uncertainty. A second limitation is the absence of BRCA1/2 mutation information in the BCSC data and hence in the model. Even though younger age at first BC diagnosis combined with a family history of BC may partly capture the effect of germline pathogenic variants, adding genetic information directly in the model may improve model accuracy as reflected in the fact that PredictCBC 2.0A has much higher AUC than that of PredictCBC 2.0B (as mentioned above). An important future work will be to update the model with BRCA1/2 status once data with this information become available.

Another limitation is that the majority of our study population consisted of White non-Hispanic US women, which may limit the generalizability of the model to other races and ethnicities. As we matched CBC cases and controls on race, race cannot be a predictor in the model. Earlier we had developed CBCRisk-Black (based on 186 cases from BCSC), which can be used for Black BC patients24. In mastectomy patients, only 82 cases are Black; such small sample sizes make it challenging to building race-specific models. As the month of diagnosis is not available in the SEER data, we defined CBC as a malignancy occurring in the contralateral breast at least one year after the initial BC diagnosis, while for BCSC, we used a 6 months duration after the initial breast cancer. Using 12 months duration would have led to a loss of 79 cases in BCSC. Another important limitation is that due to the matched case-control design of our study, we could not assess calibration, i.e., bias of the model predictions, which is typically done by comparing the expected number of cases produced by the model with the observed number of cases in a prosepctive cohort. Finally, the BCSC data used in this study were collected from women diagnosed between 1995 to 2009. Therefore, validating the model, including assessing its calibration on newer external data is warranted. CBCRisk-Mastectomy will be ready for clinical implementation after an external validation of the model.

In conclusion, CBCRisk-Mastectomy is a potentially valuable tool to assist clinicians and patients in making informed decisions regarding CPM and tailoring treatment and preventive choices by utilizing personalized risk estimates. Given the physical, psychological, and financial burdens associated with CPM, providing an evidence-based tool for personalized risk prediction is crucial for optimizing care and minimizing unnecessary CPM. Future research should focus on validating this model on external or more recent data and exploring its integration into clinical practice.

Supplementary Material

1

Acknowledgements:

Data collection was supported in part by the National Cancer Institute-funded Breast Cancer Surveillance Consortium (P01CA154292). The collection of BCSC cancer and vital status data was supported in part by several state public health departments and cancer registries (https://www.bcsc-research.org/work/acknowledgement). All statements in this report, including its findings and conclusions, are solely those of the authors and do not necessarily represent the views of the National Cancer Institute or National Institutes of Health. We thank the participating women, mammography facilities, and radiologists for the data they have provided. You can learn more about the BCSC at: http://www.bcsc-research.org/. The authors are thankful to Dr. Debu Tripathy for insightful discussion on this topic. The authors also thank two reviewers for their thorough and thoughtful review of the manuscript. Addressing their comments has led to an improved article.

Funding:

This project was partially funded by the National Cancer Institute at the National Institutes of Health (Grant Number R21CA186086–01).

Footnotes

Software: CBCRisk-Mastectomy is implemented in R package CBCRisk version 3.0 available at personal.utdallas.edu/~swati.biswas/ and github.com/ihsajal/CBCRisk.

Conflict of Interest: The authors report no proprietary or commercial interest in any product mentioned or concept discussed in this article.

Publisher's Disclaimer: This is a PDF of an article that has undergone enhancements after acceptance, such as the addition of a cover page and metadata, and formatting for readability. This version will undergo additional copyediting, typesetting and review before it is published in its final form. As such, this version is no longer the Accepted Manuscript, but it is not yet the definitive Version of Record; we are providing this early version to give early visibility of the article. Please note that Elsevier’s sharing policy for the Published Journal Article applies to this version, see: https://www.elsevier.com/about/policies-and-standards/sharing#4-published-journal-article. Please also note that, during the production process, errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Data availability:

Instructions on how to request BCSC data can be found at https://www.bcsc-research.org/. SEER data can be requested at https://seer.cancer.gov/.

References:

  • 1.Kummerow KL, Du L, Penson DF, Shyr Y, Hooks MA. Nationwide trends in mastectomy for early-stage breast cancer. JAMA Surgery. 2015;150(1):9–16. [DOI] [PubMed] [Google Scholar]
  • 2.Nash R, Goodman M, Lin CC, et al. State variation in the receipt of a contralateral prophylactic mastectomy among women who received a diagnosis of invasive unilateral early-stage breast cancer in the United States, 2004–2012. JAMA Surgery. 2017;152(7):648–657. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 3.Shaheen MS, Momeni A. Nationwide trends in contralateral prophylactic mastectomies: An analysis of 55,060 unilateral breast cancer patients. Plastic Reconstructive Surgery–Global Open. 2022;10(5):e4344. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 4.Clegg DJ, Whiteaker EN, Salomon BJ, et al. Contralateral prophylactic mastectomy in a rural population: A single-institution experience. J Surgery Open Science. 2024;18:70–77. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 5.Nichols HB, de González AB, Lacey JV Jr, Rosenberg PS, Anderson WF. Declining incidence of contralateral breast cancer in the United States from 1975 to 2006. Journal of Clinical Oncology. 2011;29(12):1564–1569. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 6.Spronk I, Schellevis FG, Burgers JS, de Bock GH, Korevaar JC. Incidence of isolated local breast cancer recurrence and contralateral breast cancer: a systematic review. The Breast. 2018;39:70–79. [DOI] [PubMed] [Google Scholar]
  • 7.Ramin C, Withrow DR, Davis Lynn BC, Gierach GL, Berrington de González A. Risk of contralateral breast cancer according to first breast cancer characteristics among women in the USA, 1992–2016. Breast Cancer Research. 2021;23:1–10. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 8.Davies KR, Cantor SB, Brewster AM. Better contralateral breast cancer risk estimation and alternative options to contralateral prophylactic mastectomy. International Journal of Women's Health. 2015:181–187. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 9.Portschy PR, Kuntz KM, Tuttle TM. Survival outcomes after contralateral prophylactic mastectomy: A decision analysis. Journal of the National Cancer Institute. 2014;106(8):dju160. [DOI] [PubMed] [Google Scholar]
  • 10.Wong SM, Freedman RA, Sagara Y, Aydogan F, Barry WT, Golshan M. Growing use of contralateral prophylactic mastectomy despite no improvement in long-term survival for invasive breast cancer. Annals of Surgery. 2017;265(3):581–589. [DOI] [PubMed] [Google Scholar]
  • 11.Jatoi I, Parsons HM. Contralateral prophylactic mastectomy and its association with reduced mortality: Evidence for selection bias. Breast Cancer Research Treatment. 2014;148:389–396. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 12.Boughey JC, Attai DJ, Chen SL, et al. Contralateral prophylactic mastectomy (CPM) consensus statement from the American Society of Breast Surgeons: Data on CPM outcomes and risks. Annals of Surgical Oncology. 2016;23:3100–3105. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 13.Hawley ST, Jagsi R, Morrow M, et al. Social and clinical determinants of contralateral prophylactic mastectomy. JAMA Surgery. 2014;149(6):582–589. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 14.Kurian AW, Lichtensztajn DY, Keegan TH, Nelson DO, Clarke CA, Gomez SL. Use of and mortality after bilateral mastectomy compared with other surgical treatments for breast cancer in California, 1998–2011. Journal of American Medical Association. 2014;312(9):902–914. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 15.Schmidt MK, Kelly JE, Brédart A, et al. EBCC-13 manifesto: Balancing pros and cons for contralateral prophylactic mastectomy. European Journal of Cancer. 2023;181:79–91. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 16.Boughey JC, Schilz SR, Van Houten HK, Zhu L, Habermann EB, Lemaine V. Contralateral prophylactic mastectomy with immediate breast reconstruction increases healthcare utilization and cost. Annals of Surgical Oncology. 2017;24:2957–2964. [DOI] [PubMed] [Google Scholar]
  • 17.Miller ME, Czechura T, Martz B, et al. Operative risks associated with contralateral prophylactic mastectomy: a single institution experience. Annals of Surgical Oncology. 2013;20:4113–4120. [DOI] [PubMed] [Google Scholar]
  • 18.Huang J, Chagpar A. Complications in patients with unilateral breast cancer who undergo contralateral prophylactic mastectomy versus unilateral mastectomy. Surgery. 2018;164(6):1347–1350. [DOI] [PubMed] [Google Scholar]
  • 19.Momoh AO, Cohen WA, Kidwell KM, et al. Tradeoffs associated with contralateral prophylactic mastectomy in women choosing breast reconstruction: Results of a prospective multicenter cohort. Annals of Surgery. 2017;266(1):158–164. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 20.Hegde JV, Wang X, Attai DJ, et al. Assessing the effect of lifetime contralateral breast cancer risk on the selection of contralateral prophylactic mastectomy for unilateral breast cancer. Clinical Breast Cancer. 2018;18(2):e205–e218. [DOI] [PubMed] [Google Scholar]
  • 21.Kenny R, Reed M, Subramanian A. Mastectomy for risk reduction or symmetry in women without high risk gene mutation: A review. International Journal of Surgery. 2018;50:60–64. [DOI] [PubMed] [Google Scholar]
  • 22.Jansen J, Serafimovska A, Glassey R, et al. The implementation of a decision aid for women with early-stage breast cancer considering contralateral prophylactic mastectomy: A pilot study. Patient Education Counseling. 2022;105(1):74–80. [DOI] [PubMed] [Google Scholar]
  • 23.Chowdhury M, Euhus D, Onega T, Biswas S, Choudhary PK. A model for individualized risk prediction of contralateral breast cancer. Breast Cancer Research Treatment. 2017;161:153–160. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 24.Sajal IH, Chowdhury M, Wang T, Euhus D, Choudhary PK, Biswas S. CBCRisk-Black: A personalized contralateral breast cancer risk prediction model for black women. Breast Cancer Research Treatment. 2022;194(1):179–186. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 25.Chowdhury M, Euhus D, Arun B, Umbricht C, Biswas S, Choudhary P. Validation of a personalized risk prediction model for contralateral breast cancer. Breast Cancer Research Treatment. 2018;170(2):415–423. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 26.Breast Cancer Surveillence Consortium. Accessed 27th July, 2024, https://www.bcsc-research.org/
  • 27.Surveillence, Epidemiology, and End-Result Program. Accessed 27th July, 2024, https://seer.cancer.gov/
  • 28.Gail MH, Brinton LA, Byar DP, et al. Projecting individualized probabilities of developing breast cancer for white females who are being examined annually. Journal of the National Cancer Institute. 1989;81(24):1879–1886. [DOI] [PubMed] [Google Scholar]
  • 29.White IR, Royston P, Wood AM. Multiple imputation using chained equations: issues and guidance for practice. Statistics in Medicine. 2011;30(4):377–399. [DOI] [PubMed] [Google Scholar]
  • 30.Chan KW, Meng X-L. Multiple improvements of multiple imputation likelihood ratio tests. Statistica Sinica. 2022;32(3):1489–1514. [Google Scholar]
  • 31.Meng X-L, Rubin DB. Performing likelihood ratio tests with multiply-imputed data sets. Biometrika. 1992;79(1):103–111. [Google Scholar]
  • 32.Van Buuren S Flexible Imputation of Missing Data. CRC press; 2018. [Google Scholar]
  • 33.Schomaker M, Heumann C. Bootstrap inference when using multiple imputation. Statistics in Medicine. 2018;37(14):2252–2266. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 34.Van Buuren S, Groothuis-Oudshoorn K. mice: Multivariate imputation by chained equations in R. Journal of statistical software. 2011;45:1–67. [Google Scholar]
  • 35.Package ‘pROC’. 2021. [Google Scholar]
  • 36.Chan KW. Multiple Improvements of Multiple Imputation Likelihood Ratio Tests. Updated 9/31/2021. Accessed 9/21/2024, 2024. https://sites.google.com/site/kwchankeith/publications/milrt?authuser=0 [Google Scholar]
  • 37.Miller ME, Muhsen S, Zabor EC, et al. Risk of contralateral breast cancer in women with ductal carcinoma in situ associated with synchronous ipsilateral lobular carcinoma in situ. Annals of Surgical Oncology. 2019;26:4317–4325. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 38.Vichapat V, Garmo H, Holmqvist M, et al. Tumor stage affects risk and prognosis of contralateral breast cancer: Results from a large Swedish-population–based study. Journal of Clinical Oncology. 2012;30(28):3478–3485. [DOI] [PubMed] [Google Scholar]
  • 39.Akdeniz D, Schmidt MK, Seynaeve CM, et al. Risk factors for metachronous contralateral breast cancer: a systematic review and meta-analysis. The Breast. 2019;44:1–14. [DOI] [PubMed] [Google Scholar]
  • 40.Lizarraga IM, Sugg SL, Weigel RJ, Scott-Conner CE. Review of risk factors for the development of contralateral breast cancer. The American Journal of Surgery. 2013;206(5):704708. [DOI] [PubMed] [Google Scholar]
  • 41.Raghavendra A, Sinha AK, Le‐Petross HT, et al. Mammographic breast density is associated with the development of contralateral breast cancer. Cancer. 2017;123(11):19351940. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 42.Vaittinen P, Hemminki K. Risk factors and age‐incidence relationships for contralateral breast cancer. International Journal of Cancer. 2000;88(6):998–1002. [DOI] [PubMed] [Google Scholar]
  • 43.Gail MH, Jatoi I. Tools for contralateral prophylactic mastectomy decision making. Journal of Clinical Oncology. 2022;40(31):3653–3659. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 44.Chen Q, Wang S. Variable selection for multiply‐imputed data with application to dioxin exposure study. 2013. [DOI] [PubMed] [Google Scholar]
  • 45.Khan SA, Kocherginsky M. Contralateral breast cancer remains a complex biologic conundrum. JAMA Oncology. 2024; [DOI] [PubMed] [Google Scholar]
  • 46.de Boer AZ, van der Hulst HC, de Glas NA, et al. Impact of older age and comorbidity on locoregional and distant breast cancer recurrence: A large population‐based study. The Oncologist. 2020;25(1):e24–e30. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 47.Wang X, Huang Y, Li L, Dai H, Song F, Chen K. Assessment of performance of the Gail model for predicting breast cancer risk: a systematic review and meta-analysis with trial sequential analysis. Breast Cancer Research. 2018;20:1–19. [DOI] [PMC free article] [PubMed] [Google Scholar]
  • 48.Giardiello D, Hooning MJ, Hauptmann M, et al. PredictCBC-2.0: a contralateral breast cancer risk prediction model developed and validated in~ 200,000 patients. Breast Cancer Research. 2022;24(1):69. [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1

Data Availability Statement

Instructions on how to request BCSC data can be found at https://www.bcsc-research.org/. SEER data can be requested at https://seer.cancer.gov/.

RESOURCES