Comparative effectiveness study of breast-conserving surgery and mastectomy in the general population: A NCDB analysis.

PURPOSE
Recent studies have revealed that breast-conserving surgery (BCS) with radiotherapy (RT) led to better survival than mastectomy in some populations. We compared the efficacy of BCS+RT and mastectomy using the National Cancer Database (NCDB, USA).


RESULTS
A total of 160,880 patients with a median follow-up of 43.4 months were included. The respective 8-year OS values were 86.5%, 72.3% and 70.4% in the BCS+RT, mastectomy alone and mastectomy+RT group, respectively (P < 0.001). After exclusion of patients with comorbidities, mastectomy (alone or with RT) remained associated with a lower OS in N0 and N1 patients. However, the OS of mastectomy+RT was equivalent to BCS+RT in N2-3 patients. Among patients aged 50 or younger, the OS benefit of BCS+RT over mastectomy alone was statistically significant (HR1.42, 95% CI 1.16-1.74), but not clinically significant (<5%) in N0 patients, whereas in N2-3 patients, the OS of BCS+RT was equivalent to mastectomy+RT (85.2% vs. 84.8%). The results of the propensity analysis were similar.


METHODS
Non-metastatic breast cancers in the NCDB from 2004-2011 were identified.The Kaplan-Meier method, Coxregression and propensity score analysis were used to compare the overall survival (OS) among patients with BCS+RT, mastectomy alone and mastectomy+RT.


CONCLUSIONS
BCS+RT resulted in improved OS compared with mastectomy ± RT in N0 and N1 patients. In N2-3 patients, BCS+RT has an OS similar to mastectomy+RT when patients with comorbidities were excluded. Among patients aged 50 or younger, the OS of BCS+RT is equivalent to mastectomy ± RT.


INTRODUCTION
The long-term survival of early-stage breast cancer patients is equivalent to either breast-conserving surgery (BCS) plus radiation therapy (RT) or modified radical mastectomy, as demonstrated in several prospective randomized controlled trials (RCTs) [1][2][3][4][5][6]. However, participants in RCT sare highly selected and may not represent the general population. Although RCT scan provide the least biased estimates for treatment comparisons, their results may not correspond to actual clinical situations [7]. In daily routine practice, physicians make decisions based on many uncontrolled factors and apply the results of RCTs to a broader range of patients. Therefore, observational studies are relied on to provide additional information regarding the comparative effectiveness of different treatments in the general population [7]. Abdulkarim, et al reported that in www.impactjournals.com/oncotarget T1-2N0M0 triple-negative (TN) patients, modified radical mastectomy without RT significantly increased the risk of local failure compared with BCS+RT [8]. This interesting finding was also observed in Adkins' study [9] but not in that by Zumsteg [10]. In addition, studies using data from national cancer registries [11][12][13] reported similar findings: that BCS+RT was associated with improved survival compared with mastectomy alone or mastectomy with RT.
In general, the findings from observational studies suffer from selection bias. For example, it is possible that patients who receive BCS+RT are more likely to have fewer comorbidities, which contributes to its superior survival rates. Information about comorbidities was usually unavailable in previous studies [11,13].Here, we retrospectively compared the long-term overall survival (OS) between BCS and mastectomy using the National Cancer Database (NCDB). The NCDB is a joint program of the Commission on Cancer of the American College of Surgeons and the American Cancer Society [7]. It includes more than 1,500 commission-accredited cancer programs in the United States and contains detailed tumor pathology information. Additionally, the insurance type, comorbidities (Charlson-Deyo score) and days of inpatient stay were collected, which enable us to identify patients with low/high comorbidities. We hypothesized that by using a large national cohort of breast cancer patients, this analysis would provide additional evidence in the relative effectiveness of the association between local therapy (BCS vs. mastectomy) and clinical outcomes. To minimize the influence of "confounding by indications", we planned to perform subgroup analyses by comorbidities or age. We hypothesized that in patients with no/few comorbidities, or younger age, there would be no differences of OS between BCS and mastectomy.

RESULTS
We identified 160,880 patients who fit the inclusion and exclusion criteria. The median age was 60 years old, and 59.1%, 34.1% and 6.0% of the patients had AJCC Stage I, II and III disease, respectively. A total of 126,569(78.7%), 26,130(16.2%) and 8,181(5.1%) patients had received BCS+RT, mastectomy alone and mastectomy+RT, respectively. As shown in Table 1, patients in the BCS group were more likely to have private insurance, fewer comorbid diseases (CD score = 0), lower tumor burden (smaller tumor and/or negative nodes), and fewer hormone receptor-positive diseases. Additionally, BCS patients were wealthier and had a higher education level, and the majority of women (83.9%) received surgery at clinics without inpatient stays after surgery.

Survival benefit of BCS over mastectomy in the entire study cohort
With a median follow-up of 43.4 months, the respective 5-year and 8-year OS values were 93.2% and 86.5% in the BCS+RT group, 83.5% and 72.3% in the mastectomy-alone group, and 83.0% and 70.4% in the mastectomy+RT group, respectively (log-rank test; P < 0.001). Univariate (Supplementary Table 1) and  multivariate analyses (Supplementary Table 2A) revealed that mastectomy (alone or with RT) was significantly associated with a lower 5-year and 8-year overall survival in patients with N0, N1 and N2-3 disease compared with BCS+RT ( Figure 1A-1C, Table 2).

Survival benefit of BCS over mastectomy in patients with fewer/no comorbidities
A total of 124,642 patients with a CD score of 0 and days of inpatient stay ≤1 were considered to have fewer/no comorbid diseases and were included in this analysis (Table 2, Figure 2). In N0 patients, BCS+RT (vs. mastectomy alone) increased the 5-year and 8-year OS by 5.0% and 7.8%, respectively. After adjustment, mastectomy alone (vs. BCS+RT) was significantly associated with a lower OS (HR 1.38, 95% CI 1.28-1.49) (Supplementary Table 2B). In N1 patients, mastectomy (alone or with RT) was significantly correlated with a lower OS (Table 2, Figure 2B). However, among N2-3 patients, mastectomy+RT (vs. BCS+RT) was no longer associated with a lower OS after adjustment (HR 1.12, 95% CI 0.97-1.28). The 5-year OS benefit of BCS+RT over mastectomy+RT was less than 5% ( Figure 2A).

Survival benefit of BCS over mastectomy varied by age
Among patients aged 50 or younger, the 5-year and 8-year survival benefit of BCS+RT (vs. mastectomy alone or with RT) was significantly lower than that of patients with age ≥50yrs (Figure 2, 3 Supplementary Table 2C). BCS+RT over mastectomy alone was statistically significant (HR 1.42, 95% CI 1.16-1.74) but not clinically significant (5-year: 2.9%; 8-year: 4.8%) in N0 patients. In N1 and N2-3 patients, BCS+RT over mastectomy+RT was not significantly correlated with an improved OS ( Table 2 Supplementary Table 2C). The survival benefit of BCS+RT vs. mastectomy (alone or with RT) in patients older than 50 were similar to the entire population (

Propensity score analysis of OS
We used propensity scores to create strata in which the possibility of having a specific treatment (BCS+RT vs. mastectomy ± RT) was similar for all patients in the same strata, regardless of their actual received treatment. Using a Cox-regression model stratified by propensity score quintile, we found that mastectomy alone (vs. BCS+RT) was associated with a worse OS in N0 (HR 1.75, 95% CI 1.66-1.85) and N1 patients(HR 1.73, 95% CI 1.60-1.88). www.impactjournals.com/oncotarget

Recent studies of the comparative effectiveness of BCS+RT vs. mastectomy
Randomized controlled trials had demonstrated that BCS has an equivalent long-term survival to mastectomy [3,4,6,[14][15][16][17][18] in early-stage breast cancer patients. The comparative effectiveness of BCS+RT vs. mastectomy in non-clinical-trial population has been investigated in administrative, observational database. Hwang, et al [13] identified 112,154 stage I and II breast cancer patients from the large population-based prospective California Cancer Registry. The authors observed that women who underwent BCS+RT had improved breast cancer specific survival(BCSS) and OS compared with those treated by mastectomy, and the survival benefit of BCS+RT was greater among " ≥50yrs & HR+" women. Brooks, et al [19] used the instrumental variable method to study the SEER-Medicare database and reported that higher mastectomy rates were associated with reduced survival. Other studies using data from national cancer registries in Norway [12], Canada [20] and the US [11,21] reported similar findings (Table 3).
Fisher et al [20]. used data from Alberta Cancer Registry (Canada) and showed that the survival benefit of BCS+RT vs. mastectomy was less significant in stage I patients, than in stage II or III patients. The underlying reason may due to the low risk of relapse events in stage I patients, rendering the survival advantages of BCS+RT less likely to be noted. However, there were two limitations of their study: 1) They did not distinguish patients who received PMRT from those who did not. Additionally, they stratified the analysis by cancer stage, rather than N-stage. Thus, in the stage II strata that contained both node-negative and nodepositive patients, the comparison of BCS vs. mastectomy was significantly confounded by the unknown PMRT status. 2) The proportion of patients with non-standard treatment (e.g. PMRT in N0 patients after mastectomy, or no PMRT in N2-3 patients after mastectomy) in their mastectomy group was unknown. Therefore, the accuracy and generalizability of their results were influenced.
In our study, we stratified the survival analysis by N-stage. We compared BCS+RT vs. mastectomy alone, BCS+RT vs. mastectomy alone vs. mastectomy+RT, BCS+RT vs. mastectomy+RT in N0, N1 and N2 patients, respectively. We believe that our design is clearer in purpose and more informative for clinical practices. Our study showed that in N0 patients that PMRT is not recommended, BCS+RT had better OS than mastectomy alone; whereas in N2-3 patients that PMRT is routinely performed, BCS+RT has equivalent OS to mastectomy+RT after adjustment (Table 2, Figure 1). These results suggested the importance of RT in the comparison of BCS and mastectomy using administrative database.

The impact of comorbid conditions and age
Confounding by indication is the major limitation for most retrospective studies. Specifically, patients with fewer comorbid conditions are more likely to receive BCS+RT and these patients are more likely to have better OS than those with many comorbidites. Land. et al [22] determined that patients with more comorbid conditions were more likely to die from breast cancer as well as other causes, using data from the Danish Breast Cancer Cooperative Group Registry. A study by Hwang et al [13] using the California Cancer Registry database showed that patients who received BCS+RT were less likely to die from cardiovascular disease, or chronic lower respiratory diseases. They inferred that mastectomy patients were more likely to have comorbid diseases, which may have influenced the surgical decision. Hence, we performed subgroup analysis by excluding patients with severe comorbidities. In the NCDB, the Charlson-Deyo (CD) score has been used to describe comorbid conditions. We defined patients with CD score = 0 (no comorbid conditions) and days of inpatient stay ≤1 as "low/no comorbidity" patients. In this subgroup of women, the 5-year OS benefit of BCS+RT was 5.0% in N0 patients (vs. mastectomy alone), and 2.9% in N2-3 patients (vs. mastectomy+RT). Compared with the results derived from our entire study population, the survival benefit of BCS+RT over mastectomy (alone or with RT) was decreased, suggesting that the significant survival benefit of BCS+RT in previous studies might be partially due to comorbid conditions. Subgroup analysis in patients with comorbidities was not performed, because the effect of "confounding by indication" may be more significant in these patients.
Survival analysis was also stratified by age. As shown in Table 2 and Figure 3, the 5-year and 8-year OS benefit of BCS+RT was less than 5% in patients aged 50 or younger, regardless of nodal status. The survival advantages of BCS+RT in patients older than 50 were similar to the entire population. Our result is consistent with previous studies. A population-based study by Cao et al [23] found that 965 patients aged 20 to 39 years with breast cancer treated from 1989 to 2003 showed no difference in the 15-year rates of BCSS. Similar results were observed in other studies [24,25]. The findings from these studies, together with ours, reveal the oncological safety of BCS+RT, particularly in young patients.   More extensive surgery may theoretically lead to better or at least equivalent outcomes as less extensive surgery in cancer treatment. Is it possible that the opposite may be true? Studies in animal models have suggested that the surgical trauma of normal tissue promotes the implantation or growth of circulating tumor cells [26][27][28][29]. Thus, with mastectomy, would a larger wound produce more cytokines to activate the distant dormant tumor cells? There is evidence from clinical studies indicating such a possibility. A randomized controlled trial published in Lancet [30] compared the efficacy of laparoscopy-assisted colectomy and open colectomy for the treatment of non-metastatic colon cancer in terms of tumor relapse and CSS. With lesser surgical treatment, laparoscopy-assisted colectomy had a significantly higher CSS. Further studies of these hypothesis and exploration of the underlying mechanisms are needed.

Benefit of PMRT
The benefit of PMRT in patients with N1 disease is controversial. A subgroup analysis of DBCG 82 randomized trials [31] suggested a similar OS benefit of RT in patients with N1 and N2-3 disease. Likewise, an EBCTCG meta-analysis using individual data from 22 trials [32,33] revealed that PMRT reduced the rate of mortality in breast cancer patients, which was independent of the number of lymph nodes involved. The National Comprehensive Cancer Network (NCCN) guidelines [34] also recommend strong consideration of PMRT in N1 patients. However, our study demonstrated that PMRT was associated with an improved 5-year OS in N2-3 patients rather than N0 or N1 patients (Table 2). Our study was not the first to question the need for PMRT in N1 patients. He, et al [35] reported that, in patients with T1-2 and N1 diseases, the survival benefit of PMRT was present only in high-risk patients. Huo, et al [36] analyzed 93,793 and 36,299 T1-2N1 breast cancer patients in the NCDB and SEER database and showed that PMRT improved OS among patients with 3 positive nodes or 2 positive nodes with tumors 2-5 cm in size, but had no benefit in patients with 1 positive node or 2 positive nodes with tumor ≤2 cm. McBride, et al [37] also observed that the benefit of PMRT on local control was highly dependent on the era of treatment: PMRT reduced the risk of 5-year LRR in patients treated in an early era (1978-1998), but not in those treated in a later era (2000)(2001)(2002)(2003)(2004)(2005)(2006)(2007), when the use of sentinel lymph node surgery, taxane chemotherapy, and aromatase inhibitors was routine. Taken together, we may reconsider the benefit of PMRT on OS among N1 breast cancer patients. Further randomized controlled studies addressing this issue are warranted.

Limitations
In this observational study, we were not able to retrieve information such as multifocality, EIC, HER2 status, BRCA status, body-mass index, detailed chemotherapy, endocrine therapy, targeted therapy and cancer-specific survival (CSS) from the NCDB. Thus, the selection biases cannot be completely diminished, although we used propensity score analysis. 1) BRCA mutationpositive patients would tend to receive mastectomy. However, a recent study showed that the 10-year OS is similar between patients with or without the BRCA mutation [38]. In addition, HER2 status is not expected to be associated with the choice of surgery (mastectomy vs. BCS). Therefore, the failure to match for BRCA status or HER2 status would not have a significant impact on our analysis. 2) Patients with positive margins, who were converted to mastectomy from BCS, were likely associated with multifocal DCIS or EIC. Hence, the inability to match for multifocality is a major limitation of this study.
3). Lack of information of BCSS in the NCDB is another limitation of our study. However, a summary of previous studies [8,9,12,20] showed that the survival benefit of BCS+RT (over mastectomy) on BCSS was similar to that of OS (Table 3). Thus, we think that OS is also a valid endpoint for our study. 4) We are not able to ensure that the chemotherapy treatments were completely comparable between groups.

Data collection
We searched the NCDB registry data between 2004 and 2011 and identified female patients with a T1-2N0-3M0 infiltrating duct breast carcinoma diagnosis. The following information was collected: facility type (community cancer program, comprehensive community cancer program, academic/research program, other program); primary payer (not insured, private insurance, Medicaid, Medicare, other government); city type (metropolitan, urban, rural); distance to hospital; median income of the state/area (2008-2012) where the patient lived; the percentage of population without a high school degree of the state/area where the patient lived; age; race; T stage; N stage; AJCC stage; laterality (left or right breast); estrogen receptor (ER) status; progesterone receptor (PR) status; grade; surgery of primary site; radiation; survival (months); vital status; primary site (LIQ, lower-inner quadrant; LOQ, lower-outer quadrant; UIQ, Upper-inner quadrant; UOQ, Upper-outer quadrant; nipple; others); days of inpatient stay; Charlson-Deyo (CD) score; lymphovascular invasion status; chemotherapy(administered as first course treatment); and hormonal therapy (administered as first course treatment). For the breast surgery code,we used the NAACCR item www.impactjournals.com/oncotarget #1290 coding rules (breast, C50.0-C50.9,http://ncdbpuf. facs.org/?q=content/breast). We defined codes 20-24 as BCS and codes 41 and 51 as mastectomy.
The data within the NCDB are rendered anonymous; therefore, the study was exempt from the Johns Hopkins Medicine Institutional Review Board review, and no consent was required.

Inclusion and exclusion criteria
We had rigorous inclusion and exclusion criteria for patient selection. All of the included patients met the following inclusion and exclusion criteria.

Statistical analyses
The chi-square test was used to compare the demographic and clinicopathological features of patients among three groups: the BCS + RT group; the mastectomyalone group; and the mastectomy + RT group. Kaplan-Meier survival analysis and an unadjusted Cox proportional hazards model were used to compare the overall survival (OS) of patients who received the different treatments, as a univariate approach. Significant factors revealed by univariate analysis were incorporated into multivariate analysis (adjusted Cox proportional hazards model) and the model was used through out the entire study. In subgroup analysis, we planned to assess the effect of surgery type on OS in patients stratified by N-stages, comorbidities and ages. The variables that were used for stratification were not included in the multivariate model during subgroup analysis.
Propensity analysis was used to compare treatment groups within strata to minimize selection bias or a lack of covariate balance. We considered that age, facility type, primary payer, primary sites, income, urban type, education, distance to hospital, CD score, race, tumor grade, ER, PR, T stage and N stage were all potential determinants for the choice of surgery. Thus, all of these variables were included as conditioning variables in a logistic model, with surgery type (BCS+RT vs. mastectomy alone or RT) as the outcome variable. The propensity score was then calculated as the probability of receiving a mastectomy (alone or with RT) for each individual. We stratified the patients into quintiles, in which patients had a similar likelihood of having received a given treatment. Using the Cox model, we estimated the effect of different treatments (BCS+RT, mastectomy alone or mastectomy+RT) on OS, with the baseline survival function varied across strata by including quintiles of the propensity scores as stratification variables.
All P values are two sided. P values <0.05 were considered statistically significant. A survival difference larger than 5% was considered clinically significant. Statistical analyses were performed using Stata/MP, version 13.0 (StataCorp LP, College Station, TX, USA).

CONCLUSIONS
This analysis of a large national cohort of patients demonstrates that BCS+RT provides a superior OS to mastectomy (alone or with RT) in N0 and N1 patients, regardless of comorbid conditions. In N2-3 patients, the survival benefit of BCS+RT (vs. mastectomy+RT) was eliminated when patients with comorbid conditions were excluded. Among patients aged 50 or younger, the BCS+RT OS is equivalent to mastectomy (alone or with RT). Because mastectomy is significantly more invasive than BCS+RT, we recommend greater efforts at educating patients to undergo BCS+RT rather than mastectomy in our routine practice, particularly for low-risk N0/N1 women.