Does local recurrence affect the rate of distant metastases and survival in patients with early-stage breast carcinoma treated with breast-conserving therapy?

Authors


Abstract

BACKGROUND

The purpose of the current analysis was to evaluate the impact of local recurrence (LR) on the development of distant metastases (DM), overall survival (OS), and cause specific survival (CSS) in patients with early-stage breast carcinoma who underwent conservative surgery (CS) and received postoperative radiotherapy (RT).

METHODS

Between 1980 and 1995, 1169 patients underwent CS and received RT. All patients were followed for > 1 year and had ≤ 4 lymph nodes involved with disease. The median duration of follow-up was 7.7 years. A Cox proportional hazards model was performed to evaluate the effect of LR on the development of DM and CSS. A matched-pair analysis that controlled for multiple prognostic factors also was performed comparing the outcomes of patients with and without LR.

RESULTS

The LR rate was 11% at 12 years. For the entire population, LR led to poorer OS and CSS rates at 12 years compared with local control (LC) (71% vs. 81% [P = 0.001] and 69% vs. 88% [P < 0.001], respectively). In a Cox multiple regression model, LR was a significant predictor of disease specific mortality. The hazard ratio (HR) associated with LR was 2.69 for mortality and 2.67 for DM (P < 0.001 and P < 0.001, respectively). The median time from surgery to the development of DM was 3.8 years for patients without LR compared with 4.7 years for patients with LR. Patients who developed LR also had two peaks in the rate of DM (at 2.5 years and at 6.5 years) compared to only one peak (at 1.5 years) for patients who did not develop LR. The impact of LR on DM still was evident in patients with small tumors (≤ 2.0 cm; P < 0.001), negative lymph nodes (P = 0.004), or both (P < 0.001). Recurrent disease that developed outside of the surgical bed region had no negative effect on survival. In the matched-pair analysis (controlling for age, tumor size, grade, number of positive lymph nodes, and estrogen receptor status), LR remained the most significant predictor of mortality (HR: mortality, 5.86; DM, 6.43).

CONCLUSIONS

The current results suggest that LR may be responsible for an increase in DM and disease specific mortality in patients who undergo CS and receive RT. This suggestion is reinforced by the distinct difference seen in the time distribution of DM after LR developed and by the fact that recurrent disease that originated outside of the surgical bed did not affect OS. These data reinforce the necessity to insure optimal LC in patients who are treated with breast-conserving therapy. Cancer 2003;97:910–9. © 2003 American Cancer Society.

DOI 10.1002/cncr.11143

The impact of local recurrence (LR) on survival in patients with early-stage breast carcinoma who are treated with breast-conserving therapy (BCT) remains controversial. Although it has been demonstrated consistently that patients who experience an LR after BCT have an increased risk of developing distant metastases (DM), it is uncertain whether an LR signals a more biologically aggressive tumor or is the nidus for future dissemination.1, 2 Many clinicians assume that an LR after BCT has no detrimental effect on survival due to the belief that breast carcinoma is a systemic disease at inception.3 Consequently, an LR is considered a marker for DM rather than a cause. In contrast, others believe that preventing an LR may improve survival by avoiding a secondary dissemination of tumor cells directly from the LR.4 This hypothesis is corroborated by recent data from three large, prospective, randomized trials in which survival was improved in high-risk, premenopausal and postmenopausal patients with breast carcinoma who received adjuvant, postmastectomy, locoregional radiotherapy (RT) and systemic therapy.5–7 It is possible that LR after BCT may have the same deleterious effect on survival that it has after mastectomy. Unfortunately, it may never be possible to prove that LR is responsible for an increase in the rate of DM. No direct experiment could ever allow a patient to be randomized to an LR group or a local control (LC) group.4, 8

Several recent studies have attempted to demonstrate (through unique statistical methodologies) how an LR may be causally related to an increased risk of developing DM. Fortin et al. and Koscielny and Tubiana showed how the time course and rate of development of DM were uniquely different in patients who experienced LR, suggesting that greater tumor aggressiveness alone could not explain these findings.4, 9 Unfortunately, regardless of how analyses are performed, the problems of competing risks and censorship can never be eliminated totally. Consequently, many clinicians are reluctant to accept a causal correlation between LR and DM based on nonrandomized data, no matter how compelling and elegant the analysis.10 The most convincing evidence of such a correlation can come only from randomized trials in which one treatment arm has both significantly fewer patients who develop LR as well as significantly fewer patients who develop DM.7, 8

The purpose of the current analysis was to evaluate further the impact of LR on the development of DM using two additional assumptions. First, a matched-pair analysis (MPA) was performed comparing outcomes in patients who developed an LR with patients who did not develop an LR to reduce competing risks for distant failure. Patients were matched for multiple factors related to the development of DM. Second, only patients who experienced a true recurrence of the index lesion were analyzed separately to eliminate the confounding effects of a new primary lesion developing in the breast (unrelated to diagnosis and treatment of the original lesion). Then, patterns of failure as well as differences in the time distribution of DM were evaluated.

MATERIALS AND METHODS

From 1980 to 1995, 1388 breasts were treated consecutively with BCT at William Beaumont Hospital, Royal Oak, Michigan. Patients with < 1 year of follow-up or > 4 lymph nodes involved with disease were excluded. Therefore, a total of 1200 breasts in 1169 patients were included in this analysis. Only patients with American Joint Committee on Cancer Stage I and II breast carcinoma were included. The following data were compiled when available: age, menopausal status, method of detection, date of detection, clinical and pathologic stage, tumor size, histologic type and grade, surgical margin status, hormone receptor status, number of positive and excised lymph nodes, dose and technique of RT, use of a reexcision, presence of an extensive intraductal component (EIC), and the use and type of systemic therapy.

Treatment

Our surgical and RT techniques have been reported previously.11–13 Briefly, conservative surgery (CS) usually consisted of a macroscopic total resection of the primary tumor with a variable rim of uninvolved breast tissue. Reexcision was performed as necessary at the discretion of the surgeon or radiation oncologist. After surgery, the whole breast was irradiated with 2 tangential fields to a median total dose of 45 grays (Gy) in 1.8-Gy or 2.0-Gy fractions. This was followed by a supplemental boost to ≥ 60 Gy in 1156 patients (96%). The median total dose was 61.0 Gy. The majority of patients (92%) underwent axillary lymph node dissection (AXLD) usually confined to levels I–II of the axilla. A total of 127 patients (11%) also received separate RT to the supraclavicular fossa/axilla to a median total dose of 50 Gy (generally in patients with positive axillary lymph nodes or in patients who did not undergo axillary surgery). A total of 215 patients (18%) received adjuvant systemic chemotherapy. Chemotherapy varied during the years of study but generally included cyclophosphamide, methotrexate, and 5-fluorouracil or doxorubicin-based regimens. Four hundred sixteen patients (36%) received tamoxifen. Patients were followed every 3 months by their radiation oncologist or surgeon during the first 2 years and every 6 months thereafter. Baseline mammography was performed at 6 months after the completion of RT and yearly thereafter.

Statistical Analysis: All Patients

An ipsilateral LR was defined as the reappearance of disease in the treated breast before or at the time of metastases. To evaluate the effects of different types of LR on the development of DM and on survival, recurrences were classified by their location within the breast according to the criteria established by Recht et al.14 A true recurrence/marginal miss (TR/MM) failure was defined as a recurrence of the index lesion within the region of the lumpectomy cavity or the boost field (if performed). An elsewhere failure (E failure) was defined as a recurrence developing outside of the lumpectomy bed or boost field and was felt to be unrelated to the treatment of the index lesion (e.g., a de novo neoplasm). Unclear or borderline lesions always were classified as TR/MM. Contralateral breast failure was defined as the subsequent development of breast carcinoma in the opposite, untreated breast. Overall survival (OS) reflects all deaths, disease-related or otherwise. Cause specific survival (CSS) was based on death attributed to breast carcinoma.

Estimations of likelihood events for local failure, regional failure, distant failure, CSS and OS were calculated according to the Kaplan–Meier method for the entire patient population. Statistical differences between curves were calculated using the log-rank test. The association of categoric variables with each group was analyzed using the Fisher exact test (two-tailed). For continuous variables, the Student unpaired t test was used to determine the significance of the difference between two sample means. Multiple regression analysis was performed using the Cox proportional hazards model. A P value of ≤ 0.05 was considered statistically significant. All time intervals were calculated from the date of diagnosis. Statistical analysis was performed with SYSTAT software (version 10.0; SPSS, Inc., Chicago, IL).

Statistical Analysis: MPA

To reduce the effects of confounding variables that contribute to DM and CSS, an MPA was performed comparing outcome in patients with and without the development of an LR. (i.e., patients were matched to eliminate the possibility of unequal distributions of critical variables that independently can affect DM and CSS). Only 51 of 71 recurrences that were classified as a TR/MM failures were used in the MPA. Each of these 51 patients was matched randomly with two of the remaining 1095 patients who had not experienced an LR (1:2 match). Outcome (DM and CSS) was blinded for both groups of patients. The RANDOMIZE statement was used with the RND function within Microsoft Visual Basic software (Microsoft Corporation, Redmond, WA) to generate random numbers between 1 and 1095. This was performed to eliminate any possibility of selection bias in the matched group. Each patient with a TR/MM was matched with two unique patients without LR. Patients were matched for 1) tumor size ± 5 mm, 2) age ± 10 years, 3) the same number of positive lymph nodes (one vs. two vs. three vs. four vs. no AXLD), 4) the same tumor grade (Grade 1 vs. Grade 2 vs. Grade 3 vs. unknown), and 5) estrogen receptor status (positive vs. negative vs. unknown).

RESULTS

All Patients

The mean duration of follow-up for all patients was 7.7 years (median, 6.7 years; range, 1.1–17.8 years). A total of 74 patients (6%) developed an LR in the ipsilateral breast, translating into 6-year and 12-year LR rates of 4% and 11%, respectively. Complete actuarial failure data are summarized in Table 1. Mean values for selected characteristics in patients with and without LR are presented in Table 2. The only differences between the two groups were younger age (mean of 51.9 years vs. 58.6 years; P < 0.001) and a longer mean follow-up (9.9 years vs. 7.5 years; P < 0.001) for patients who developed an LR.

Table 1. Actuarial Failure Data for All Patients (N = 1169 patients)
EndpointActuarial failure (%)
At 6 yrsAt 12 yrs
  1. TR/MM: true recurrence/marginal miss.

Local recurrence4.011.0
 TR/MM3.07.0
 Elsewhere0.43.0
Regional failure3.04.0
Distant metastases10.015.0
Disease free survival85.069.0
Overall survival92.080.0
Cause specific survival93.086.0
Table 2. Mean Values for Selected Patient Characteristics (All patients)
CharacteristicNo LRLRP valuea
  • LR: local recurrence; RT: radiation therapy; Gy: grays.

  • a

    Unpaired t test.

Age at diagnosis (yrs)58.651.9< 0.001
Tumor size (cm)1.61.90.10
Grade2.02.00.76
No. of lymph nodes sampled13.513.60.83
No. of positive lymph nodes0.360.390.77
Total RT dose (Gy)61.161.10.99
Follow-up (yrs)7.59.9< 0.001
Time to LR (yrs)5.7
Time to regional failure (yrs)3.05.30.07
Time to distant metastasis (yrs)3.84.70.23

Survival Analysis

A total of 141 patients died (12%), of which 110 deaths (9%) were related to breast carcinoma. The actuarial CSS rate was 86% at 12 years. Table 3 lists failure results for patients with and without an LR. At 12 years, the CSS rate was 88% in patients without LR and 69% in patients with an LR (P < 0.001). Table 4 lists factors that were associated with CSS in the Cox multiple regression analysis. Local failure was the most powerful predictor of mortality with a hazard ratio (HR) of 2.69. Additional factors that were associated with mortality included positive lymph nodes (HR, 2.48), grade (HR, 2.68), tumor size (HR, 1.36), and estrogen receptor status (HR, 1.98).

Table 3. Twelve-Year Data Based on Failure Status (All patients)
Endpoint (at 12 yrs)No LR (%)LR (%)Log-rank P valueOR (95%CI)OR (P value)
  1. LR: local recurrence; OR: odds ratio; 95%CI: 95% confidence interval.

Regional failure316< 0.0015.69 (2.56–12.64)< 0.001
Distant metastases1333< 0.0013.86 (2.26–6.12)< 0.001
Overall survival8171< 0.12.54 (1.44–4.46)0.001
Cause specific survival8869< 0.0014.13 (2.37–7.21)< 0.00
Table 4. Multivariate Analysis (All patients)
FactorCause specific survivalDistant metastases
P valueHRP valueHR
  • HR: hazard ratio; ER: estrogen receptor; RT: radiation therapy.

  • a

    Not included in multivariate analysis, because it was not significant on univariate analysis.

Local recurrence< 0.0012.69< 0.0012.67
Positive (vs. negative) lymph nodes< 0.0012.48< 0.0012.15
Higher grade< 0.0012.68< 0.0012.32
Greater size (cm)0.0011.260.011.19
ER negative (vs. positive)0.0041.980.12
Close/positive margin (vs. negative margin)0.12a
Younger age0.360.24
Lower RT dosea0.020.91

Factors Associated with DM

A total of 130 patients (11%) developed DM, for a 12-year actuarial DM rate of 15%. The rate of DM in patients who developed an LR was 33% compared with 13% in patients who did not develop an LR (P < 0.001). Factors that were associated with the development of DM on multiple regression analysis also are listed in Table 4. Local failure was relatively the most significant predictor of DM, with an HR of 2.67. Additional factors that were associated with DM included positive lymph nodes (HR, 2.15), grade (HR, 2.32), and tumor size (HR, 1.19). Local failure still was associated with an increased risk of DM in patients with tumors that measured ≥ 2.0 cm (P < 0.001), patients with negative lymph nodes (P = 0.004), or both (22% vs. 8% at 10 years, respectively; P < 0.001).

Table 5 explores differences in the frequency of several variables that independently may affect the development of DM or CSS. Patients who developed an LR were significantly more likely to have the following characteristics: younger age (P < 0.001), clinical detection (P = 0.001), earlier diagnosis date (P < 0.001), positive reexcision (P < 0.001), close/positive margins (P < 0.001), and no adjuvant tamoxifen (P = 0.01).

Table 5. Additional Selected Patient Characteristics (All patients)
CharacteristicAll patients (%)aMatched pairs (%)a
No LRLRP valueNo LRTR/MMP value
  • LR: local recurrence; EIC: extensive intraductal component.

  • a

    All numbers indicate the % of breasts.

Age (yrs)      
 < 502849< 0.00138430.60
 ≥ 5072516257
Tumor size (cm)      
 ≤ 2.077700.2669710.85
 > 2.023303129
Lymph node status      
 Negative74760.4780801.0
 Positive18201414
 No dissection8466
Grade      
 119220.2716161.0
 225151616
 319202020
 Unknown37434545
Histology      
 Infiltrating ductal80800.8785780.54
 Infiltrating lobular98610
 Other1112912
ER status      
 Positive57460.1539391.0
 Negative16182424
 Unknown27363737
Method of detection      
 Mammographic48270.00143260.04
 Clinical52735774
Year of diagnosis      
 1980–19851739< 0.00129390.17
 1986–199032413422
 1991–199552203622
Reexcision      
 None4043< 0.00137470.005
 Negative3412318
 Positive26453145
Final margin status      
 Negative8151< 0.00175510.002
 Close (≤ 2 mm)10201516
 Positive516318
 Unknown412716
EIC      
 No88820.2087840.63
 Yes12181316
Tamoxifen      
 No64780.0178800.84
 Yes36222220
Chemotherapy      
 No82800.6481860.55
 Yes18201914

Time Course of DM

The mean (± standard deviation) time to the development of DM was 3.8 ± 2.8 years in patients who did not develop an LR compared with 4.7 ± 3.3 years in patients who developed an LR (P = 0.23). Patients who developed an LR had two peaks in the rate of development of DM (at 2.5 years and 6.5 years) compared with only one peak (at 1.5 years) in patients who did not develop an LR (Fig. 1). In addition, there was a very close association observed between the time to LR and the time (and rate) of DM, as illustrated in Figure 2. Patients who developed an LR within 3 years of treatment (n = 18 patients) had both a significantly greater risk of developing DM and a much shorter time to the appearance of DM compared with patients who developed an LR at 3–6 years (n = 23 patients) and patients who developed an LR after 6 years (n = 19 patients; P = 0.001). A similar trend also was found for regional failure (data not presented). Patients who developed a regional failure within the first 3 years of treatment had a significantly greater risk of developing DM compared with patients who developed a regional failure after 6 years.

Figure 1.

Hazard ratios for the development of distant metastasis (all patients).

Figure 2.

Distant metastasis calculated from the date of local recurrence (all patients).

Risk Factors for DM among Patients with LR

Factors that were associated with DM among patients who developed an LR included positive lymph nodes (63% vs. 20%; P = 0.01), clinical detection versus mammographic detection (36% vs. 12%, respectively; P = 0.03), total tumor bed RT dose ≤ 55 Gy (50% vs. 27%; P = 0.002), and regional failure (78% vs. 22%; P < 0.001) (see Table 6).

Table 6. Ten-Year Actuarial Rate of Distant Metastases Based on Failure Status
CharacteristicNo LRLR
No. of breasts10-yr DM (%)P valueNo. of breasts10-yr DM (%)P value
  1. LR: local recurrence; DM: distant metastases; ER: estrogen receptor; RT: radiotherapy; GY: grays.

Age (yrs)      
 < 50310170.00136320.50
 > 50816103827
Tumor size (cm)      
 ≤ 2.08629< 0.00152320.66
 > 2.0264212223
Lymph node status      
 Negative834105620
 Positive20419< 0.00115630.01
 No dissection8826333
Grade      
 12097< 0.00116100.13
 2280131139
 3219201533
 Unknown418103238
ER status      
 Positive641110.0234200.08
 Negative179151341
 Unknown306122736
Detection      
 Mammographic5376< 0.00120120.03
 Clinical589175436
Year of diagnosis      
 1980–198519020< 0.00129320.57
 1986–1990355123027
 1991–19955811531
Reexcision      
 None44611< 0.00132280.59
 Negative3878954
 Positive293193325
Final margin      
 Negative917110.0838400.36
 Close (≤ 2 mm)115171514
 Positive5225128
 Unknown427935
Total RT dose (Gy)      
 ≤ 55.038250.0048500.002
 > 55.01088116627
Tamoxifen      
 No716120.4858250.26
 Yes410121656
Chemotherapy      
 No92110< 0.00159230.09
 Yes200201555
Regional failure      
 No109910< 0.0016522< 0.001
 Yes2787978
All patients1126147429

Factors Associated with LR

The 6-year and 12-year actuarial rates of LR were 4% and 11%, respectively. Of 74 LRs that developed, 51 LRs (69%) were classified as a TR/MM, and 16% were E failures. The remaining failures were either skin (n = 6 LRs), diffuse (n = 2 LRs) or nonclassifiable (n = 3 LRs). Factors that were associated with LR were young age, clinical detection, earlier year of detection, positive reexcision, close/positive margins, and no tamoxifen. In a Cox multiple regression model, young age (P < 0.01) and margin status (P < 0.01) were associated independently with LR.

MPA

To reduce the confounding effects of competing risks on DM, an MPA was performed comparing outcome in patients with and without LR. In addition, to eliminate the confounding effects from recurrences unrelated to treatment of the index lesion (e.g., E failures), only the 51 patients who had TR/MM failures were used in the MPA. These 51 patients were matched with 102 unique patients (out of 1095) with no LR (1:2 match). Patients were matched for tumor size (± 5 mm), the same number of positive lymph nodes (0 vs. 1 vs. 2 vs. 3 vs. 4 vs. no ALND), age ± 10 years, same grade (Grade 1 vs. Grade 2 vs. Grade 3 vs. unknown), and the same estrogen receptor status (positive vs. negative vs. unknown). The mean values for selected patient characteristics of the MPA are listed in Table 7. Differences in the distribution of other critical variables also can be found in Table 5. The only remaining differences in clinical, pathologic, and treatment-related characteristics between the two patient groups in the MPA included reexcision status, margin status, and method of detection. There were no statistically significant differences in the distribution of all other critical variables, including patient age and length of follow-up.

Table 7. Mean Values for Selected Patient Characteristics: Matched-Pair Analysis
CharacteristicNo LRTRMMP valueaE failure
  • LR: local recurrence; TRMM: true recurrence/marginal miss; E failure: failure elsewhere; Gy: grays; RF: regional failure; DM: distant metastasis.

  • a

    Unpaired t test.

Age at diagnosis (yrs)53.453.00.8456.1
Tumor size (cm)1.91.90.981.5
Grade2.12.11.01.5
Lymph nodes    
 No. sampled14.513.20.2615.8
 No. positive0.220.221.00.25
Total dose (Gy)61.860.80.2162.0
Follow-up (yrs)9.210.30.0810.5
Time to LR (yrs)5.77.3
Time to RF (yrs)2.25.51.7
Time to DM (yrs)4.25.40.421.7

Table 8 lists 12-year actuarial outcome data between patients with and without LR. Patients who experienced TR/MM more frequently experienced regional lymph node failure (15% vs. 1%; P = 0.005) and DM (32% vs. 8%; P = 0.002) and had a worse CSS (74% vs. 93% P = 0.01).

Table 8. Twelve-Year Actuarial Data by Failure Group: Matched-Pair Analysis
Endpoint (at 12 yrs)Matched pair
No LRTRMMP valueE failure
  1. LR: local recurrence; TRMM: true recurrence/marginal miss; E failure: failure elsewhere.

Regional failure (%)1150.0058
Distant metastasis (%)8320.0028
Overall survival (%)85720.1392
Cause specific survival (%)93740.0192

Factors Associated with DM and CSS in the MPA

Table 9 lists factors that were associated with DM and CSS on multivariate analysis. The HR associated with LR was 6.43 for DM and 5.86 for mortality. The only other factor that was associated significantly with both DM and CSS in the MPA was negative estrogen receptor status. The development of E failures had no statistically significant impact on the rate of DM or CSS.

Table 9. Multivariate Analysis of Matched Pairs
FactorCSSDM
P valueHRP valueHR
  1. CSS: cause specific survival; DM: distant metastases; HR: hazard ratio; TR/MM: true recurrence/marginal miss; ER: estrogen receptor; RT: radiotherapy; Gy: grays.

TR/MM0.0085.860.0076.43
Negative ER status (vs. positive)0.0076.110.0095.85
Positive lymph nodes (vs. negative)0.810.81
Lower RT dose (Gy)0.190.30

Time Distribution of DM in Patients with LR

The time from completion of treatment to any given event in the 23 patients who developed DM or regional lymph node failure was analyzed. Again, the mean interval to the appearance of DM was longer for patients with LR compared with the interval for patients who achieved LC (5.4 years vs. 4.2 years, respectively). In addition, DM almost uniformly occurred sequentially after LR.

Risk Factors for DM in Patients who Developed LR

Factors that were associated with DM in patients who experienced a TR/MM failure were examined. Negative estrogen receptor status, regional lymph node failure, a lower RT dose, and a shorter interval to LR all were associated with DM. Age at diagnosis, tumor size, number of lymph nodes sampled, lymph node status, tumor grade, method of detection, year of diagnosis, and the use of chemotherapy or tamoxifen were not associated with DM after LR.

DISCUSSION

The current study was conducted to help clarify the impact of LR on the rate of DM and survival in patients with early-stage breast carcinoma who are treated with BCT. Although these results may be influenced by the retrospective nature of the study and the inability to control for all competing risks or censorship, they provide additional evidence suggesting that an LR is associated independently with the development of DM and may be responsible for a reduction in survival of patients with breast carcinoma. When analyzed for all 1169 patients or when restricted only to the 153 patients in the MPA, LR resulted in a dramatic increase in the rate of DM and a profound reduction in CSS. This effect was seen in patients with small tumors, negative lymph nodes, or both. In addition, the time course of development of DM was uniquely different in patients who developed an LR. Compared with patients who did not develop an LR, the time to DM in patients with LR was substantially longer and occurred almost uniformly after an LR (regardless of when the LR developed). Patients who experienced an LR also developed two separate peaks in the rate of DM (compared with patients who did not develop an LR), suggesting that LR may be responsible for a certain percentage of DM in these patients. Adding additional strength to this argument, recurrences that developed elsewhere in the breast (recurrences unrelated to the treatment of the index lesion) did not result either in an increase in the rate of DM or in a reduction in survival. These results provide additional indirect evidence suggesting that an LR may be related causally to DM and survival, reinforcing the necessity to provide the highest LC rate possible.

Although it has been shown in six prospective, randomized trials that BCT produces survival equivalent to mastectomy, sufficient clinical data have now emerged suggesting that small but real differences in the rate of DM and survival may be related to the development of LR. Unfortunately, establishing a cause-and-effect relation between LR and DM using retrospective data has been extremely difficult due to the effects of competing risks and the statistical bias introduced from analyzing an event (LR) that occurs at a variable time after primary treatment. Gelman and Harris pointed out eloquently that convincing evidence for a survival detriment from LR may be obtained best from formal randomized trials.8 Even if such trials could be performed, however, establishing a survival benefit would be difficult, because it most likely would be quite small and require large numbers of patients.2 Thus, currently, proof that an LR is a direct cause of DM can be extrapolated only from retrospective data with the inherent limitations discussed above.

Several recent retrospective studies with large numbers of patients have addressed the impact of LR on the development of DM. Koscielny and Tubiana reviewed the outcome of 4144 consecutive patients with breast carcinoma who underwent radical surgery without receiving chemotherapy at the Institut Gustave Roussy from 1954 to 1975. Patients who developed LR, as consistently observed previously, had a much greater rate of DM over 20 years (84%) compared with patients who were without LR (46%).9 This was true even when stratifying by various prognostic factors, such as tumor size and lymph node status. In addition, those authors demonstrated that the appearance of DM in patients without LR was highest in the first year after treatment and decreased continuously. In contrast, in patients with LR, the appearance of DM was delayed, with the rate increasing during the first year. More importantly, the authors then showed how DM occurred subsequent to LR in many patients. Assuming that metastases occurred prior to diagnosis, the median growth duration of DM was calculated at 67 months in patients without LR and 194 months in patients with LR. This indicates that LR does not reflect biologic aggressiveness independently. The authors argued that, because it has been shown that tumor kinetics are more rapid in patients with a poorer prognosis, their data imply that, in many patients with LR, DM must occur after treatment.

In a similar fashion, Fortin et al. reviewed their institution's outcome in treating 2030 patients with breast carcinoma.4 LR led to poorer survival at 10 years compared with LC (55% vs. 75%; P < 0.00), as expected. In a Cox multivariate model, LR was a powerful predictor of mortality. The relative risk associated with LR was 3.6 for mortality and 5.1 for DM (P < 0.00). In patients with LR, the rate of DM peaked at 5–6 years, whereas it peaked at 2 years for patients who achieved LC. The mean time between surgery and DM was 1050 days for patients without LR and 1650 days for patients with LR (P < 0.00).

The objectives of the current study were to examine the impact of LR on DM and survival by reducing the effect of competing risks using a MPA and to eliminate the confounding effects of recurrences elsewhere in the breast by not including them in statistical analyses. Using these two techniques, the probability of documenting an independent effect of LR on DM and survival, theoretically, was increased. Similar to the previous studies outlined above, our results clearly document an increase in DM and disease specific mortality with LR. Likewise, our data also establish a remarkably similar time course for the development of DM in patients with and without LR. Perhaps the most unique findings from our study relate to the interrelations observed between the time from surgery to LR and the subsequent time to and rate of DM. Figure 2 shows that patients who developed an LR shortly after surgery (< 3 years) not only had a higher rate of DM but also had a shorter time to their development. Because the interval of time increased from surgery to local recurrence, so did the time to and freedom from DM. In addition, similar to the study by Fortin et al.,4 patients who developed an LR experienced a second delayed peak in the rate of development of DM that most logically can be attributed to a secondary dissemination of tumor cells due to the LR event.

Despite all of the efforts made to eliminate bias, confounding effects, and competing risks, it still is conceivable that LR merely reflects greater tumor aggressiveness. The issue of censorship, as pointed out by others, can never be eliminated fully. Patients who develop DM are generally less likely to be evaluated for the presence or development of LR. At the same time, once an LR has been documented, a patient is more likely to be followed and evaluated for the development of DM. Clearly, these two factors may have profound effects on the interpretation of statistical data based on time dependent events. Unfortunately, it is impossible to know whether these effects can be eliminated completely when analyzing potentially biased retrospective data.

The only conclusive method to prove that LR results in an increased risk of mortality, as discussed above, would be a Phase III study in which patients who are randomized to receive additional local therapy (e.g., RT) not only experienced a lower rate of LR but also a reduction in DM and disease specific death. Several recent studies have demonstrated just such an effect.7, 15 In the study by Ragaz et al. of postmastectomy RT,7 the rate of locoregional recurrence was reduced by 56% (relative risk, 0.44; 95% confidence interval, 0.26–0.77; P = 0.003), and the rate of systemic recurrence was reduced by 34% (relative risk, 0.66; 95% confidence interval, 0.49–0.89; P = 0.006) in the group of patients who were treated with combined therapy. Those authors believe their data support the concept that locoregional recurrence is not only a marker of systemic disease but also, in some patients, a potential source for its future dissemination (similar to the Stockholm trial in patients with positive lymph nodes16).

Prospective, randomized trials of lumpectomy alone versus lumpectomy with RT also have suggested a potential increased risk of DM and reduced survival in patients who did not receive RT. Unfortunately, these differences in outcome have been quite small; consequently, proving a statistically significant association has been difficult.2 This point is discussed eloquently in a recent review by Fowble,2 who points out that there are three patterns of LR that can develop after treatment with BCT. First, an LR may be an isolated event that is salvageable with additional local therapy. This type of LR theoretically should have no impact on survival, and data suggest that it accounts for 60% of all local failures after CS and RT. Second, an LR can develop simultaneously with DM. This type of recurrence supports the concept of a local failure as a marker for established distant disease, and it has been observed in about 10% of patients who have local failures after BCT. The third type of failure assumes that an LR can lead to an increased risk of DM through a direct causal relation. This is the only type of LR that, if it is prevented, may lead to an improvement in survival. Assuming that 30% of patients who develop an LR will experience this type of failure (a number supported by a review of multiple studies on CS and RT), even a 20–30% improvement in the LC rate, at best, may translate into a maximum survival advantage of only 6–9% (by reducing the subsequent development of DM). Most randomized trials exploring the use of RT after lumpectomy or mastectomy are expected to have a much smaller improvement in the LC rate. Consequently, the potential survival advantage from RT also would be expected to be much smaller, necessitating either larger patient numbers and/or an extended follow-up to demonstrate statistical significance with RT.

Assuming that LR results in an independent increase in DM, it would be important to identify and reduce the factors responsible for this event. In our MPA, the only statistically significant differences in clinical, pathologic, or treated-related characteristics between patients with LR and without LR were variables related to the adequacy of tumor excision. Patients who developed an LR more frequently had close/positive margins, did not undergo reexcision, or underwent a positive reexcision. In addition, when analyzing the factors associated with LR, margin status and young age were the only significant variables on multivariate analysis. These results are similar to most published studies that have addressed this issue and suggest that it may be possible to influence the rate of DM by reducing the risk of LR through improved surgical removal.

Several recent studies also have confirmed the importance of adequacy of excision and the effect that margin positivity can have on DM. Voogd et al.17 analyzed data from two randomized clinical trials (European Organization for Research and Treatment of Cancer trial 10801 and Danish Breast Carcinoma Cooperative Group trial DBCG-82TM)in which 879 of 1772 patients with Stage I and II breast carcinoma underwent CS and received RT. Representative slides were available for review in 91% of patients. Microscopic involvement of excision margins was associated independently with distant disease. The risk of DM in these patients was twice as high compared with the risk of DM in patients with negative margins. This independent negative effect of positive margins on the development of DM also was observed in studies by Fortin et al.,4 Schnitt et al.,18 Macmillan et al.,19 and DiBiase et al.20 (see Table 10). In the current study, patients who received a total tumor bed dose ≤ 55 Gy (n = 46 patients) also experienced a significantly greater 10-year actuarial risk of developing LR (18% vs. 8%; P = 0.01) and DM (30% vs. 13%; P < 0.001) compared with patients who received > 55 Gy. All of these data clearly suggest that serious consideration must be given to the concept that differences in local treatment (whether through the addition of RT and/or through more comprehensive tumor resection) will affect outcome.

Table 10. Studies Showing an Increase in Metastases or a Decrease in Survival in Patients with Positive/Close Margins
InstitutionNo. of patientsStudy parametersOverall findingsResults
  1. EORTC: European Organization for Research and Treatment of Cancer; BCT: breast-conserving therapy; DM: distant metastases; DFS: disease free survival; ER: estrogen receptor.

Netherlands EORTC trials171772Two randomized trials: complete pathologic review (91% of patients); mastectomy and BCTIndependent increase in distant disease with macroscopic positive margins with BCTTen-yr risk of distant disease; 60% vs. 29% in patients with positive margins
Quebec42030Retrospective analysis of BCT patients only; no pathology reviewIndependent negative effect on survival with positive marginsClose margins = increase in DM (28% vs. 17% of patients (P = 0.01)
Joint Center for Radiation Therapy18181Retrospective review of BCT patients onlyIncrease in distant failure with positive margins; however, increase in positive lymph nodes with positive marginsDistant failure risk: 25% vs. 14% in patients with positive margins within 5 yrs
Thomas Jefferson University20453Retrospective review of BCT patients onlyDecrease in survival at 10 yrs with positive margins84% vs. 78% 10-yr survival with negative margins (P = 0.047)
Glasgow University19300Retrospective review of BCT patients onlyPatients with positive margins were two times more likely to develop DM vs. patients with negative marginsDistant DFS 50% with positive martins and negative ER status vs. 90% with negative margins and positive ER status

CONCLUSIONS

Our results suggest that LR may be responsible for an increase in DM and disease specific mortality in patients who are treated with BCT. This is demonstrated by the finding of a distinct difference in the time distribution of DM after LR (even after controlling for multiple prognostic factors) and by the fact that recurrent disease that originated outside of the surgical bed did not affect OS. These data suggest that efforts should be made to optimize LC in patients who are treated with BCT.

Ancillary