World J Surg (2014) 38:1070–1076 DOI 10.1007/s00268-013-2387-9
A Simple Risk Score to Predict the Presence of Non-Sentinel Lymph Node Metastases in Breast Cancer Patients with a Positive Sentinel Node Raquel F. D. van la Parra • Petronella G. M. Peer Wilfred K. de Roos • Miranda F. Ernst • Johannes H. W. de Wilt • Koop Bosscha
•
Published online: 5 December 2013 Ó Socie´te´ Internationale de Chirurgie 2013
Abstract Background Historically, completion axillary lymph node dissection (cALND) is recommended in sentinel lymph node (SLN)-positive patients. However, the high rate of negative non-sentinel nodes (NSNs) in cALND and the reported low axillary recurrence rates have led to a more conservative approach. A risk score was developed to identify a patient’s individual risk for NSN metastases. Methods Data of 182 SLN-positive patients who underwent cALND were used for risk score development. The risk score, consisting of pathological tumor size (B20/ [20 mm), lymphovascular invasion (no/yes), extracapsular extension (no/yes), size of metastases (B2/[2 mm), and number of positive SLNs (1/[1), was subsequently validated on an external population (n = 180).
Results The area under the receiver operating characteristic curve was 0.78 (95 % CI 0.71–0.85) in the original population and 0.78 (95 % CI 0.70–0.85) in the validation population. Based on the predicted risk for positive NSNs, three groups were defined: low risk (B20 %), intermediate risk (21–50 %), and high risk ([50 %). In total, 88 patients met the Z0011 inclusion criteria and none of them had a high predicted risk. Of the 199 non-Z0011 patients, 67 (33.7 %) had low risk, 96 (48.2 %) had intermediate risk, and 36 (18.1 %) had high risk. Conclusion A simple risk score, integrating just five clinicopathological variables, was developed that may assist in individual decision making regarding ALND in SLN-positive patients outside of the Z0011 trial.
Introduction R. F. D. van la Parra (&) Department of Surgery, The Netherlands Cancer Institute-Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands e-mail: [email protected] P. G. M. Peer Department for Health Evidence, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands W. K. de Roos Department of Surgery, Gelderse Vallei Hospital, Ede, The Netherlands M. F. Ernst K. Bosscha Department of Surgery, Jeroen Bosch Hospital, ’s-Hertogenbosch, The Netherlands J. H. W. de Wilt Department of Surgery, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands
123
Over the past 20 years, axillary management in breast cancer has evolved from routine axillary lymph node dissection (ALND) to sentinel lymph node (SLN) biopsy, which has been proven to reliably identify nodal involvement in early-stage breast cancer. Although ALND can be safely avoided in SLN-negative patients, it has generally been recommended in patients with positive SLNs. However, a more conservative approach of the axilla has recently been proposed since only 40–60 % of patients are found to have additional non-sentinel node (NSN) metastases [1]. A considerable number of patients are thus exposed to the morbidity of ALND without a potential therapeutic benefit. The recently published results of the American College of Surgeons Oncology Group (ACOSOG) Z0011 trial, which compared standard ALND with no further specific axillary treatment in SLN-positive patients, demonstrated that the axillary recurrence rate was
World J Surg (2014) 38:1070–1076
low for both the SLN-only group and the ALND group [2]. Moreover, long-term results also demonstrated that overall survival was not compromised in the different treatment arms [3]. However, the trial results did only apply to patients with favorable tumor characteristics, e.g. patients with a T1 or T2 tumor treated with breast conservative surgery (including tangential field irradiation) with less than three involved SLNs and no extracapsular extension (ECE). Since only a few breast cancer patients meet the Z0011 criteria as demonstrated by Gu¨th et al. [4], a risk score could still be valuable in decision making regarding completion ALND (cALND) in SLN-positive patients outside the Z0011 trial inclusion criteria. Several predictive tools have been created over the past years to identify patients with the lowest risk of NSN metastases [5–15]. The Memorial Sloan Kettering Cancer Center (MSKCC) nomogram has been the most utilized [5]. Unfortunately, underestimation of especially the lowrisk groups has been reported in some series validating this nomogram. In a recent meta-analysis of our group, eight clinicopathological variables were identified, each of which elevates the likelihood of NSN metastases at a relevant (odds ratio C2.0) and statistically significant level [1]. Our goal was to create a new, simple model based on these clinically most relevant variables to predict the likelihood of NSN metastases. This nomogram was then studied for patients who met the Z0011 criteria and for those who did not.
1071
Patients who received neoadjuvant therapy were excluded. All patients routinely underwent preoperative axillary ultrasonography. If nodal metastases were documented preoperatively by fine needle aspiration cytology, patients underwent axillary dissection without SLN biopsy. These patients were also excluded from the study. Predictive variables Our previous meta-analyses identified eight clinicopathological variables that corresponded with the highest likelihood of NSN metastases: method of detection of SLN metastases (IHC only vs. other), size of the SLN metastases (B2/[2 mm), number of negative SLNs (B1/[1), number of positive SLNs (1/[1), ratio of positive SLNs (=number of positive SLNs divided by total number of SLNs, B50/ [50 %), extracapsular tumor extension, ECE (no/yes), tumor size (B20/[20 mm), and lymphovascular invasion, LVI (no/yes; [1]). In a previous study, we demonstrated that the method of detection relates to the size of metastases [16]. Basically, macrometastases are detected by standard haematoxylin and eosin (HE), micrometastases are detected by serial HE, and isolated tumor cells are detected by immunohistochemistry. Since the size of the metastases was known in both populations, we decided to exclude the variable ‘method of detection’ from our model, leaving seven variables. SLN mapping
Materials and methods Populations Two independent patient populations (from different teaching hospitals, where the first author worked as a surgical registrar) were used: the JBH (Jeroen Bosch Hospital) population was used for development of the model (development group) and included patients with SLN metastases who underwent ALND between January 2000 and December 2007. These were filtered from the original cohort of 913 breast cancer patients who underwent SLN biopsy. Of these, 261 had a positive SLN and 182 underwent cALND. A level I–II axillary dissection was routinely performed in case of a positive SLN. To demonstrate reliability, the model was validated on the GVH (Gelderse Vallei Hospital) population (validation group), which included patients with SLN metastases who underwent ALND between January 2004 and December 2010. These were filtered from the validation cohort of 941 breast cancer patients who underwent SLN biopsy. Of these, 260 had a positive SLN and 180 underwent cALND.
In the development group, lymphatic mapping with SLN biopsy was performed using both blue dye and radioisotope, as described previously [16]. In the validation group, lymphatic mapping was performed using radioisotope with or without blue dye at the discretion of the surgeon. All blue and/or radioactive nodes counting 10-fold ex vivo relative to the background were regarded as SLNs. Histopathological evaluation Frozen section analyses were only performed in the GVH group. SLNs with a diameter of less than 5 mm were frozen intact, whereas SLNs with a diameter of more than 5 mm were bisected longitudinally and frozen. Frozen sections of the SLN were taken from the optimal crosssectional surface with a microtome setting of 4 lm. Definitive pathological analysis in both hospitals consisted of serially sectioning the SLN at four levels of 250-lm intervals along the longitudinal axis for permanent section. From every section, two parallel slides of 4 lm were made. All slides were stained with HE and with anticytokeratin 8. The nodes removed by cALND were stained with HE.
123
1072
World J Surg (2014) 38:1070–1076
Statistical analysis Patient and tumor data were collected from the databases for each variable. The criteria for accurate model development, as stated by Coutant et al. [17] were applied. To deal with the problem of missing values in the predictors (LVI and ECE), multiple imputed datasets were constructed. The predictive distribution of a dichotomous factor was obtained by logistic regression on the observed data. A random drawing from this predictive distribution was imputed for the missing value. When each missing value was imputed, a complete dataset was obtained. This process was repeated 20 times. The relation between the predictors and the presence or absence of NSN metastases was assessed by the pooled result of the multivariable logistic regression analyses on each imputed JBH dataset. The ability of the pooled model to discriminate between patients with and without NSN metastases in the GVH dataset was quantified by the pooled area under the receiver operating characteristic Table 1 Clinicopathological variables and their association with the incidence of additional NSN metastases in the development and the validation group
curve (AUC). GVH patients were grouped into intervals according to their pooled predicted probability of NSN metastasis (B0.05, 0.05 to B0.10, 0.10 to B0.20, etc.). A calibration plot was drawn showing the actual (women who actually had positive NSNs) versus the mean of the pooled predicted percentage for each interval. To construct an easily applicable risk score, we transformed the regression coefficients of the predictors in the pooled model to integers according to their relative contribution to the risk estimation. Statistical analyses were conducted with SPSS 16.0.
Results Patient characteristics for both populations are presented in Table 1. There was a comparable distribution for tumor size, the number of positive SLNs, and metastasis size between both populations. LVI was more prevalent in the JBH population (37.2 vs. 17.7 %), however, positive SLN
Jeroen Bosch Hospital
Gelderse Vallei Hospital
Development group (n = 182)
Validation group (n = 180)
Positive NSN
Positive NSN
Tumor size (mm) B20
114 (62.6)
29 (25.4)
114 (63.3)
28 (24.6)
[20
68 (37.4)
28 (41.2)
66 (36.7)
23 (34.8)
Absent
108 (62.8)
33 (30.6)
121 (82.3)
37 (30.6)
Present
64 (37.2)
23 (35.9)
26 (17.7)
11 (42.3)
Missing
10
LVI
No. positive SLNs 1
33
149 (81.9)
42 (28.2)
152 (84.4)
38 (25.0)
33 (18.1)
15 (45.5)
28 (15.6)
13 (46.4)
B1
163 (89.6)
54 (33.1)
148 (82.7)
39 (26.4)
[1
19 (10.4)
3 (15.8)
31 (17.3)
11 (35.5)
[1 No. negative SLNs
Missing
1
Size of metastasis (mm) B2
55 (30.9)
4 (7.3)
[2
123 (69.1)
53 (43.1)
Missing
4
59 (33.3)
5 (8.5)
118 (66.7)
46 (39.0)
3
ECE No
123 (70.7)
28 (22.8)
86 (74.8)
22 (25.6)
Yes
51 (29.3)
29 (56.9)
29 (25.2)
22 (75.9)
All data are presented as n (%)
Missing
ECE extracapsular extension, LVI lymphovascular invasion, NSN non-sentinel nodes, SLNs sentinel lymph nodes
Z0011 Non-Z0011
47 (27.3) 125 (72.7)
Uncategorizeda
8
a
Due to missing ECE status
123
Total
8
65 13 (27.7) 44 (35.2)
41 (35.7) 74 (64.3)
9 (22.0) 35 (47.3)
65 57 (31.3)
51 (28.3)
World J Surg (2014) 38:1070–1076
1073
rate was almost equal in both centers (35.9 and 42.3 %) for patients with LVI. Almost 70 % of our population were patients who did not meet the Z0011 inclusion criteria due to irradical excision (n = 16), multicentricity (n = 34), presence of ECE (n = 80), more than three SLNs removed (n = 1), mastectomy performed (n = 157), and/or tumor size over 50 mm (n = 3). Additional NSN positivity was lower in patients who met the Z0011 criteria (25 vs. 40 % of the non-Z0011 patients). After cALND, a median total of 12 nodes (range 4–23) were harvested in the JBH group, and 15 nodes (range 6–37) in the GVH group. Initially, three predictive models were developed that differed in the way the variables positive and/or negative SLNs were incorporated (e.g. model I only SLN positive, model II SLN positive and negative, model III ratio of positive SLNs). Model I included the following dichotomous variables: tumor size, LVI, number of positive SLNs, metastasis size, and ECE. All three models had an AUC of [0.7 on external validation, but model I performed the best on calibration of the low predicted risks (B0.05 or B0.10), because it corresponded best to the reference line. Models II and III, which included the number of SLN-negative patients performed worse on predicting low probabilities. Based on this performance, model I became the final model. Using the formula of the logistic regression model, a patient’s probability of NSN metastases based on her clinicopathological profile was established. To facilitate use in daily practice, the final model was simplified to an applicable scoring rule. Regression coefficients of the
Fig. 1 Graphical representation of total scores and corresponding predicted risks for positive non-sentinel nodes
variables in the final model were rounded to integers according to their relative contribution to the risk estimate (Table 2). As such, five points were given for the presence of LVI, six points for a tumor size [2 cm, 7 points for the presence of more than one positive SLN, 14 points for the presence of ECE, and 20 points for a metastases size [2 mm. Subsequently, a sum score is obtained for each patient. The total score ranges from 0 to 52. Each score corresponds with a predicted percentage of NSN metastases, which can be extracted from Fig. 1. The final model was validated externally on the GVH population demonstrating an AUC of 0.78 (95 % CI
Table 2 Risk score to predict the likelihood of NSN metastases in patients with a positive SLN Variables
Coefficients
Point values
LVI Absent Present
0 0.432
5
0.527
6
Tumor size (cm) B2 [2
0
No. positive SLNs B1 [1
0 0.622
7
1.247
14
1.716
20
ECE No Yes
0
Size of metastasis (mm) B2 [2
0
Model formula P ¼ eð3:072þ0:432LVIþ0:527tumour sizeþ0:622no1 positive SNsþ1:247ECEþ1:716size of metastasesÞ ECE extracapsular extension, LVI lymphovascular invasion, NSN non-sentinel node, SLN sentinel lymph nodes
123
1074
World J Surg (2014) 38:1070–1076
0.70–0.85). Patients were classified into intervals of the predicted risks, and actual probabilities were plotted against the predicted probability of NSN metastases (Fig. 2). The false-negative rate was 3.3 % in the group with a predicted probability of NSN metastases of \5 %.
Three risk groups were constructed based on the predicted probabilities (Table 3): a low risk group (predicted risk B20 % for NSN metastases), intermediate risk (21–50 %), and high risk ([50 %). In the low-risk group, 4 of 58 patients (6.9 %) had NSN metastases. For the intermediate- and high-risk groups this was 27 of 96 patients (28.1 %) and 20 of 26 (76.9 %), respectively. Of the 287 breast cancer patients with a known ECE status from both populations, 88 patients (30.7 %) met the Z0011 inclusion criteria, and of these, 39 (44.3 %) were classified as low risk and 49 (55.7 %) as intermediate risk. Of the 199 patients who did not meet the Z0011 inclusion criteria, 67 (33.7 %) had a low risk, 96 (48.2 %) an intermediate risk, and 36 (18.1 %) a high risk of NSN metastases (Table 4).
Discussion
Fig. 2 Calibration plot for risk score to predict positive non-sentinel nodes. The entire cohort was separated into nine groups of varying sizes according to their predicted percentages (1–10, 11–20 %, etc.). For each group, the actual percentage was calculated. A calibration plot was drawn showing for each interval the actual versus predicted percentage. If the model was perfect, all dots would lie on the line, with a slope of 1
Table 3 Actual percentage for NSN metastases in the validation group Risks
Intervals
Total
Positive NSNs, n (%) No
Yes (=actual)
Low
0 to 0.20
58
54 (93.1)
4 (6.9)
Intermediate
0.21 to B0.50
96
69 (71.9)
27 (28.1)
High
0.51 to B0.80
26
6 (23.1)
20 (76.9)
180
129 (71.7)
51 (28.3)
Total NSN non-sentinel node
Table 4 Total number of patients with or without Z0011 criteria Risks
Intervals
Total
Z0011, n (%)
Non-Z0011, n (%)
Low
0 to 0.20
106
39 (44.3)
67 (33.7)
Intermediate
0.21 to B0.50
145
49 (55.7)
96 (48.2)
High
0.51 to B0.80
36
–
36 (18.1)
88*
199*
Total
287
* Numbers calculated on patients with a known extracapsular extension status
123
A simple risk score, integrating just five clinicopathological variables, was developed to assist in individual decision making regarding ALND in SLN-positive patients outside of the Z0011 trial. With this risk score, three groups were established with low, intermediate, or high risk on additional NSNs in the ALND. In total, 88 patients met the Z0011 inclusion criteria and none of these patients had a high predicted risk. Of the 199 patients who did not meet the Z0011 inclusion criteria, 67 (33.7 %) had a low risk, 96 (48.2 %) an intermediate risk, and 36 (18.1 %) a high risk of NSN metastases. The risk of additional NSN metastases is \7 % in the low-risk group, 28.1 % for the intermediate-risk group, and 76.9 % for the high-risk group. External validation of the risk score demonstrated considerable discrimination (AUC 0.78). The false-negative rate was 3.3 % in the group with a predicted probability of NSN metastases of \5 %. ALN status is traditionally regarded as an important prognosticator in breast cancer. ALND is considered the standard of care when the SLN is positive, because it could give additional information on the total number of involved nodes and it was assumed to reduce the local recurrence rate. In our study groups, only 57/182 (31.3 %) and 51/180 (28.3 %) patients with SLN metastases had positive NSNs, which illustrates overtreatment of the axilla in the majority of patients and predisposes to associated morbidity such as lymphedema and axillary numbness. Several predictive models have been developed over past years in order to try to identify patients at low risk of NSN metastases [6–15]. These models identified different parameters as significant predictors of NSN metastases. Several parameters have been more commonly included in
World J Surg (2014) 38:1070–1076
models and were also identified as independent predictors in our previous meta-analysis [1]. The predictive models for NSN metastases developed thus far, including the MSKCC model, were derived from multivariable analyses of their original populations, which varied in size from 71 to 791 patients, with a variable number of predictors (between 3 and 7) [6–15]. External validation of our risk score has been performed; this was only performed in some of the other published models [5, 8, 12–14]. The MSKCC nomogram does not contain a cut-off score that mandates the performance of an ALND. Also, the MSKCC nomogram tends to underestimate the probability of NSN metastases in the lower range. The present risk score is based on the clinically most significant predictors of NSN metastases as identified in a previous meta-analysis [1]. Since our risk score is only based on five dichotomous variables, the score is easy to calculate and the associated risk can be extracted from the figure. The risk score is unique in that it provides cut-off scores, below which cALND can be safely avoided and above which cALND should be performed. Three levels of risk were identified with an associated axillary treatment recommendation based on the estimates for NSN involvement of 6.9, 28.1, and 76.9 %, respectively. We propose to avoid ALND in the low-risk group since the associated risk of additional NSN metastases is only 6.9 %. ALND should be discussed in the intermediate-risk group (associated risk 28.1 %). We strongly recommend performing ALND in the high-risk group (associated risk 76.9 %) until other studies have shown that ALND can be safely avoided. The high rate of negative NSNs in ALND and the low axillary recurrence rates reported in recent trials [18] have led to a shift towards a more conservative approach of the axilla with a declining role of ALND. In the ACOSOG Z0011 study, SLN-positive patients with early-stage breast cancer were randomized to ALND or not. At a median follow-up of 6.3 years, there were no statistically significant differences in local recurrence (1.8 % in the SLN-only group, n = 446 vs. 3.6 % in the ALND arm, n = 445; [2]). Furthermore, 5-year overall survival was 91.9 versus 92.5 % for the SLN-only group, and disease-free survival was 82.2 versus 83.8 % for the SLN-only group [3]. Although results are impressive, there are some aspects of the Z0011 study that limit its general applicability. The selection of a low-risk population and the effect of adjuvant systemic therapy on residual microscopic disease may be reflected in this low recurrence rate. Second, all patients received tangential field irradiation of the breast, which covers at least part of the axilla. Therefore, all patients received some kind of axillary treatment, which may in part account for the low rates of axillary recurrence.
1075
In light of the Z0011 trial, some have questioned the need for risk scores/nomograms in clinical decision making. However, the Z0011 results only apply to low-risk patients treated with breast-conserving surgery. Gu¨th et al. [4] tested the applicability of the Z0011 criteria on their patient population, which comprised 389 consecutive breast cancer patients who were treated with SLN biopsy. Only 9 % (35 patients) of their population met the Z0011 inclusion criteria, which demonstrates that the Z0011 results only apply to a minority of all patients who underwent SLN biopsy for breast cancer. In the present study, only 88 patients (30.7 %) met the Z0011 criteria (Table 4). This percentage is higher than that reported in the study by Gu¨th et al. [4], since we only reported the number of SLN-positive patients, while Gu¨th et al. reported on all SLN-positive and -negative patients. So, if ALND is to be avoided in the Z0011 patients, this still leaves 69.3 % of SLN-positive patients outside of the Z0011 inclusion criteria, for whom we have to decide whether or not ALND should be performed. Recently, Milgrom et al. [19] showed a similar outcome in SLN-positive patients with early-stage breast cancer who underwent total mastectomy without cALND or radiotherapy. However, of the 210 mastectomy patients with a positive SLN, there were only 19 patients (9.1 %) with N1 (macrometastatic) disease [113 patients (54 %) had N0(i?) disease, and 78 patients (37 %) N1mic]. The numbers are too small to draw firm conclusions and to avoid ALND in SLN-positive mastectomy patients. In daily practice, the score described in the present study can be used for an individualized risk estimate in patients with SLN metastases. Since predictive scores provide a percentage as outcome, a low-risk group was identified with an associated risk for NSN metastases of \7 %. Since the risk score gives a risk estimate, other factors are also involved in our decision making, e.g. patient age, comorbidity, patient preference, and surgeon’s experience. Some limitations apply to this study. The risk score was developed and validated in a relatively small patient population. However, some previous nomograms were developed in even smaller populations [6, 7, 9, 11]. To prove its strength, it needs to be validated prospectively on larger (external) populations. Due to the retrospective nature of the study, we had some missing variables. However, to deal with the problem of missing values, we constructed multiple imputed datasets. Furthermore, preoperative axillary ultrasound was routinely performed in both populations. Preoperative axillary ultrasound selects patients with a greater nodal burden, so it is likely that its routine use would reduce overall axillary disease burden in those undergoing SLN biopsy. Since it was applied in both the development and the validation population this could explain the good results on external validation. Despite
123
1076
World J Surg (2014) 38:1070–1076
differences in the SLN mapping technique and pathological analyses (frozen section in GVH) between both hospitals, validation remained good. This demonstrates the strength of the model when applied to a dataset outside the institution of origin.
Conclusion We succeeded in developing a simple risk score, integrating only five clinicopathological variables to provide an individualized risk estimate of the likelihood for NSN metastases when the SLN is positive. Three levels of risk were identified, with an associated axillary treatment recommendation based on the estimates for NSN involvement of 6.9, 28.1, and 76.9 %, respectively. This tool may assist in individual decision making regarding ALND in SLNpositive patients, especially in those who do not meet the Z0011 trial inclusion criteria. Acknowledgments The authors thank W. Lemmens for performing the data analysis. No grant support was received for this work. Conflict of Interest
The authors have no conflicts of interest.
References 1. van la Parra RF, Peer PG, Ernst MF et al (2011) Meta-analysis of predictive factors for non-sentinel lymph node metastases in breast cancer patients with a positive SLN. Eur J Surg Oncol 37:290–299 2. Giuliano AE, McCall L, Beitsch P et al (2010) Locoregional recurrence after sentinel lymph node dissection with or without axillary dissection in patients with sentinel lymph node metastases: the American College of Surgeons Oncology Group Z0011 randomized trial. Ann Surg 252:426–432; discussion 32–33 3. Giuliano AE, Hunt KK, Ballman KV et al (2011) Axillary dissection vs no axillary dissection in women with invasive breast cancer and sentinel node metastasis: a randomized clinical trial. JAMA 305:569–575 4. Gu¨th U, Myrick ME, Viehl CT et al (2012) The post ACOSOG Z0011 era: does our new understanding of breast cancer really change clinical practice? Eur J Surg Oncol 38:645–650 5. Van Zee KJ, Manasseh DM, Bevilacqua JL et al (2003) A nomogram for predicting the likelihood of additional nodal metastases in breast cancer patients with a positive sentinel node biopsy. Ann Surg Oncol 10:1140–1151
123
6. Hwang RF, Krishnamurthy S, Hunt KK et al (2003) Clinicopathologic factors predicting involvement of nonsentinel axillary nodes in women with breast cancer. Ann Surg Oncol 10:248–254 7. Saidi RF, Dudrick PS, Remine SG, Mittal VK (2004) Nonsentinel lymph node status after positive sentinel lymph node biopsy in early breast cancer. Am Surg 70:101–105 8. Degnim AC, Reynolds C, Pantvaidya G et al (2005) Nonsentinel node metastasis in breast cancer patients: assessment of an existing and a new predictive nomogram. Am J Surg 190:543– 550 9. Barranger E, Coutant C, Flahault A et al (2005) An axilla scoring system to predict non-sentinel lymph node status in breast cancer patients with sentinel lymph node involvement. Breast Cancer Res Treat 91:113–119 10. Chagpar AB, Scoggins CR, Martin RC II et al (2006) Prediction of sentinel lymph node-only disease in women with invasive breast cancer. Am J Surg 192:882–887 11. Pal A, Provenzano E, Duffy SW et al (2008) A model for predicting non-sentinel lymph node metastatic disease when the sentinel lymph node is positive. Br J Surg 95:302–309 12. Kohrt HE, Olshen RA, Bermas HR et al (2008) New models and online calculator for predicting non-sentinel lymph node status in sentinel lymph node positive breast cancer patients. BMC Cancer 8:66 13. Cho J, Han W, Lee JW et al (2008) A scoring system to predict nonsentinel lymph node status in breast cancer patients with metastatic sentinel lymph nodes: a comparison with other scoring systems. Ann Surg Oncol 15:2278–2286 14. Coufal O, Pavlik T, Fabian P et al (2009) Predicting non-sentinel lymph node status after positive sentinel biopsy in breast cancer: what model performs the best in a Czech population? Pathol Oncol Res 15:733–740 15. Perhavec A, Perme MP, Hocevar M et al (2010) Ljubljana nomograms for predicting the likelihood of non-sentinel lymph node metastases in breast cancer patients with a positive sentinel lymph node. Breast Cancer Res Treat 119:357–366 16. van la Parra RF, Ernst MF, Bevilacqua JL et al (2009) Validation of a nomogram to predict the risk of nonsentinel lymph node metastases in breast cancer patients with a positive sentinel node biopsy: validation of the MSKCC breast nomogram. Ann Surg Oncol 16:1128–1135 17. Coutant C, Olivier C, Lambaudie E et al (2009) Comparison of models to predict nonsentinel lymph node status in breast cancer patients with metastatic sentinel lymph nodes: a prospective multicenter study. J Clin Oncol 27:2800–2808 18. Francissen CM, Dings PJ, van Dalen T et al (2012) Axillary recurrence after a tumor-positive sentinel lymph node biopsy without axillary treatment: a review of the literature. Ann Surg Oncol 19:4140–4149 19. Milgrom S, Cody H, Tan L et al (2012) Characteristics and outcomes of sentinel node-positive breast cancer patients after total mastectomy without axillary-specific treatment. Ann Surg Oncol 19:3762–3770
A Simple Risk Score to Predict the Presence of Non-Sentinel Lymph Node Metastases in Breast Cancer Patients with a Positive Sentinel Node Raquel F. D. van la Parra • Petronella G. M. Peer Wilfred K. de Roos • Miranda F. Ernst • Johannes H. W. de Wilt • Koop Bosscha
•
Published online: 5 December 2013 Ó Socie´te´ Internationale de Chirurgie 2013
Abstract Background Historically, completion axillary lymph node dissection (cALND) is recommended in sentinel lymph node (SLN)-positive patients. However, the high rate of negative non-sentinel nodes (NSNs) in cALND and the reported low axillary recurrence rates have led to a more conservative approach. A risk score was developed to identify a patient’s individual risk for NSN metastases. Methods Data of 182 SLN-positive patients who underwent cALND were used for risk score development. The risk score, consisting of pathological tumor size (B20/ [20 mm), lymphovascular invasion (no/yes), extracapsular extension (no/yes), size of metastases (B2/[2 mm), and number of positive SLNs (1/[1), was subsequently validated on an external population (n = 180).
Results The area under the receiver operating characteristic curve was 0.78 (95 % CI 0.71–0.85) in the original population and 0.78 (95 % CI 0.70–0.85) in the validation population. Based on the predicted risk for positive NSNs, three groups were defined: low risk (B20 %), intermediate risk (21–50 %), and high risk ([50 %). In total, 88 patients met the Z0011 inclusion criteria and none of them had a high predicted risk. Of the 199 non-Z0011 patients, 67 (33.7 %) had low risk, 96 (48.2 %) had intermediate risk, and 36 (18.1 %) had high risk. Conclusion A simple risk score, integrating just five clinicopathological variables, was developed that may assist in individual decision making regarding ALND in SLN-positive patients outside of the Z0011 trial.
Introduction R. F. D. van la Parra (&) Department of Surgery, The Netherlands Cancer Institute-Antoni van Leeuwenhoek Hospital, Plesmanlaan 121, 1066 CX Amsterdam, The Netherlands e-mail: [email protected] P. G. M. Peer Department for Health Evidence, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands W. K. de Roos Department of Surgery, Gelderse Vallei Hospital, Ede, The Netherlands M. F. Ernst K. Bosscha Department of Surgery, Jeroen Bosch Hospital, ’s-Hertogenbosch, The Netherlands J. H. W. de Wilt Department of Surgery, Radboud University Nijmegen Medical Center, Nijmegen, The Netherlands
123
Over the past 20 years, axillary management in breast cancer has evolved from routine axillary lymph node dissection (ALND) to sentinel lymph node (SLN) biopsy, which has been proven to reliably identify nodal involvement in early-stage breast cancer. Although ALND can be safely avoided in SLN-negative patients, it has generally been recommended in patients with positive SLNs. However, a more conservative approach of the axilla has recently been proposed since only 40–60 % of patients are found to have additional non-sentinel node (NSN) metastases [1]. A considerable number of patients are thus exposed to the morbidity of ALND without a potential therapeutic benefit. The recently published results of the American College of Surgeons Oncology Group (ACOSOG) Z0011 trial, which compared standard ALND with no further specific axillary treatment in SLN-positive patients, demonstrated that the axillary recurrence rate was
World J Surg (2014) 38:1070–1076
low for both the SLN-only group and the ALND group [2]. Moreover, long-term results also demonstrated that overall survival was not compromised in the different treatment arms [3]. However, the trial results did only apply to patients with favorable tumor characteristics, e.g. patients with a T1 or T2 tumor treated with breast conservative surgery (including tangential field irradiation) with less than three involved SLNs and no extracapsular extension (ECE). Since only a few breast cancer patients meet the Z0011 criteria as demonstrated by Gu¨th et al. [4], a risk score could still be valuable in decision making regarding completion ALND (cALND) in SLN-positive patients outside the Z0011 trial inclusion criteria. Several predictive tools have been created over the past years to identify patients with the lowest risk of NSN metastases [5–15]. The Memorial Sloan Kettering Cancer Center (MSKCC) nomogram has been the most utilized [5]. Unfortunately, underestimation of especially the lowrisk groups has been reported in some series validating this nomogram. In a recent meta-analysis of our group, eight clinicopathological variables were identified, each of which elevates the likelihood of NSN metastases at a relevant (odds ratio C2.0) and statistically significant level [1]. Our goal was to create a new, simple model based on these clinically most relevant variables to predict the likelihood of NSN metastases. This nomogram was then studied for patients who met the Z0011 criteria and for those who did not.
1071
Patients who received neoadjuvant therapy were excluded. All patients routinely underwent preoperative axillary ultrasonography. If nodal metastases were documented preoperatively by fine needle aspiration cytology, patients underwent axillary dissection without SLN biopsy. These patients were also excluded from the study. Predictive variables Our previous meta-analyses identified eight clinicopathological variables that corresponded with the highest likelihood of NSN metastases: method of detection of SLN metastases (IHC only vs. other), size of the SLN metastases (B2/[2 mm), number of negative SLNs (B1/[1), number of positive SLNs (1/[1), ratio of positive SLNs (=number of positive SLNs divided by total number of SLNs, B50/ [50 %), extracapsular tumor extension, ECE (no/yes), tumor size (B20/[20 mm), and lymphovascular invasion, LVI (no/yes; [1]). In a previous study, we demonstrated that the method of detection relates to the size of metastases [16]. Basically, macrometastases are detected by standard haematoxylin and eosin (HE), micrometastases are detected by serial HE, and isolated tumor cells are detected by immunohistochemistry. Since the size of the metastases was known in both populations, we decided to exclude the variable ‘method of detection’ from our model, leaving seven variables. SLN mapping
Materials and methods Populations Two independent patient populations (from different teaching hospitals, where the first author worked as a surgical registrar) were used: the JBH (Jeroen Bosch Hospital) population was used for development of the model (development group) and included patients with SLN metastases who underwent ALND between January 2000 and December 2007. These were filtered from the original cohort of 913 breast cancer patients who underwent SLN biopsy. Of these, 261 had a positive SLN and 182 underwent cALND. A level I–II axillary dissection was routinely performed in case of a positive SLN. To demonstrate reliability, the model was validated on the GVH (Gelderse Vallei Hospital) population (validation group), which included patients with SLN metastases who underwent ALND between January 2004 and December 2010. These were filtered from the validation cohort of 941 breast cancer patients who underwent SLN biopsy. Of these, 260 had a positive SLN and 180 underwent cALND.
In the development group, lymphatic mapping with SLN biopsy was performed using both blue dye and radioisotope, as described previously [16]. In the validation group, lymphatic mapping was performed using radioisotope with or without blue dye at the discretion of the surgeon. All blue and/or radioactive nodes counting 10-fold ex vivo relative to the background were regarded as SLNs. Histopathological evaluation Frozen section analyses were only performed in the GVH group. SLNs with a diameter of less than 5 mm were frozen intact, whereas SLNs with a diameter of more than 5 mm were bisected longitudinally and frozen. Frozen sections of the SLN were taken from the optimal crosssectional surface with a microtome setting of 4 lm. Definitive pathological analysis in both hospitals consisted of serially sectioning the SLN at four levels of 250-lm intervals along the longitudinal axis for permanent section. From every section, two parallel slides of 4 lm were made. All slides were stained with HE and with anticytokeratin 8. The nodes removed by cALND were stained with HE.
123
1072
World J Surg (2014) 38:1070–1076
Statistical analysis Patient and tumor data were collected from the databases for each variable. The criteria for accurate model development, as stated by Coutant et al. [17] were applied. To deal with the problem of missing values in the predictors (LVI and ECE), multiple imputed datasets were constructed. The predictive distribution of a dichotomous factor was obtained by logistic regression on the observed data. A random drawing from this predictive distribution was imputed for the missing value. When each missing value was imputed, a complete dataset was obtained. This process was repeated 20 times. The relation between the predictors and the presence or absence of NSN metastases was assessed by the pooled result of the multivariable logistic regression analyses on each imputed JBH dataset. The ability of the pooled model to discriminate between patients with and without NSN metastases in the GVH dataset was quantified by the pooled area under the receiver operating characteristic Table 1 Clinicopathological variables and their association with the incidence of additional NSN metastases in the development and the validation group
curve (AUC). GVH patients were grouped into intervals according to their pooled predicted probability of NSN metastasis (B0.05, 0.05 to B0.10, 0.10 to B0.20, etc.). A calibration plot was drawn showing the actual (women who actually had positive NSNs) versus the mean of the pooled predicted percentage for each interval. To construct an easily applicable risk score, we transformed the regression coefficients of the predictors in the pooled model to integers according to their relative contribution to the risk estimation. Statistical analyses were conducted with SPSS 16.0.
Results Patient characteristics for both populations are presented in Table 1. There was a comparable distribution for tumor size, the number of positive SLNs, and metastasis size between both populations. LVI was more prevalent in the JBH population (37.2 vs. 17.7 %), however, positive SLN
Jeroen Bosch Hospital
Gelderse Vallei Hospital
Development group (n = 182)
Validation group (n = 180)
Positive NSN
Positive NSN
Tumor size (mm) B20
114 (62.6)
29 (25.4)
114 (63.3)
28 (24.6)
[20
68 (37.4)
28 (41.2)
66 (36.7)
23 (34.8)
Absent
108 (62.8)
33 (30.6)
121 (82.3)
37 (30.6)
Present
64 (37.2)
23 (35.9)
26 (17.7)
11 (42.3)
Missing
10
LVI
No. positive SLNs 1
33
149 (81.9)
42 (28.2)
152 (84.4)
38 (25.0)
33 (18.1)
15 (45.5)
28 (15.6)
13 (46.4)
B1
163 (89.6)
54 (33.1)
148 (82.7)
39 (26.4)
[1
19 (10.4)
3 (15.8)
31 (17.3)
11 (35.5)
[1 No. negative SLNs
Missing
1
Size of metastasis (mm) B2
55 (30.9)
4 (7.3)
[2
123 (69.1)
53 (43.1)
Missing
4
59 (33.3)
5 (8.5)
118 (66.7)
46 (39.0)
3
ECE No
123 (70.7)
28 (22.8)
86 (74.8)
22 (25.6)
Yes
51 (29.3)
29 (56.9)
29 (25.2)
22 (75.9)
All data are presented as n (%)
Missing
ECE extracapsular extension, LVI lymphovascular invasion, NSN non-sentinel nodes, SLNs sentinel lymph nodes
Z0011 Non-Z0011
47 (27.3) 125 (72.7)
Uncategorizeda
8
a
Due to missing ECE status
123
Total
8
65 13 (27.7) 44 (35.2)
41 (35.7) 74 (64.3)
9 (22.0) 35 (47.3)
65 57 (31.3)
51 (28.3)
World J Surg (2014) 38:1070–1076
1073
rate was almost equal in both centers (35.9 and 42.3 %) for patients with LVI. Almost 70 % of our population were patients who did not meet the Z0011 inclusion criteria due to irradical excision (n = 16), multicentricity (n = 34), presence of ECE (n = 80), more than three SLNs removed (n = 1), mastectomy performed (n = 157), and/or tumor size over 50 mm (n = 3). Additional NSN positivity was lower in patients who met the Z0011 criteria (25 vs. 40 % of the non-Z0011 patients). After cALND, a median total of 12 nodes (range 4–23) were harvested in the JBH group, and 15 nodes (range 6–37) in the GVH group. Initially, three predictive models were developed that differed in the way the variables positive and/or negative SLNs were incorporated (e.g. model I only SLN positive, model II SLN positive and negative, model III ratio of positive SLNs). Model I included the following dichotomous variables: tumor size, LVI, number of positive SLNs, metastasis size, and ECE. All three models had an AUC of [0.7 on external validation, but model I performed the best on calibration of the low predicted risks (B0.05 or B0.10), because it corresponded best to the reference line. Models II and III, which included the number of SLN-negative patients performed worse on predicting low probabilities. Based on this performance, model I became the final model. Using the formula of the logistic regression model, a patient’s probability of NSN metastases based on her clinicopathological profile was established. To facilitate use in daily practice, the final model was simplified to an applicable scoring rule. Regression coefficients of the
Fig. 1 Graphical representation of total scores and corresponding predicted risks for positive non-sentinel nodes
variables in the final model were rounded to integers according to their relative contribution to the risk estimate (Table 2). As such, five points were given for the presence of LVI, six points for a tumor size [2 cm, 7 points for the presence of more than one positive SLN, 14 points for the presence of ECE, and 20 points for a metastases size [2 mm. Subsequently, a sum score is obtained for each patient. The total score ranges from 0 to 52. Each score corresponds with a predicted percentage of NSN metastases, which can be extracted from Fig. 1. The final model was validated externally on the GVH population demonstrating an AUC of 0.78 (95 % CI
Table 2 Risk score to predict the likelihood of NSN metastases in patients with a positive SLN Variables
Coefficients
Point values
LVI Absent Present
0 0.432
5
0.527
6
Tumor size (cm) B2 [2
0
No. positive SLNs B1 [1
0 0.622
7
1.247
14
1.716
20
ECE No Yes
0
Size of metastasis (mm) B2 [2
0
Model formula P ¼ eð3:072þ0:432LVIþ0:527tumour sizeþ0:622no1 positive SNsþ1:247ECEþ1:716size of metastasesÞ ECE extracapsular extension, LVI lymphovascular invasion, NSN non-sentinel node, SLN sentinel lymph nodes
123
1074
World J Surg (2014) 38:1070–1076
0.70–0.85). Patients were classified into intervals of the predicted risks, and actual probabilities were plotted against the predicted probability of NSN metastases (Fig. 2). The false-negative rate was 3.3 % in the group with a predicted probability of NSN metastases of \5 %.
Three risk groups were constructed based on the predicted probabilities (Table 3): a low risk group (predicted risk B20 % for NSN metastases), intermediate risk (21–50 %), and high risk ([50 %). In the low-risk group, 4 of 58 patients (6.9 %) had NSN metastases. For the intermediate- and high-risk groups this was 27 of 96 patients (28.1 %) and 20 of 26 (76.9 %), respectively. Of the 287 breast cancer patients with a known ECE status from both populations, 88 patients (30.7 %) met the Z0011 inclusion criteria, and of these, 39 (44.3 %) were classified as low risk and 49 (55.7 %) as intermediate risk. Of the 199 patients who did not meet the Z0011 inclusion criteria, 67 (33.7 %) had a low risk, 96 (48.2 %) an intermediate risk, and 36 (18.1 %) a high risk of NSN metastases (Table 4).
Discussion
Fig. 2 Calibration plot for risk score to predict positive non-sentinel nodes. The entire cohort was separated into nine groups of varying sizes according to their predicted percentages (1–10, 11–20 %, etc.). For each group, the actual percentage was calculated. A calibration plot was drawn showing for each interval the actual versus predicted percentage. If the model was perfect, all dots would lie on the line, with a slope of 1
Table 3 Actual percentage for NSN metastases in the validation group Risks
Intervals
Total
Positive NSNs, n (%) No
Yes (=actual)
Low
0 to 0.20
58
54 (93.1)
4 (6.9)
Intermediate
0.21 to B0.50
96
69 (71.9)
27 (28.1)
High
0.51 to B0.80
26
6 (23.1)
20 (76.9)
180
129 (71.7)
51 (28.3)
Total NSN non-sentinel node
Table 4 Total number of patients with or without Z0011 criteria Risks
Intervals
Total
Z0011, n (%)
Non-Z0011, n (%)
Low
0 to 0.20
106
39 (44.3)
67 (33.7)
Intermediate
0.21 to B0.50
145
49 (55.7)
96 (48.2)
High
0.51 to B0.80
36
–
36 (18.1)
88*
199*
Total
287
* Numbers calculated on patients with a known extracapsular extension status
123
A simple risk score, integrating just five clinicopathological variables, was developed to assist in individual decision making regarding ALND in SLN-positive patients outside of the Z0011 trial. With this risk score, three groups were established with low, intermediate, or high risk on additional NSNs in the ALND. In total, 88 patients met the Z0011 inclusion criteria and none of these patients had a high predicted risk. Of the 199 patients who did not meet the Z0011 inclusion criteria, 67 (33.7 %) had a low risk, 96 (48.2 %) an intermediate risk, and 36 (18.1 %) a high risk of NSN metastases. The risk of additional NSN metastases is \7 % in the low-risk group, 28.1 % for the intermediate-risk group, and 76.9 % for the high-risk group. External validation of the risk score demonstrated considerable discrimination (AUC 0.78). The false-negative rate was 3.3 % in the group with a predicted probability of NSN metastases of \5 %. ALN status is traditionally regarded as an important prognosticator in breast cancer. ALND is considered the standard of care when the SLN is positive, because it could give additional information on the total number of involved nodes and it was assumed to reduce the local recurrence rate. In our study groups, only 57/182 (31.3 %) and 51/180 (28.3 %) patients with SLN metastases had positive NSNs, which illustrates overtreatment of the axilla in the majority of patients and predisposes to associated morbidity such as lymphedema and axillary numbness. Several predictive models have been developed over past years in order to try to identify patients at low risk of NSN metastases [6–15]. These models identified different parameters as significant predictors of NSN metastases. Several parameters have been more commonly included in
World J Surg (2014) 38:1070–1076
models and were also identified as independent predictors in our previous meta-analysis [1]. The predictive models for NSN metastases developed thus far, including the MSKCC model, were derived from multivariable analyses of their original populations, which varied in size from 71 to 791 patients, with a variable number of predictors (between 3 and 7) [6–15]. External validation of our risk score has been performed; this was only performed in some of the other published models [5, 8, 12–14]. The MSKCC nomogram does not contain a cut-off score that mandates the performance of an ALND. Also, the MSKCC nomogram tends to underestimate the probability of NSN metastases in the lower range. The present risk score is based on the clinically most significant predictors of NSN metastases as identified in a previous meta-analysis [1]. Since our risk score is only based on five dichotomous variables, the score is easy to calculate and the associated risk can be extracted from the figure. The risk score is unique in that it provides cut-off scores, below which cALND can be safely avoided and above which cALND should be performed. Three levels of risk were identified with an associated axillary treatment recommendation based on the estimates for NSN involvement of 6.9, 28.1, and 76.9 %, respectively. We propose to avoid ALND in the low-risk group since the associated risk of additional NSN metastases is only 6.9 %. ALND should be discussed in the intermediate-risk group (associated risk 28.1 %). We strongly recommend performing ALND in the high-risk group (associated risk 76.9 %) until other studies have shown that ALND can be safely avoided. The high rate of negative NSNs in ALND and the low axillary recurrence rates reported in recent trials [18] have led to a shift towards a more conservative approach of the axilla with a declining role of ALND. In the ACOSOG Z0011 study, SLN-positive patients with early-stage breast cancer were randomized to ALND or not. At a median follow-up of 6.3 years, there were no statistically significant differences in local recurrence (1.8 % in the SLN-only group, n = 446 vs. 3.6 % in the ALND arm, n = 445; [2]). Furthermore, 5-year overall survival was 91.9 versus 92.5 % for the SLN-only group, and disease-free survival was 82.2 versus 83.8 % for the SLN-only group [3]. Although results are impressive, there are some aspects of the Z0011 study that limit its general applicability. The selection of a low-risk population and the effect of adjuvant systemic therapy on residual microscopic disease may be reflected in this low recurrence rate. Second, all patients received tangential field irradiation of the breast, which covers at least part of the axilla. Therefore, all patients received some kind of axillary treatment, which may in part account for the low rates of axillary recurrence.
1075
In light of the Z0011 trial, some have questioned the need for risk scores/nomograms in clinical decision making. However, the Z0011 results only apply to low-risk patients treated with breast-conserving surgery. Gu¨th et al. [4] tested the applicability of the Z0011 criteria on their patient population, which comprised 389 consecutive breast cancer patients who were treated with SLN biopsy. Only 9 % (35 patients) of their population met the Z0011 inclusion criteria, which demonstrates that the Z0011 results only apply to a minority of all patients who underwent SLN biopsy for breast cancer. In the present study, only 88 patients (30.7 %) met the Z0011 criteria (Table 4). This percentage is higher than that reported in the study by Gu¨th et al. [4], since we only reported the number of SLN-positive patients, while Gu¨th et al. reported on all SLN-positive and -negative patients. So, if ALND is to be avoided in the Z0011 patients, this still leaves 69.3 % of SLN-positive patients outside of the Z0011 inclusion criteria, for whom we have to decide whether or not ALND should be performed. Recently, Milgrom et al. [19] showed a similar outcome in SLN-positive patients with early-stage breast cancer who underwent total mastectomy without cALND or radiotherapy. However, of the 210 mastectomy patients with a positive SLN, there were only 19 patients (9.1 %) with N1 (macrometastatic) disease [113 patients (54 %) had N0(i?) disease, and 78 patients (37 %) N1mic]. The numbers are too small to draw firm conclusions and to avoid ALND in SLN-positive mastectomy patients. In daily practice, the score described in the present study can be used for an individualized risk estimate in patients with SLN metastases. Since predictive scores provide a percentage as outcome, a low-risk group was identified with an associated risk for NSN metastases of \7 %. Since the risk score gives a risk estimate, other factors are also involved in our decision making, e.g. patient age, comorbidity, patient preference, and surgeon’s experience. Some limitations apply to this study. The risk score was developed and validated in a relatively small patient population. However, some previous nomograms were developed in even smaller populations [6, 7, 9, 11]. To prove its strength, it needs to be validated prospectively on larger (external) populations. Due to the retrospective nature of the study, we had some missing variables. However, to deal with the problem of missing values, we constructed multiple imputed datasets. Furthermore, preoperative axillary ultrasound was routinely performed in both populations. Preoperative axillary ultrasound selects patients with a greater nodal burden, so it is likely that its routine use would reduce overall axillary disease burden in those undergoing SLN biopsy. Since it was applied in both the development and the validation population this could explain the good results on external validation. Despite
123
1076
World J Surg (2014) 38:1070–1076
differences in the SLN mapping technique and pathological analyses (frozen section in GVH) between both hospitals, validation remained good. This demonstrates the strength of the model when applied to a dataset outside the institution of origin.
Conclusion We succeeded in developing a simple risk score, integrating only five clinicopathological variables to provide an individualized risk estimate of the likelihood for NSN metastases when the SLN is positive. Three levels of risk were identified, with an associated axillary treatment recommendation based on the estimates for NSN involvement of 6.9, 28.1, and 76.9 %, respectively. This tool may assist in individual decision making regarding ALND in SLNpositive patients, especially in those who do not meet the Z0011 trial inclusion criteria. Acknowledgments The authors thank W. Lemmens for performing the data analysis. No grant support was received for this work. Conflict of Interest
The authors have no conflicts of interest.
References 1. van la Parra RF, Peer PG, Ernst MF et al (2011) Meta-analysis of predictive factors for non-sentinel lymph node metastases in breast cancer patients with a positive SLN. Eur J Surg Oncol 37:290–299 2. Giuliano AE, McCall L, Beitsch P et al (2010) Locoregional recurrence after sentinel lymph node dissection with or without axillary dissection in patients with sentinel lymph node metastases: the American College of Surgeons Oncology Group Z0011 randomized trial. Ann Surg 252:426–432; discussion 32–33 3. Giuliano AE, Hunt KK, Ballman KV et al (2011) Axillary dissection vs no axillary dissection in women with invasive breast cancer and sentinel node metastasis: a randomized clinical trial. JAMA 305:569–575 4. Gu¨th U, Myrick ME, Viehl CT et al (2012) The post ACOSOG Z0011 era: does our new understanding of breast cancer really change clinical practice? Eur J Surg Oncol 38:645–650 5. Van Zee KJ, Manasseh DM, Bevilacqua JL et al (2003) A nomogram for predicting the likelihood of additional nodal metastases in breast cancer patients with a positive sentinel node biopsy. Ann Surg Oncol 10:1140–1151
123
6. Hwang RF, Krishnamurthy S, Hunt KK et al (2003) Clinicopathologic factors predicting involvement of nonsentinel axillary nodes in women with breast cancer. Ann Surg Oncol 10:248–254 7. Saidi RF, Dudrick PS, Remine SG, Mittal VK (2004) Nonsentinel lymph node status after positive sentinel lymph node biopsy in early breast cancer. Am Surg 70:101–105 8. Degnim AC, Reynolds C, Pantvaidya G et al (2005) Nonsentinel node metastasis in breast cancer patients: assessment of an existing and a new predictive nomogram. Am J Surg 190:543– 550 9. Barranger E, Coutant C, Flahault A et al (2005) An axilla scoring system to predict non-sentinel lymph node status in breast cancer patients with sentinel lymph node involvement. Breast Cancer Res Treat 91:113–119 10. Chagpar AB, Scoggins CR, Martin RC II et al (2006) Prediction of sentinel lymph node-only disease in women with invasive breast cancer. Am J Surg 192:882–887 11. Pal A, Provenzano E, Duffy SW et al (2008) A model for predicting non-sentinel lymph node metastatic disease when the sentinel lymph node is positive. Br J Surg 95:302–309 12. Kohrt HE, Olshen RA, Bermas HR et al (2008) New models and online calculator for predicting non-sentinel lymph node status in sentinel lymph node positive breast cancer patients. BMC Cancer 8:66 13. Cho J, Han W, Lee JW et al (2008) A scoring system to predict nonsentinel lymph node status in breast cancer patients with metastatic sentinel lymph nodes: a comparison with other scoring systems. Ann Surg Oncol 15:2278–2286 14. Coufal O, Pavlik T, Fabian P et al (2009) Predicting non-sentinel lymph node status after positive sentinel biopsy in breast cancer: what model performs the best in a Czech population? Pathol Oncol Res 15:733–740 15. Perhavec A, Perme MP, Hocevar M et al (2010) Ljubljana nomograms for predicting the likelihood of non-sentinel lymph node metastases in breast cancer patients with a positive sentinel lymph node. Breast Cancer Res Treat 119:357–366 16. van la Parra RF, Ernst MF, Bevilacqua JL et al (2009) Validation of a nomogram to predict the risk of nonsentinel lymph node metastases in breast cancer patients with a positive sentinel node biopsy: validation of the MSKCC breast nomogram. Ann Surg Oncol 16:1128–1135 17. Coutant C, Olivier C, Lambaudie E et al (2009) Comparison of models to predict nonsentinel lymph node status in breast cancer patients with metastatic sentinel lymph nodes: a prospective multicenter study. J Clin Oncol 27:2800–2808 18. Francissen CM, Dings PJ, van Dalen T et al (2012) Axillary recurrence after a tumor-positive sentinel lymph node biopsy without axillary treatment: a review of the literature. Ann Surg Oncol 19:4140–4149 19. Milgrom S, Cody H, Tan L et al (2012) Characteristics and outcomes of sentinel node-positive breast cancer patients after total mastectomy without axillary-specific treatment. Ann Surg Oncol 19:3762–3770
