Validation of Three Breast Cancer Nomograms and a New Formula for Predicting Non-sentinel Lymph Node Status

Breast cancer, which accounts for 28% of female cancers and approximately 14% of cancer-related deaths, is the most common cancer type in females. According to the statistics from 2004-2008, a woman’s risk of the lifetime probability of being diagnosed with an invasive breast cancer has increased from 1/13-1/8 (Jemal et al., 2010). Axillary lymph node involvement is the most important prognostic factor in patients with breast cancer. In the case of axillary lymph node involvement, not only the treatment modality will be changed, but also the life expectancy will decrease (Chen et al., 2010). Nowadays, sentinel lymph node (SLN) biopsy is considered as the standard approach in patients without clinical diagnosis of axillary metastasis. Complementary axillary dissection is performed as a continuation of standard therapy in patients


Introduction
Breast cancer, which accounts for 28% of female cancers and approximately 14% of cancer-related deaths, is the most common cancer type in females.According to the statistics from 2004-2008, a woman's risk of the lifetime probability of being diagnosed with an invasive breast cancer has increased from 1/13-1/8 (Jemal et al., 2010).
Axillary lymph node involvement is the most important prognostic factor in patients with breast cancer.In the case of axillary lymph node involvement, not only the treatment modality will be changed, but also the life expectancy will decrease (Chen et al., 2010).Nowadays, sentinel lymph node (SLN) biopsy is considered as the standard approach in patients without clinical diagnosis of axillary metastasis.Complementary axillary dissection is performed as a continuation of standard therapy in patients RESEARCH ARTICLE Validation of Three Breast Cancer Nomograms and a New Formula for Predicting Non-sentinel Lymph Node Status Serhan Derici 1 *, Ali Sevinc 2 , Omer Harmancioglu 2 , Serdar Saydam 2 , Mehmet Kocdor 2 , Suleyman Aksoy 3 , Tufan Egeli 2 , Tulay Canda 2 , Hulya Ellidokuz 2 , Solen Derici 1 with sentinel lymph node (SLN) metastasis.However, studies have revealed that 40-70% of the patients with SLN metastasis undergoing complementary axillary dissection do not have an additional non-SLN metastasis (Van Zee et al., 2003;Chen et al., 2010) After the above-mentioned determinations, studies have been conducted in many centers in order to create models that are capable of predicting the risk of non-SLN metastasis in patients with SLN metastasis.Van Zee et al. (2003) developed a nomogram based on the records of the Memorial Sloan-Kettering Cancer Center (MSKCC) using a retrospective analysis method.They also tested the nomogram prospectively, and found the area under the receiver operating characteristic (ROC) curve to be 0.77 (Van Zee et al., 2003).Validation studies were performed in many centers after the nomogram was published.However, obtaining different results in different patient populations have led the researchers in other centers to seek new nomograms.Barranger et al. (2005) retrospectively reviewed the data from the Hospital Tenon in Paris city of France and developed a new nomogram (Barranger et al., 2005).In England, Pal et al. (2008) applied the MSKCC nomogram to their patients in 2008 and found the AUC of the ROC curve to be 0.68.Based on this result, they found the MSKCC nomogram not suitable for their patients and published the Cambridge nomogram (Pal et al., 2008).Kohrt et al. (2008) from the Stanford University School of Medicine evaluated the MSKCC nomogram using their own data and reported the AUC of the ROC curve to be 0.62.Depending on this finding, concluded that the MSKCC nomogram was not suitable for their patients, and they published the Stanford nomogram with an AUC of the ROC curve of 0.74 (Kohrt et al., 2008).
In our clinic, 41% of the patients with SLN metastasis, which was detected on SLN biopsy between July 2008 and August 2010, did not have non-SLN metastasis.The present study aimed to evaluate the efficacy of mostly accepted three nomograms, the MSKCC, Tenon and Stanford nomograms, in our patient population.

Materials and Methods
Files of the patients who underwent surgical procedure with the diagnosis of primary breast cancer in the Dokuz Eylul University Hospital between July 2008 and August 2010 were retrospectively reviewed.Among 289 patients who underwent successful SLN biopsy, 170 patients with positive SLN biopsy were included in the study.Patients who were thought to have clinical axillary lymph node metastasis, those with distant metastasis, and those who received neoadjuvant therapy or have >T3 tumor were excluded from the study.
During SLN evaluation; formalin-fixed paraffin embedded tissue sections of all the sentinel and nonsentinel lymph nodes were examined after frozen section diagnosis.Four deeper levels of 4 µm sections from formalin-fixed parafin-embedded tissue blocks of lymph nodes with 25 µm intervals each, were performed.The first, third, and fourth sections were examined with hematoxylin-eosin (H&E) staining, whereas the second section was reserved for IHC.IHC of the lymph node sections were carried out by streptavidin biotin peroxidase method using AE1/AE3 (1:100 dilution, MS-343, Neomarker, USA), which is a monoclonal antibody cocktail reactive with cytokeratins.
Diagnosis was established by frozen imprint in 165 patients, whereas it was established by further analyses [hematoxylin and eosin (H&E) staining, serial section examination, immunohistochemistry (IHC) assay] in 5 patients.
The age of all patients was recorded.The following tumor characteristics were recorded: location, type, size, grade, multifocality, the Scarff-Bloom-Richardson (SBR) grade, nuclear grade, mitotic grade, histologic grade, ratio and type of intraductal component, lymphocytic infiltration status, presence of elastosis, presence of invasion to lymph vessel, blood vessel, skin or fascia, estrogen receptor (ER) and progesterone receptor (PR) status, and ER and PR staining intensities.The method used in the detection of SLN, and the number of positive and negative SLNs were determined; proportion of SLN was calculated.Capsular invasion status and size of metastasis were recorded for SLN metastasis.

Nomograms
The MSKCC nomogram, which was published in 2003 by Van Zee et al. (2003), is the first known nomogram.This nomogram includes the following nine variables: presence or absence of frozen examination, tumor diameter, histologic and nuclear grade of tumor, lymphovascular invasion (LVI), multifocality, ER status, method of detection of SLN, number of positive SLNs, and number of negative SLNs.In that model, 702 patients with SLN metastasis were evaluated retrospectively, and the relation of each variable with SLN metastasis was assessed by a multivariate logistic regression analysis.A calculation chart showing the score of each variable was created (Figure 1).The authors then validated the method by prospectively applying this calculation chart to another group comprising 373 patients.A ROC curve was drawn for this method, and the AUC was reported to be 0.77 (Van Zee et al., 2003;www.mskcc.org/applications/nomograms).
A new scoring system, which was developed in Tenon Hospital in France by Barranger et al. (2005), was published in 2005.They found that non-SLN metastasis was correlated with the tumor size, presence of macrometastasis, method of detection of SLN metastasis, number of positive SLN, proportion of positive SLN, LVI status, and size of SLN metastasis.Based on these data, "the Tenon axilla scoring system" was developed.This scoring system is based on the following three variables: 1) ratio of number of positive SLN to the total number of dissected SLN, 2) presence of macrometastasis, 3) histological tumor size.
The likelihood of non-SNL metastasis in breast cancer patients with SNL involvement is calculated obtaining a score between 0 and 7 with these variables (Table 1).A score of <3.5 indicates the absence of non-SLN metastasis with a probability of 97.3% (Barranger et al., 2005).
In  of detection of SLN metastasis) were found significant in univariate analysis.Based on multivariate analyses, they concluded that only three of these parameters (tumor size, presence of LVI, and size of SLN metastasis) could be used to predict non-SLN metastasis (www3-hrpdcc.stanford.edu/nsln-calculator).
The study patients were divided into two groups as those without non-SLN metastasis and those with non-SLN metastasis.Calculations were performed for each patient according to the MSKCC, Tenon and Stanford nomograms.
All statistics including the ROC curve analysis were performed using the Statistical Package for the Social Sciences (SPSS, Inc., Chicago, IL, USA) version 15.0.The efficacies of MSKCC, Tenon and Stanford nomograms were evaluated using ROC curves.A ROC curve was drawn, and an AUC was calculated for each nomogram (an AUC of the ROC curve between 0.7 and 0.8 was considered good, an AUC of the ROC curve between 0.8 and 0.9 indicated that the nomogram had excellent discrimination, and an AUC of the ROC curve of 0.5 indicated that there is no discrimination).

Statistical analysis
In univariate analysis, the Student's t-test, one-way

Results
Calculation was performed for each of the three nomograms, and an AUC of the ROC curve was calculated to evaluate the efficacy of each nomogram (Figure 2).The AUC values of the ROC curves were found to be 0.713, 0.534, and 0.671 for the MSKCC, Stanford, and Tenon nomograms, respectively.
Results of the univariate analyses are presented in Table 2. Univariate analysis revealed significant differences between the patients with and without non-SLN metastasis in the following parameters: multifocality (p=0.001),PR status (p=0.05),PR staining intensity (p=0.022),number of negative SLN (p=0.001),proportion of metastatic SLN (number of metastatic SLN/total number of SLN; p=0.001), size of SLN metastasis (p=0046), SLN capsular invasion (p=0.001).A multivariate analysis was performed for the above-mentioned significant variables and the variables used in all of the three nomograms.The logistic regression analysis yielded Exp.(B) values for multifocality, LVI, number of positive SLN, and SLN capsular invasion (Table 3).

Discussion
Breast cancer is the most common cancer in women.Recently, conservative methods are preferred for the treatment of breast cancer to reduce the potential morbidity.SLN biopsy has been accepted as the standard therapy approach in many centers.With the use of SLN biopsy, it is possible to prevent lymphedema, nerve injury, limitation of shoulder motion, and chronic pain, which are likely to occur due to unnecessary axillary dissection (Latosinsky et al., 2008).However, review of a 15-year data has revealed that 40-70% of the patients with metastasis detected on SLN biopsy have no additional axillary lymph node metastasis (Chen et al., 2010).
Retrospectively evaluating the data of Memorial Sloan-Kettering Cancer Center, Van Zee et al. (2003) developed a nomogram (the MSKCC nomogram) in 2003, and then prospectively validated this nomogram.They found the AUC value to be 0.77.In the present study, we found the AUC value to be 0.71 for the MSKCC nomogram.
In their study, Kohrt et al. (2008) validated the MSKCC nomogram and found the AUC value to be 0.62.They developed the Stanford nomogram and reported the AUC value for this nomogram to be 0.74 (Kohrt et al., 2008).In our study, we found the AUC value for the Stanford nomogram to be 0.53.Barranger et al. (2005) defined the Tenon nomogram in 2008 based on the data from Hospital Tenon in Paris.
In the present study, we found the AUC value to be 0.67 for the Tenon nomogram.
In 2008, Pal et al. (2008) from England validated the MSKCC nomogram using their own data and found the AUC value as 0.68.They found the AUC value as 0.84 for the Cambridge nomogram, which they developed based on their own data.
In the first validation study performed in Turkey for the MSKCC, Cambridge, Stanford, and Tenon nomograms, the predictability of these nomograms were evaluated (Gur et al., 2010).In that particular study, the AUC values were determined to be 0.70, 0.73, 0.71, and 0.58 for the MSKCC, Stanford, Cambridge, and Tenon nomograms.They found the AUC value to be 0.80 for the nomogram (MF08-01) that they developed based on their own data (Gur et al., 2010).
The results of the validation studies performed for the MSKCC nomogram, which was the first nomogram, are summarized in Table 4 (Unal et al., 2008).
Tumor size is the common variable used in the MSKCC, Tenon and Stanford nomograms, whereas LVI and tumor grade are the common variables used in the MSKCC and Stanford nomograms.While the variables including number of positive SLN, number of negative SLN, method of detection of SLN metastasis, ER status, and multifocality are used only in the MSKCC nomogram, SLN micro/macrometastasis status and proportion of positive SLN are used only in the Tenon nomogram.Size of SLN metastasis is used only in the Stanford nomogram (Table 4).
In the present study, when the obtained data and the nomograms were reviewed, we observed that SLN capsular invasion, which had the second highest OR in the present series, was used in none of the three nomograms.The other intriguing finding was that three of the parameters for which OR was calculated in the present series, however, were separately evaluable in the three nomograms.This was considered as the greatest inadequacy of the validated nomograms for the present series (Table 4).
Based on the data of the patients from the Dokuz Eylül University Breast Tumor Group, a risk assessment formula was developed using these outcomes.When the formula was verified for the patients in the present study, the AUC for the ROC curve was calculated as 0.814 (Figure 3).
The new formula; p=1/ [1+exp -1x(4, 113+MFx0, 857+PrSLNx2,  686+LVIx1, 065+SLNkapINVx2, 058) ] In conclusion, there is no method other than the nomograms developed to assess the likelihood of non-SLN metastasis; however, there are limitations concerning the use of these nomograms.Primarily, the best outcomes of the nomograms are obtained based on the data of the clinic, in which the nomogram was developed.In order to overcome this limitation, each step of SLN biopsy procedure should be standardized, and the nomograms should be revised with multicenter, even with   In the present study, we evaluated the MSKCC, Tenon and Stanford nomograms.We concluded that the MSKCC nomogram is in the limits of applicability, whereas the Tenon and Stanford nomograms are not available for our patients.From this point of view, based on the data of patients of the Dokuz Eylül University Breast Tumor Group, we developed a new risk assessment system, which was intended to be verified prospectively.Thus, we aimed to contribute to the large-scale scoring systems that would be developed in the future.
Efficacy of the risk assessment formula defined in the present study should be prospectively controlled both in our clinic and in other centers.In the event of obtaining adequate evidence about applicability of this new formula, a conclusion about the risk of non-SLN metastasis could be reached together with the outcomes of another nomogram, the efficacy of which has been proven (e.g. the MSKCC nomogram).
It may be suggested that, in clinical practice, these nomograms can be used in the near future to better inform the patients about their potential risks".Nevertheless, as the information on cancer biology and behavior is enhanced and stronger outcomes for the nomograms are obtained, the rate of axilla-preserving surgical procedures may increase in the future.

Table 2 . Significant Variables in the Univariate Analysis
*PR: progesterone receptor, SLN: sentinel lymph node Figure 2. Receiver Operating Characteristic (ROC) Curve.MSKCC, Tenon and stanford nomograms Figure 3. Receiver Operating Characteristic (ROC) Curve.The area under the curve (AUC) was 0.814 using our