- Email: [email protected]
Development and Validation of a Clinical Prediction Model for N2 Lymph Node Metastasis in Non-Small Cell Lung Cancer
Development and Validation of a Clinical Prediction Model for N2 Lymph Node Metastasis in Non-Small Cell Lung Cancer Kezhong Chen, MD, Fang Yang, MD, Guanchao Jiang, MD, Jianfeng Li, MD, and Jun Wang, MD
Background. The true incidence of occult N2 lymph node metastasis in patients with clinical N0 non-small cell lung cancer (NSCLC) remains controversial. Estimation of the probability of N2 lymph node metastasis can assist physicians when making diagnosis and treatment decisions. Methods. We reviewed the medical records of 605 patients (group A) and 211 patients (group B) with computed tomography–defined N0 NSCLC that had an exact tumor-node-metastasis stage after surgery. Logistic regression analysis of group A’s clinical characteristics was used to estimate the independent predictors of N2 lymph node metastasis. A prediction model was then built and internally validated by using cross validation and externally validated in group B. The model was also compared with 2 previously described models. Results. We identified 4 independent predictors of N2 disease: a younger age; larger tumor size; central
tumor location; and adenocarcinoma or adenosquamous carcinoma pathology. The model showed good calibration (Hosmer-Lemeshow test: p [ 0.96) with an area under the receiver operating characteristic curve (AUC) of 0.756 (95% confidence interval, 0.699 to 0.813). The AUC of our model was better than those of the other models when validated with independent data. Conclusions. Our prediction model estimated the pretest probability of N2 disease in computed tomography-defined N0 NSCLC and was more accurate than the existing models. Use of our model can be of assistance when making clinical decisions about invasive or expensive mediastinal staging procedures.
A
PET/CT prohibits its routine application in many countries [6]. Several studies have therefore attempted to determine which factors are associated with occult mediastinal lymph node metastasis in patients with NSCLC [7, 8], because successful prediction could aid in selecting candidates for either preoperative cervical mediastinoscopy or endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA). However, it is difficult to select and interpret diagnostic test results correctly without estimating the patient’s pretest probability of metastasis. Shafazand and Gould [9] of the Veterans Affairs Palo Alto Health Care System identified 6 independent predictors of N2 metastasis in 566 patients with NSCLC; the area under the receiver operating characteristic (ROC) curve was 0.70. Zhang and colleagues at the Fudan Clinic [10] retrospectively reviewed the medical records of 530 patients with CT-defined T1N0 NSCLC and identified 4 independent predictors of metastasis. The accuracy of the Fudan model is similar to that of the VA model, with an area under the ROC curve (AUC) of 0.726. However, neither of the 2 models has external validation, which is warranted to investigate the generality of the prediction model before it is implemented in clinical practice. We conducted this retrospective study to assess the real risk factors of N2 disease among patients with CT-defined
ccurate staging of non-small cell lung cancer (NSCLC) is crucial for selecting the optimal therapy. Patients with tumor-node-metastasis stages IA to IIB NSCLC are widely considered to be suitable surgical candidates, while controversy remains on the role of surgery for patients with stage IIIA NSCLC (N2 disease). This controversy makes preoperative mediastinal staging particularly important for patients who are potentially suitable surgical candidates. The F-18-fluorodeoxyglucose positron emission tomography combined with thoracic computed tomography (18F-FDG PET/CT) is increasingly utilized for staging of mediastinal lymph nodes [1]. However, a few studies have raised concerns about the underestimation of N2 disease when PET/CT results are negative [2, 3]. In patients who have been clinically staged as negative for N2 disease, the incidence of occult mediastinal lymph node metastasis ranges from 5.6% to 18% based on the T stage [4]. One study demonstrated that the sensitivity of PET/CT for detecting malignant nodal involvement was only 32.4% in nodes less than 10 mm [5]. In addition, the high expense of Accepted for publication June 7, 2013. Address correspondence to Dr Wang, Department of Thoracic Surgery, Peking University People’s Hospital, No. 11 Xizhimen South St, Xicheng District, Beijing, Ot 100044, P. R. China; e-mail: [email protected].
Ó 2013 by The Society of Thoracic Surgeons Published by Elsevier Inc
(Ann Thorac Surg 2013;96:1761–8) Ó 2013 by The Society of Thoracic Surgeons
0003-4975/$36.00 http://dx.doi.org/10.1016/j.athoracsur.2013.06.038
GENERAL THORACIC
Department of Thoracic Surgery, Peking University People’s Hospital, Beijing, People’s Republic of China
1762
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
N0 NSCLC. We developed a clinically useful prediction model, which we validated both internally and externally, to estimate the pretest probability of N2 metastasis in patients with NSCLC. We also compared the accuracy of our model with that of the VA and Fudan models by calculating the AUC. It is important to note that the value of this model lies in its ability to estimate the pretest probability, which can then be used to guide the subsequent selection of diagnostic tests; the model is not intended for use as a stand-alone diagnostic method.
Ann Thorac Surg 2013;96:1761–8
University People’s Hospital. The techniques used to dissect the lymph nodes were standardized. Mediastinal lymph node dissection or systematic lymph node sampling were acceptable procedures (American College of Surgeons Oncology Group criteria) [13].
Established Mathematical Models The Veterans Affairs model: The pretest probability of N2 nodal involvement ¼ ex/(1þex). x ¼ 1:806 þ ð0:955 adenoÞ þ ð0:876
GENERAL THORACIC
Material and Methods
abnormal CXRÞ þ ð0:749 centralÞ þ 0:485 primary symptom þ ð0:435 sizeÞ ð0:408
Patient Eligibility
age > 65Þ:
Between September 2006 and August 2011, we reviewed the medical records of 781 patients with a pathologic diagnosis of NSCLC who underwent surgical resection with lymph node dissection. All of the patients underwent contrast-enhanced thoracic CT scans at the Peking University People’s Hospital before surgery. Other routine preoperative examinations included chest radiograph, cardiopulmonary function tests, abdominal and adrenal gland ultrasonography or CT, brain magnetic resonance imaging or CT, and a bone scan. A lymph node was considered to be positive if the short axis exceeded 1 cm on a thoracic CT image. One hundred fourteen patients who had been diagnosed with CT-defined N1 or N2 NSCLC before surgery were excluded. Thirty-two patients with pure bronchioalveolar carcinoma (BAC) [11] were excluded because BAC was defined as a noninvasive adenocarcinoma in accordance with the revised World Health Organization classification [12], 11 patients were excluded because of a history of lung malignancy within 5 years, and 19 patients were excluded because of incomplete data. In total, 605 patients were enrolled into group A, and their clinical data were analyzed to create the mathematical model. Clinical data were also collected from an additional 211 patients who were enrolled with the same inclusion criteria between September 2011 and December 2012 (group B).
Clinical Variables We collected clinical data of patients in Groups A and B that included the age and gender of the patient, symptoms at presentation, smoking history, average number of cigarettes per year, family history of cancer, tumor histology and location, and tumor size measured by CT. We also analyzed the chest radiographs (CXR) of patients in group B to determine the appearance of the mediastinum. Tumors were considered to be centrally located if the center of the tumor was in the medial third of the lung parenchyma (adjacent to the mediastinum) and peripherally located if the center of the tumor was within the lateral two-thirds of the lung parenchyma.
The Fudan model. The pretest probability of N2 nodal involvement ¼ ex/(1þex). x ¼ 3:449 þ ð1:018 diameterÞ þ ð1:164 locationÞ þ ð1:263 histologyÞ ð0:026 ageÞ:
Statistical Analysis Univariate analysis of the correlation between mediastinal lymph node metastasis and the categoric variables was conducted by using the Fisher exact test or Pearson c2 test. An independent sample t test was used to assess the association between N2 disease and continuous variables, which are expressed as the mean values and standard deviations. Variables with a p value less than 0.1 were subjected to a binary logistic regression analysis. We developed the model by using stepwise logistic regression, with factors that were statistically significant at the 0.05 level remaining in the final model as the dependent variables. The Hosmer-Lemeshow goodness-of-fit test (p > 0.05) was used to evaluate the model’s fit. We calibrated the model by dividing the sample into 5 equal groups that were based on the predicted probability of N2 metastasis and then plotting the median probability of each quintile versus the observed frequency of N2 metastasis. The accuracy of the model was assessed with AUC. We internally validated the model by using a crossvalidation procedure [14]. We divided the data into 10 equal test sets by sampling randomly. For each set, we calculated the predicted probability of metastasis by using parameters that were estimated from a logistic regression model that used data from the other 9 sets. These 10 logistic regressions had identical specifications; each used 90% of the data. We then built the ROC curve and calculated the AUC by using the predicted probabilities that were generated by the cross validation. We used PASW SPSS for Windows version 18.0 (Chicago, IL) and SAS for Windows Version 9.1 (Cary, NC) for statistical analysis.
Surgical Resection
Results
All of the surgical resections and lymph node dissections were conducted by thoracic surgeons at Peking
A total of 605 patients met the inclusion criteria and were enrolled into group A to develop our prediction
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
model. A positive N2 lymph node was identified in 93 patients (15.4%). The mean age of all of the patients was 62.5 10.9 years. The patients with N2 disease were significantly younger than those without N2 disease (p ¼ 0.048). The mean tumor diameter was 2.91 1.62 cm, and the diameter was significantly larger in patients with mediastinal nodal involvement than in those without mediastinal nodal involvement (p < 0.001). The tumors of patients with positive N2 nodes were more likely to be centrally located than peripherally located (36.7% compared with 10.7%, respectively; p < 0.001). The incidence of N2 disease in patients with adenocarcinoma or adenosquamous carcinoma was 17.5%, while the incidence in patients whose tumors exhibited other histologies (including large cell carcinoma and squamous carcinoma) was 9.4%, which is significantly less (p ¼ 0.02). Patients with a tumor in the lower lobe were somewhat more likely to have N2 involvement (p ¼ 0.03) (Table 1). By using multivariate logistic regression analysis, we identified 4 independent predictors of mediastinal lymph node metastasis in the prediction model: a younger age at diagnosis, central tumor location, larger tumor size, and
1763
adenocarcinoma or adenosquamous carcinoma histology (Table 2). To estimate the probability that a patient has mediastinal lymph node metastasis, we developed an equation that is based on the results of the binary logistic regression analysis. The pretest probability of N2 nodal involvement ¼ ex/(1þex), with x ¼ 0.032 age þ 1.921 central þ 0.324 size þ 1.833 path2.685, where e is the base of the natural logarithm. The units for diameter and age are centimeters and years, respectively. The location equals 1 if the tumor is centrally located or 0 if it is peripherally located. The histology equals 1 if the tumor exhibits adenocarcinoma or adenosquamous carcinoma histology or 0 if it exhibits other histologies. The HosmerLemeshow goodness-of-fit test, which was not significant (p ¼ 0.96), revealed that the predicted likelihood was of high concordance to the observed likelihood. The calibration of the model was excellent; for participants in each quintile of predicted probability, the observed frequency of mediastinal lymph node metastasis closely matched the predicted probability (Fig 1). The AUC (Fig 2) was good (0.756; 95% confidence interval [CI], 0.699 to 0.813). Internal validation by cross validation
Table 1. Univariate Analysis of Data Collected From Patients Included in Group A Variable Age (years) Tumor size (cm) Sex Male Female Symptoms No Yes Smoking history No Yes Smoking quantity (pack-years) Tumor history No Yes Family tumor history No Yes Tumor location Peripheral Central Right Left Middle or lower Upper Histology Adenocarcinoma/mixed Others
All Patients (n ¼ 605)
Patients With N2 Disease (n ¼ 93)
Patients Without N2 Disease (n ¼ 512)
p Value
62.46 10.86 2.91 1.62
60.33 11.91 3.59 1.75
62.81 10.61 2.79 1.57
0.048 <0.001
320 285
52 41
268 244
0.53
343 262
49 44
294 218
0.40
357 248 14 22.7
48 45 13.9 22.8
309 203 14.8 22.3
0.12
577 28
89 4
488 24
0.87
557 48
89 4
468 44
0.21
496 109 364 241 268 337
53 40 60 33 51 42
443 69 304 208 217 295
<0.001
446 159
78 15
368 144
0.02
0.72
0.35 0.03
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1764
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Ann Thorac Surg 2013;96:1761–8
Table 2. Multivariate Logistic Regression Analysis 95% CI Variable
Regression Coefficient
p Value
OR
Lower
Upper
0.032 0.324 1.921 1.833 2.685
0.004 <0.001 <0.001 <0.001 0.001
0.968 1.383 6.830 6.252 0.068
0.947 1.190 3.849 3.014
0.990 1.607 12.122 12.971
Age Tumor size Central Adenocarcinoma/mixed Constant CI ¼ confidence interval;
OR ¼ odds ratio.
GENERAL THORACIC
demonstrated that the predicted probability had a similar AUC (0.750; 95% CI, 0.694 to 0.808). Patients of group B were enrolled to allow external validation of the model (Table 3). Because all of our participants were diagnosed with NSCLC without lymph node enlargement on CT scans, only 2 patients in group B had an abnormal CXR. The accuracy of our prediction model was good, with an AUC of 0.757 (95% CI, 0.660 to 0.853), while that of the VA model is 0.668 (95% CI, 0.558 to 0.777) (Fig 3). Although our model was not developed specifically for patients with T1 NSCLC, we also compared it with the Fudan model, which was developed for patients with T1 NSCLC whose tumors were 3 cm or less, by using data from group B patients whose tumors were 3 cm or less. This criterion was fulfilled by 133 patients, but 78 patients were excluded because their tumors were more than 3 cm in diameter. The AUC of our model was 0.806 (95% CI, 0.679 to 0.934), while that of the Fudan model was 0.767 (95% CI, 0.634 to 0.899) (Fig 4).
Fig 1. The calibration curve for the clinical prediction model. The curve shows that the observed frequency of N2 disease was similar to the predicted probability for patients in each quintile (0.034, 0.041), (0.080, 0.083), (0.109, 0.099), (0.149, 0.132), and (0.396, 0.413).
Comment Credible and accurate detection of mediastinal lymph node metastasis is critical for staging of NSCLC. Mediastinoscopy is the traditional method for mediastinal lymph node staging. However, as an invasive method, the operative risk cannot be neglected. Some clinicians believe that mediastinoscopy should be performed only when a swollen mediastinal lymph node is detected on a preoperative CT scan [15]. The EBUS-TBNA is accepted as a safer and less invasive method for mediastinal lymph node staging [16]. Nevertheless, little evidence exists to support the effectiveness of EBUS-TBNA for staging of lymph nodes that are less than 1 cm in diameter. The PET/CT has been widely used as a noninvasive method for staging of mediastinal lymph nodes, in spite of its high cost. However, PET/CT may not always be suitable for staging of mediastinal lymph nodes in patients with NSCLC [1]. Al-Sarraf and colleagues [3] found occult mediastinal metastases in 16% of 153 patients with NSCLC that was clinically staged as clinical N0 or N1 by PET/CT. Carnochan and colleagues [17] have reported an even higher false negative rate. Some authors have observed that PET/CT was not sensitive detecting malignant involvement of lymph nodes that were less than 10 mm [5, 18]. Careful assessment of each individual’s clinical characteristics should be carried out when the lymph nodes exhibit negative mediastinal uptake of 18F-FDG-PET, especially when they are less than 10 mm. By using clinicopathologic data and logistic regression analysis, we developed a prediction model to estimate the pretest probability of mediastinal lymph node metastasis. The prediction model displayed good calibration and accuracy, and was internally validated by cross validation and externally validated by an independent dataset. As far as we know, this is the first N2-disease prediction model that has both internal and external validation. The 4 predictors in our model have been identified in other studies. Ramalingam and colleagues and Bourke and colleagues discovered that younger patients were less likely to present with local-stage disease [19] and had a higher incidence of more advanced disease [20]. These findings could arise if physicians and younger patients are both reluctant to consider a diagnosis of lung cancer, and this reluctance leads to a delayed cancer diagnosis. In a study by Lee and colleagues [8], the incidence of occult N2 disease ranged from 21.6% to as high as 26.7% in cases
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Fig 2. The receiver operating characteristic curve of our model (group A). The AUC was 0.756 (95% CI, 0.699-0.813).
1765
of NSCLC with centrally located tumors exceeding 2 cm in diameter, while the incidence in cases with peripherally located tumors was only 2.9%. Yang and colleagues [21] concluded that lymph node status had a significant relationship to tumor diameter in cases of NSCLC; the percentage of cases with no metastases (N0 or M0) was 70.8% for tumors 20 mm or less in diameter compared with 58.88% for tumors of 21 to 30 mm, 48.03% for tumors of 31 to 50 mm, 47.55% for tumors of 51 to 70 mm, and 33.33% for tumors 71 mm or greater. Suzuki and colleagues [22] determined that adenocarcinoma histology was a predictive factor for pathologic N2 in patients with NSCLC who had a negative mediastinum on a CT scan, and the findings of Kanzakia and colleagues were similar [7]. One advantage of our prediction model is its practicality. It can be used to predict the pretest probability of N2 metastasis in patients whose NSCLC has been defined by CT as N0, which is a potent reference value for the
Table 3. Patients Characteristics in Group B
Age (years) Size (cm) Sex Male Female Symptom No Yes Smoke history No Yes Smoke quantity (pack-years) Tumor history No Yes Family tumor history No Yes Abnormal CXR No Yes Tumor location Peripheral Central Right Left Middle or lower Upper Histology Adeno/mixed Others CXR ¼ chest X-ray.
All (n ¼ 211)
Patients With N2 Disease (n ¼ 21)
Patients Without N2 Disease (n ¼ 190)
61.41 9.97 3.01 1.38
56.05 11.62 3.66 1.35
62.00 9.63 2.93 1.35
0.009 0.021
128 83
13 8
115 75
0.902
92 119
10 11
82 108
0.696
117 94 15.1 21.8
11 10 10.7 15.4
106 84 15.6 22.3
0.766
197 14
20 1
177 13
0.716
200 11
21 0
179 11
0.607
209 2
19 2
190 0
<0.001
189 22 130 81 88 123
16 5 14 7 10 11
173 17 116 74 78 112
0.034
147 64
16 5
131 59
0.493
p Value
0.33
0.616 0.563
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1766
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Ann Thorac Surg 2013;96:1761–8
Fig 3. The receiver operating characteristic curves of our model and the Veterans Affairs (VA) model for all patients of group B. The AUC of our model was 0.757 (95% CI, 0.660-0.853), while that of the VA model was 0.668 (95% CI, 0.558-0.777).
GENERAL THORACIC
decision of which diagnostic test to perform. The quantified probability determined by our prediction model allows clinicians to make decisions more objectively and provides a more convenient option to get the patient involved in the decision-making process. For example, if a 70-year-old patient has a peripherally located squamous carcinoma that is 3.0 cm in diameter, then the pretest probability of N2 nodes in this patient would be 1.9%,
Fig 4. The receiver operating characteristic curves of our model and the Fudan model for group B patients with a tumor of less than 3 cm. The AUC of our model was 0.806 (95% CI, 0.679-0.934), while that of the Fudan model was 0.767 (95% CI, 0.634-0.899).
which is in a lower predicted probability quintile of our model, and the clinician would advise surgery directly. If the histology is not squamous carcinoma, but rather adenocarcinoma, then the calculated pretest probability would be 10.7%, which is in the median predicted probability quintile of our model, and PET would be needed for further assessment. If the adenocarcinoma is centrally located, not peripherally located, then the calculated
pretest probability would be 45.0%, which is in a higher predicted probability quintile of our model, and mediastinoscopy could be the best first option. Furthermore, in the patient with a high pretest probability of N2 metastasis, a negative PET result could not exclude N2 involvement. For example, a 45-year-old patient with a centrally located adenocarcinoma that is 3.0 cm in diameter has a pretest probability of N2 nodes of 64.6%. In a recent meta-analysis, the pooled weighted sensitivity and specificity of PET for detecting mediastinal lymph node involvement were 0.76 and 0.88, respectively [1]. Based on Bayes’ theorem, the calculated post-test probability of N2 involvement is approximately 33.2%, which could not rule out the use of mediastinoscopy. As we have noted before, some studies have raised concerns about the underestimation of N2 disease when the results of PET/CT are negative. Two other groups have also developed prediction models. However, CT findings were not included in the VA model because only half of the patients in their sample received CT scans; this fact restricted the model’s predictive ability. In addition, tumor size was a categoric predictor variable, not a continuous predictor variable in this model; accordingly, it could not stratify the risk of N2 disease in patients with different tumor sizes into unequal risk categories. Our model identified the same 4 clinical factors as the Fudan model, though they have dissimilar inclusion criteria. However, the ratio of each factor was greatly different between the 2 models. The factor “tumor size” occupies a larger proportion in their model. This feature may be due to the fact that the cases they used to develop their model were all T1 NSCLC, while our model was not developed by assessing cases of a particular T stage. Because the Fudan study only enrolled patients with T1 NSCLC, the applied value of the model is limited to some degree. In our opinion, it is important to note that a prediction model is neither a stand-alone method nor a precise determination criterion. The most important and practical benefit of this mathematically derived prediction model is that the estimated pretest probability of N2 disease can be used as a reference for patients with cN0 NSCLC before other invasive or noninvasive tests are performed, whatever the T stage is. Neither the VA nor the Fudan model had an external validation; therefore, they may not be sufficient and the reported performances may not be indicative of the models’ performances in future patients [23]. We externally validated our prediction model in an independent cohort of patients and compared our model with the VA and Fudan models by determining the AUC of the ROC curves. The results indicate that our model is better than both the VA and Fudan models. While this simple clinical prediction model can calculate the pretest probability of N2 disease, there are no definite thresholds for N2 metastasis. Different clinicians may choose different thresholds. However, this model provides a quantified probability that can be used as an objective reference value to aid decision-making. In addition, although this is the first N2-disease prediction
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
1767
model that has both internal and external validation, the sample size for external validation was not sufficient. Hence, more data need to be evaluated in the future. In conclusion, we have developed a 4-item predictor model for N2 metastasis in patients with CT-defined N0 NSCLC. The model was internally validated by using cross validation and externally validated with data from an independent group. The model can assist physicians in making clinical decisions about invasive or expensive procedures for mediastinal lymph node staging.
References 1. Lv YL, Yuan DM, Wang K, et al. Diagnostic performance of integrated positron emission tomography/computed tomography for mediastinal lymph node staging in nonsmall cell lung cancer. J Thorac Oncol 2011;6:1350–8. 2. Birim O, Kappetein AP, Stijnen T, Bogers AJ. Meta-analysis of positron emission tomographic and computed tomographic imaging in detecting mediastinal lymph node metastases in nonsmall cell lung cancer. Ann Thorac Surg 2005;79:375–82. 3. Al-Sarraf N, Aziz R, Gately K, et al. Pattern and predictors of occult mediastinal lymph node involvement in non-small cell lung cancer patients with negative mediastinal uptake on positron emission tomography. Eur J Cardiothorac Surg 2008;33:104–9. 4. Meyers BF, Haddad F, Siegel BA, et al. Cost-effectiveness of routine mediastinoscopy in computed tomography- and positron emission tomography-screened patients with stage I lung cancer. J Thorac Cardiovasc Surg 2006;131:822–9. 5. Bill e A, Pelosi E, Skanjeti A, et al. Preoperative intrathoracic lymph node staging in patients with non-small-cell lung cancer: accuracy of integrated positron emission tomography and computed tomography. Eur J Cardiothorac Surg 2009;36: 440–5. 6. Whitson BA, Groth SS, Maddaus MA. Recommendations for optimal use of imaging studies to clinically stage mediastinal lymph nodes in non-small-cell lung cancer patients. Lung Cancer 2008;61:177–85. 7. Kanzaki R, Higashiyama M, Fujiwara A, et al. Occult mediastinal lymph node metastasis in NSCLC patients diagnosed as clinical N0-1 by preoperative integrated FDG-PET/CT and CT: risk factors, pattern, and histopathological study. Lung Cancer 2011;71:333–7. 8. Lee PC, Port JL, Korst RJ, Liss Y, Meherally DN, Altorki NK. Risk factors for occult mediastinal metastases in clinical stage i non-small cell lung cancer. Ann Thorac Surg 2007;84: 177–81. 9. Shafazand S, Gould MK. Clinical prediction rule to estimate the probability of mediastinal metastasis in patients with non-small cell lung cancer. J Thorac Oncol 2006;1:953–9. 10. Zhang Y, Sun Y, Xiang J, Zhang Y, Hu H, Chen H. A prediction model for N2 disease in T1 non-small cell lung cancer. J Thorac Cardiovasc Surg 2012;144:1360–4. 11. Park HK, Jeon K, Koh WJ, et al. Occult nodal metastasis in patients with non-small cell lung cancer at clinical stage IA by PET/CT. Respirology 2010;15:1179–84. 12. Goldstraw P, Crowley J, Chansky K, et al. The IASLC Lung Cancer Staging Project: proposals for the revision of the TNM stage groupings in the forthcoming (seventh) edition of the TNM Classification of malignant tumours. J Thorac Oncol 2007;2:706–14. 13. Osarogiagbon RU, Allen JW, Farooq A, Wu JT. Objective review of mediastinal lymph node examination in a lung cancer resection cohort. J Thorac Oncol 2012;7:390–6. 14. Gould MK, Ananth L, Barnett PG; Veterans Affairs SNAP Cooperative Study Group. A clinical model to estimate the pretest probability of lung cancer in patients with solitary pulmonary nodules. Chest 2007;131:383–8.
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1768
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
GENERAL THORACIC
15. Takamochi K, Nagai K, Suzuki K, Yoshida J, Ohde Y, Nishiwaki Y. Clinical predictors of N2 disease in non-small cell lung cancer. Chest 2000;117:1577–82. 16. Yasufuku K, Pierre A, Darling G, et al. A prospective controlled trial of endobronchial ultrasound-guided transbronchial needle aspiration compared with mediastinoscopy for mediastinal lymph node staging of lung cancer. J Thorac Cardiovasc Surg 2011;142:1393–400. 17. Carnochan FM, Walker WS. Positron emission tomography may underestimate the extent of thoracic disease in lung cancer patients. Eur J Cardiothorac Surg 2009;35:781–5. 18. Nomori H, Watanabe K, Ohtsuka T, Naruke T, Suemasu K, Uno K. The size of metastatic foci and lymph nodes yielding false negative and false-positive lymph node staging with positron emission tomography in patients with lung cancer. J Thorac Cardiovasc Surg 2004;127:1087–92.
Ann Thorac Surg 2013;96:1761–8
19. Ramalingam S, Pawlish K, Gadgeel S, Demers R, Kalemkerian GP. Lung cancer in young patients: analysis of a Surveillance, Epidemiology, and End Results database. J Clin Oncol 1998;16:651–7. 20. Bourke W, Milstein D, Giura R, et al. Lung cancer in young adults. Chest 1992;102:1723–9. 21. Yang F, Chen H, Xiang J, et al. Relationship between tumor size and disease stage in non-small cell lung cancer. BMC Cancer 2010;10:474. 22. Suzuki K, Nagai K, Yoshida J, Nishimura M, Takahashi K, Nishiwaki Y. Clinical predictors of N2 disease in the setting of a negative computed tomographic scan in patients with lung cancer. J Thorac Cardiovasc Surg 1999;117:593–8. 23. Bleeker SE, Moll HA, Steyerberg EW, et al. External validation is necessary in prediction research: a clinical example. J Clin Epidemiol 2003;56:826–32.
Notice From the American Board of Thoracic Surgery The 2013 Part I (written) examination will be held on Monday, November 25, 2013, at multiple sites throughout the United States using an electronic format. The closing date for registration was August 15, 2013. Those wishing to be considered for examination must apply online at www.abts.org. To be admissible to the Part II (oral) examination, a candidate must have successfully completed the Part I (written) examination.
Ó 2013 by The Society of Thoracic Surgeons Published by Elsevier Inc
A candidate applying for admission to the certifying examination must fulfill all the requirements of the Board in force at the time the application is received. Please address all communications to the American Board of Thoracic Surgery, 633 N St. Clair St, Suite 2320, Chicago, IL 60611; telephone: (312) 202-5900; fax: (312) 202-5960; e-mail: [email protected].
Ann Thorac Surg 2013;96:1768
0003-4975/$36.00
Background. The true incidence of occult N2 lymph node metastasis in patients with clinical N0 non-small cell lung cancer (NSCLC) remains controversial. Estimation of the probability of N2 lymph node metastasis can assist physicians when making diagnosis and treatment decisions. Methods. We reviewed the medical records of 605 patients (group A) and 211 patients (group B) with computed tomography–defined N0 NSCLC that had an exact tumor-node-metastasis stage after surgery. Logistic regression analysis of group A’s clinical characteristics was used to estimate the independent predictors of N2 lymph node metastasis. A prediction model was then built and internally validated by using cross validation and externally validated in group B. The model was also compared with 2 previously described models. Results. We identified 4 independent predictors of N2 disease: a younger age; larger tumor size; central
tumor location; and adenocarcinoma or adenosquamous carcinoma pathology. The model showed good calibration (Hosmer-Lemeshow test: p [ 0.96) with an area under the receiver operating characteristic curve (AUC) of 0.756 (95% confidence interval, 0.699 to 0.813). The AUC of our model was better than those of the other models when validated with independent data. Conclusions. Our prediction model estimated the pretest probability of N2 disease in computed tomography-defined N0 NSCLC and was more accurate than the existing models. Use of our model can be of assistance when making clinical decisions about invasive or expensive mediastinal staging procedures.
A
PET/CT prohibits its routine application in many countries [6]. Several studies have therefore attempted to determine which factors are associated with occult mediastinal lymph node metastasis in patients with NSCLC [7, 8], because successful prediction could aid in selecting candidates for either preoperative cervical mediastinoscopy or endobronchial ultrasound-guided transbronchial needle aspiration (EBUS-TBNA). However, it is difficult to select and interpret diagnostic test results correctly without estimating the patient’s pretest probability of metastasis. Shafazand and Gould [9] of the Veterans Affairs Palo Alto Health Care System identified 6 independent predictors of N2 metastasis in 566 patients with NSCLC; the area under the receiver operating characteristic (ROC) curve was 0.70. Zhang and colleagues at the Fudan Clinic [10] retrospectively reviewed the medical records of 530 patients with CT-defined T1N0 NSCLC and identified 4 independent predictors of metastasis. The accuracy of the Fudan model is similar to that of the VA model, with an area under the ROC curve (AUC) of 0.726. However, neither of the 2 models has external validation, which is warranted to investigate the generality of the prediction model before it is implemented in clinical practice. We conducted this retrospective study to assess the real risk factors of N2 disease among patients with CT-defined
ccurate staging of non-small cell lung cancer (NSCLC) is crucial for selecting the optimal therapy. Patients with tumor-node-metastasis stages IA to IIB NSCLC are widely considered to be suitable surgical candidates, while controversy remains on the role of surgery for patients with stage IIIA NSCLC (N2 disease). This controversy makes preoperative mediastinal staging particularly important for patients who are potentially suitable surgical candidates. The F-18-fluorodeoxyglucose positron emission tomography combined with thoracic computed tomography (18F-FDG PET/CT) is increasingly utilized for staging of mediastinal lymph nodes [1]. However, a few studies have raised concerns about the underestimation of N2 disease when PET/CT results are negative [2, 3]. In patients who have been clinically staged as negative for N2 disease, the incidence of occult mediastinal lymph node metastasis ranges from 5.6% to 18% based on the T stage [4]. One study demonstrated that the sensitivity of PET/CT for detecting malignant nodal involvement was only 32.4% in nodes less than 10 mm [5]. In addition, the high expense of Accepted for publication June 7, 2013. Address correspondence to Dr Wang, Department of Thoracic Surgery, Peking University People’s Hospital, No. 11 Xizhimen South St, Xicheng District, Beijing, Ot 100044, P. R. China; e-mail: [email protected].
Ó 2013 by The Society of Thoracic Surgeons Published by Elsevier Inc
(Ann Thorac Surg 2013;96:1761–8) Ó 2013 by The Society of Thoracic Surgeons
0003-4975/$36.00 http://dx.doi.org/10.1016/j.athoracsur.2013.06.038
GENERAL THORACIC
Department of Thoracic Surgery, Peking University People’s Hospital, Beijing, People’s Republic of China
1762
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
N0 NSCLC. We developed a clinically useful prediction model, which we validated both internally and externally, to estimate the pretest probability of N2 metastasis in patients with NSCLC. We also compared the accuracy of our model with that of the VA and Fudan models by calculating the AUC. It is important to note that the value of this model lies in its ability to estimate the pretest probability, which can then be used to guide the subsequent selection of diagnostic tests; the model is not intended for use as a stand-alone diagnostic method.
Ann Thorac Surg 2013;96:1761–8
University People’s Hospital. The techniques used to dissect the lymph nodes were standardized. Mediastinal lymph node dissection or systematic lymph node sampling were acceptable procedures (American College of Surgeons Oncology Group criteria) [13].
Established Mathematical Models The Veterans Affairs model: The pretest probability of N2 nodal involvement ¼ ex/(1þex). x ¼ 1:806 þ ð0:955 adenoÞ þ ð0:876
GENERAL THORACIC
Material and Methods
abnormal CXRÞ þ ð0:749 centralÞ þ 0:485 primary symptom þ ð0:435 sizeÞ ð0:408
Patient Eligibility
age > 65Þ:
Between September 2006 and August 2011, we reviewed the medical records of 781 patients with a pathologic diagnosis of NSCLC who underwent surgical resection with lymph node dissection. All of the patients underwent contrast-enhanced thoracic CT scans at the Peking University People’s Hospital before surgery. Other routine preoperative examinations included chest radiograph, cardiopulmonary function tests, abdominal and adrenal gland ultrasonography or CT, brain magnetic resonance imaging or CT, and a bone scan. A lymph node was considered to be positive if the short axis exceeded 1 cm on a thoracic CT image. One hundred fourteen patients who had been diagnosed with CT-defined N1 or N2 NSCLC before surgery were excluded. Thirty-two patients with pure bronchioalveolar carcinoma (BAC) [11] were excluded because BAC was defined as a noninvasive adenocarcinoma in accordance with the revised World Health Organization classification [12], 11 patients were excluded because of a history of lung malignancy within 5 years, and 19 patients were excluded because of incomplete data. In total, 605 patients were enrolled into group A, and their clinical data were analyzed to create the mathematical model. Clinical data were also collected from an additional 211 patients who were enrolled with the same inclusion criteria between September 2011 and December 2012 (group B).
Clinical Variables We collected clinical data of patients in Groups A and B that included the age and gender of the patient, symptoms at presentation, smoking history, average number of cigarettes per year, family history of cancer, tumor histology and location, and tumor size measured by CT. We also analyzed the chest radiographs (CXR) of patients in group B to determine the appearance of the mediastinum. Tumors were considered to be centrally located if the center of the tumor was in the medial third of the lung parenchyma (adjacent to the mediastinum) and peripherally located if the center of the tumor was within the lateral two-thirds of the lung parenchyma.
The Fudan model. The pretest probability of N2 nodal involvement ¼ ex/(1þex). x ¼ 3:449 þ ð1:018 diameterÞ þ ð1:164 locationÞ þ ð1:263 histologyÞ ð0:026 ageÞ:
Statistical Analysis Univariate analysis of the correlation between mediastinal lymph node metastasis and the categoric variables was conducted by using the Fisher exact test or Pearson c2 test. An independent sample t test was used to assess the association between N2 disease and continuous variables, which are expressed as the mean values and standard deviations. Variables with a p value less than 0.1 were subjected to a binary logistic regression analysis. We developed the model by using stepwise logistic regression, with factors that were statistically significant at the 0.05 level remaining in the final model as the dependent variables. The Hosmer-Lemeshow goodness-of-fit test (p > 0.05) was used to evaluate the model’s fit. We calibrated the model by dividing the sample into 5 equal groups that were based on the predicted probability of N2 metastasis and then plotting the median probability of each quintile versus the observed frequency of N2 metastasis. The accuracy of the model was assessed with AUC. We internally validated the model by using a crossvalidation procedure [14]. We divided the data into 10 equal test sets by sampling randomly. For each set, we calculated the predicted probability of metastasis by using parameters that were estimated from a logistic regression model that used data from the other 9 sets. These 10 logistic regressions had identical specifications; each used 90% of the data. We then built the ROC curve and calculated the AUC by using the predicted probabilities that were generated by the cross validation. We used PASW SPSS for Windows version 18.0 (Chicago, IL) and SAS for Windows Version 9.1 (Cary, NC) for statistical analysis.
Surgical Resection
Results
All of the surgical resections and lymph node dissections were conducted by thoracic surgeons at Peking
A total of 605 patients met the inclusion criteria and were enrolled into group A to develop our prediction
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
model. A positive N2 lymph node was identified in 93 patients (15.4%). The mean age of all of the patients was 62.5 10.9 years. The patients with N2 disease were significantly younger than those without N2 disease (p ¼ 0.048). The mean tumor diameter was 2.91 1.62 cm, and the diameter was significantly larger in patients with mediastinal nodal involvement than in those without mediastinal nodal involvement (p < 0.001). The tumors of patients with positive N2 nodes were more likely to be centrally located than peripherally located (36.7% compared with 10.7%, respectively; p < 0.001). The incidence of N2 disease in patients with adenocarcinoma or adenosquamous carcinoma was 17.5%, while the incidence in patients whose tumors exhibited other histologies (including large cell carcinoma and squamous carcinoma) was 9.4%, which is significantly less (p ¼ 0.02). Patients with a tumor in the lower lobe were somewhat more likely to have N2 involvement (p ¼ 0.03) (Table 1). By using multivariate logistic regression analysis, we identified 4 independent predictors of mediastinal lymph node metastasis in the prediction model: a younger age at diagnosis, central tumor location, larger tumor size, and
1763
adenocarcinoma or adenosquamous carcinoma histology (Table 2). To estimate the probability that a patient has mediastinal lymph node metastasis, we developed an equation that is based on the results of the binary logistic regression analysis. The pretest probability of N2 nodal involvement ¼ ex/(1þex), with x ¼ 0.032 age þ 1.921 central þ 0.324 size þ 1.833 path2.685, where e is the base of the natural logarithm. The units for diameter and age are centimeters and years, respectively. The location equals 1 if the tumor is centrally located or 0 if it is peripherally located. The histology equals 1 if the tumor exhibits adenocarcinoma or adenosquamous carcinoma histology or 0 if it exhibits other histologies. The HosmerLemeshow goodness-of-fit test, which was not significant (p ¼ 0.96), revealed that the predicted likelihood was of high concordance to the observed likelihood. The calibration of the model was excellent; for participants in each quintile of predicted probability, the observed frequency of mediastinal lymph node metastasis closely matched the predicted probability (Fig 1). The AUC (Fig 2) was good (0.756; 95% confidence interval [CI], 0.699 to 0.813). Internal validation by cross validation
Table 1. Univariate Analysis of Data Collected From Patients Included in Group A Variable Age (years) Tumor size (cm) Sex Male Female Symptoms No Yes Smoking history No Yes Smoking quantity (pack-years) Tumor history No Yes Family tumor history No Yes Tumor location Peripheral Central Right Left Middle or lower Upper Histology Adenocarcinoma/mixed Others
All Patients (n ¼ 605)
Patients With N2 Disease (n ¼ 93)
Patients Without N2 Disease (n ¼ 512)
p Value
62.46 10.86 2.91 1.62
60.33 11.91 3.59 1.75
62.81 10.61 2.79 1.57
0.048 <0.001
320 285
52 41
268 244
0.53
343 262
49 44
294 218
0.40
357 248 14 22.7
48 45 13.9 22.8
309 203 14.8 22.3
0.12
577 28
89 4
488 24
0.87
557 48
89 4
468 44
0.21
496 109 364 241 268 337
53 40 60 33 51 42
443 69 304 208 217 295
<0.001
446 159
78 15
368 144
0.02
0.72
0.35 0.03
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1764
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Ann Thorac Surg 2013;96:1761–8
Table 2. Multivariate Logistic Regression Analysis 95% CI Variable
Regression Coefficient
p Value
OR
Lower
Upper
0.032 0.324 1.921 1.833 2.685
0.004 <0.001 <0.001 <0.001 0.001
0.968 1.383 6.830 6.252 0.068
0.947 1.190 3.849 3.014
0.990 1.607 12.122 12.971
Age Tumor size Central Adenocarcinoma/mixed Constant CI ¼ confidence interval;
OR ¼ odds ratio.
GENERAL THORACIC
demonstrated that the predicted probability had a similar AUC (0.750; 95% CI, 0.694 to 0.808). Patients of group B were enrolled to allow external validation of the model (Table 3). Because all of our participants were diagnosed with NSCLC without lymph node enlargement on CT scans, only 2 patients in group B had an abnormal CXR. The accuracy of our prediction model was good, with an AUC of 0.757 (95% CI, 0.660 to 0.853), while that of the VA model is 0.668 (95% CI, 0.558 to 0.777) (Fig 3). Although our model was not developed specifically for patients with T1 NSCLC, we also compared it with the Fudan model, which was developed for patients with T1 NSCLC whose tumors were 3 cm or less, by using data from group B patients whose tumors were 3 cm or less. This criterion was fulfilled by 133 patients, but 78 patients were excluded because their tumors were more than 3 cm in diameter. The AUC of our model was 0.806 (95% CI, 0.679 to 0.934), while that of the Fudan model was 0.767 (95% CI, 0.634 to 0.899) (Fig 4).
Fig 1. The calibration curve for the clinical prediction model. The curve shows that the observed frequency of N2 disease was similar to the predicted probability for patients in each quintile (0.034, 0.041), (0.080, 0.083), (0.109, 0.099), (0.149, 0.132), and (0.396, 0.413).
Comment Credible and accurate detection of mediastinal lymph node metastasis is critical for staging of NSCLC. Mediastinoscopy is the traditional method for mediastinal lymph node staging. However, as an invasive method, the operative risk cannot be neglected. Some clinicians believe that mediastinoscopy should be performed only when a swollen mediastinal lymph node is detected on a preoperative CT scan [15]. The EBUS-TBNA is accepted as a safer and less invasive method for mediastinal lymph node staging [16]. Nevertheless, little evidence exists to support the effectiveness of EBUS-TBNA for staging of lymph nodes that are less than 1 cm in diameter. The PET/CT has been widely used as a noninvasive method for staging of mediastinal lymph nodes, in spite of its high cost. However, PET/CT may not always be suitable for staging of mediastinal lymph nodes in patients with NSCLC [1]. Al-Sarraf and colleagues [3] found occult mediastinal metastases in 16% of 153 patients with NSCLC that was clinically staged as clinical N0 or N1 by PET/CT. Carnochan and colleagues [17] have reported an even higher false negative rate. Some authors have observed that PET/CT was not sensitive detecting malignant involvement of lymph nodes that were less than 10 mm [5, 18]. Careful assessment of each individual’s clinical characteristics should be carried out when the lymph nodes exhibit negative mediastinal uptake of 18F-FDG-PET, especially when they are less than 10 mm. By using clinicopathologic data and logistic regression analysis, we developed a prediction model to estimate the pretest probability of mediastinal lymph node metastasis. The prediction model displayed good calibration and accuracy, and was internally validated by cross validation and externally validated by an independent dataset. As far as we know, this is the first N2-disease prediction model that has both internal and external validation. The 4 predictors in our model have been identified in other studies. Ramalingam and colleagues and Bourke and colleagues discovered that younger patients were less likely to present with local-stage disease [19] and had a higher incidence of more advanced disease [20]. These findings could arise if physicians and younger patients are both reluctant to consider a diagnosis of lung cancer, and this reluctance leads to a delayed cancer diagnosis. In a study by Lee and colleagues [8], the incidence of occult N2 disease ranged from 21.6% to as high as 26.7% in cases
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Fig 2. The receiver operating characteristic curve of our model (group A). The AUC was 0.756 (95% CI, 0.699-0.813).
1765
of NSCLC with centrally located tumors exceeding 2 cm in diameter, while the incidence in cases with peripherally located tumors was only 2.9%. Yang and colleagues [21] concluded that lymph node status had a significant relationship to tumor diameter in cases of NSCLC; the percentage of cases with no metastases (N0 or M0) was 70.8% for tumors 20 mm or less in diameter compared with 58.88% for tumors of 21 to 30 mm, 48.03% for tumors of 31 to 50 mm, 47.55% for tumors of 51 to 70 mm, and 33.33% for tumors 71 mm or greater. Suzuki and colleagues [22] determined that adenocarcinoma histology was a predictive factor for pathologic N2 in patients with NSCLC who had a negative mediastinum on a CT scan, and the findings of Kanzakia and colleagues were similar [7]. One advantage of our prediction model is its practicality. It can be used to predict the pretest probability of N2 metastasis in patients whose NSCLC has been defined by CT as N0, which is a potent reference value for the
Table 3. Patients Characteristics in Group B
Age (years) Size (cm) Sex Male Female Symptom No Yes Smoke history No Yes Smoke quantity (pack-years) Tumor history No Yes Family tumor history No Yes Abnormal CXR No Yes Tumor location Peripheral Central Right Left Middle or lower Upper Histology Adeno/mixed Others CXR ¼ chest X-ray.
All (n ¼ 211)
Patients With N2 Disease (n ¼ 21)
Patients Without N2 Disease (n ¼ 190)
61.41 9.97 3.01 1.38
56.05 11.62 3.66 1.35
62.00 9.63 2.93 1.35
0.009 0.021
128 83
13 8
115 75
0.902
92 119
10 11
82 108
0.696
117 94 15.1 21.8
11 10 10.7 15.4
106 84 15.6 22.3
0.766
197 14
20 1
177 13
0.716
200 11
21 0
179 11
0.607
209 2
19 2
190 0
<0.001
189 22 130 81 88 123
16 5 14 7 10 11
173 17 116 74 78 112
0.034
147 64
16 5
131 59
0.493
p Value
0.33
0.616 0.563
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1766
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
Ann Thorac Surg 2013;96:1761–8
Fig 3. The receiver operating characteristic curves of our model and the Veterans Affairs (VA) model for all patients of group B. The AUC of our model was 0.757 (95% CI, 0.660-0.853), while that of the VA model was 0.668 (95% CI, 0.558-0.777).
GENERAL THORACIC
decision of which diagnostic test to perform. The quantified probability determined by our prediction model allows clinicians to make decisions more objectively and provides a more convenient option to get the patient involved in the decision-making process. For example, if a 70-year-old patient has a peripherally located squamous carcinoma that is 3.0 cm in diameter, then the pretest probability of N2 nodes in this patient would be 1.9%,
Fig 4. The receiver operating characteristic curves of our model and the Fudan model for group B patients with a tumor of less than 3 cm. The AUC of our model was 0.806 (95% CI, 0.679-0.934), while that of the Fudan model was 0.767 (95% CI, 0.634-0.899).
which is in a lower predicted probability quintile of our model, and the clinician would advise surgery directly. If the histology is not squamous carcinoma, but rather adenocarcinoma, then the calculated pretest probability would be 10.7%, which is in the median predicted probability quintile of our model, and PET would be needed for further assessment. If the adenocarcinoma is centrally located, not peripherally located, then the calculated
pretest probability would be 45.0%, which is in a higher predicted probability quintile of our model, and mediastinoscopy could be the best first option. Furthermore, in the patient with a high pretest probability of N2 metastasis, a negative PET result could not exclude N2 involvement. For example, a 45-year-old patient with a centrally located adenocarcinoma that is 3.0 cm in diameter has a pretest probability of N2 nodes of 64.6%. In a recent meta-analysis, the pooled weighted sensitivity and specificity of PET for detecting mediastinal lymph node involvement were 0.76 and 0.88, respectively [1]. Based on Bayes’ theorem, the calculated post-test probability of N2 involvement is approximately 33.2%, which could not rule out the use of mediastinoscopy. As we have noted before, some studies have raised concerns about the underestimation of N2 disease when the results of PET/CT are negative. Two other groups have also developed prediction models. However, CT findings were not included in the VA model because only half of the patients in their sample received CT scans; this fact restricted the model’s predictive ability. In addition, tumor size was a categoric predictor variable, not a continuous predictor variable in this model; accordingly, it could not stratify the risk of N2 disease in patients with different tumor sizes into unequal risk categories. Our model identified the same 4 clinical factors as the Fudan model, though they have dissimilar inclusion criteria. However, the ratio of each factor was greatly different between the 2 models. The factor “tumor size” occupies a larger proportion in their model. This feature may be due to the fact that the cases they used to develop their model were all T1 NSCLC, while our model was not developed by assessing cases of a particular T stage. Because the Fudan study only enrolled patients with T1 NSCLC, the applied value of the model is limited to some degree. In our opinion, it is important to note that a prediction model is neither a stand-alone method nor a precise determination criterion. The most important and practical benefit of this mathematically derived prediction model is that the estimated pretest probability of N2 disease can be used as a reference for patients with cN0 NSCLC before other invasive or noninvasive tests are performed, whatever the T stage is. Neither the VA nor the Fudan model had an external validation; therefore, they may not be sufficient and the reported performances may not be indicative of the models’ performances in future patients [23]. We externally validated our prediction model in an independent cohort of patients and compared our model with the VA and Fudan models by determining the AUC of the ROC curves. The results indicate that our model is better than both the VA and Fudan models. While this simple clinical prediction model can calculate the pretest probability of N2 disease, there are no definite thresholds for N2 metastasis. Different clinicians may choose different thresholds. However, this model provides a quantified probability that can be used as an objective reference value to aid decision-making. In addition, although this is the first N2-disease prediction
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
1767
model that has both internal and external validation, the sample size for external validation was not sufficient. Hence, more data need to be evaluated in the future. In conclusion, we have developed a 4-item predictor model for N2 metastasis in patients with CT-defined N0 NSCLC. The model was internally validated by using cross validation and externally validated with data from an independent group. The model can assist physicians in making clinical decisions about invasive or expensive procedures for mediastinal lymph node staging.
References 1. Lv YL, Yuan DM, Wang K, et al. Diagnostic performance of integrated positron emission tomography/computed tomography for mediastinal lymph node staging in nonsmall cell lung cancer. J Thorac Oncol 2011;6:1350–8. 2. Birim O, Kappetein AP, Stijnen T, Bogers AJ. Meta-analysis of positron emission tomographic and computed tomographic imaging in detecting mediastinal lymph node metastases in nonsmall cell lung cancer. Ann Thorac Surg 2005;79:375–82. 3. Al-Sarraf N, Aziz R, Gately K, et al. Pattern and predictors of occult mediastinal lymph node involvement in non-small cell lung cancer patients with negative mediastinal uptake on positron emission tomography. Eur J Cardiothorac Surg 2008;33:104–9. 4. Meyers BF, Haddad F, Siegel BA, et al. Cost-effectiveness of routine mediastinoscopy in computed tomography- and positron emission tomography-screened patients with stage I lung cancer. J Thorac Cardiovasc Surg 2006;131:822–9. 5. Bill e A, Pelosi E, Skanjeti A, et al. Preoperative intrathoracic lymph node staging in patients with non-small-cell lung cancer: accuracy of integrated positron emission tomography and computed tomography. Eur J Cardiothorac Surg 2009;36: 440–5. 6. Whitson BA, Groth SS, Maddaus MA. Recommendations for optimal use of imaging studies to clinically stage mediastinal lymph nodes in non-small-cell lung cancer patients. Lung Cancer 2008;61:177–85. 7. Kanzaki R, Higashiyama M, Fujiwara A, et al. Occult mediastinal lymph node metastasis in NSCLC patients diagnosed as clinical N0-1 by preoperative integrated FDG-PET/CT and CT: risk factors, pattern, and histopathological study. Lung Cancer 2011;71:333–7. 8. Lee PC, Port JL, Korst RJ, Liss Y, Meherally DN, Altorki NK. Risk factors for occult mediastinal metastases in clinical stage i non-small cell lung cancer. Ann Thorac Surg 2007;84: 177–81. 9. Shafazand S, Gould MK. Clinical prediction rule to estimate the probability of mediastinal metastasis in patients with non-small cell lung cancer. J Thorac Oncol 2006;1:953–9. 10. Zhang Y, Sun Y, Xiang J, Zhang Y, Hu H, Chen H. A prediction model for N2 disease in T1 non-small cell lung cancer. J Thorac Cardiovasc Surg 2012;144:1360–4. 11. Park HK, Jeon K, Koh WJ, et al. Occult nodal metastasis in patients with non-small cell lung cancer at clinical stage IA by PET/CT. Respirology 2010;15:1179–84. 12. Goldstraw P, Crowley J, Chansky K, et al. The IASLC Lung Cancer Staging Project: proposals for the revision of the TNM stage groupings in the forthcoming (seventh) edition of the TNM Classification of malignant tumours. J Thorac Oncol 2007;2:706–14. 13. Osarogiagbon RU, Allen JW, Farooq A, Wu JT. Objective review of mediastinal lymph node examination in a lung cancer resection cohort. J Thorac Oncol 2012;7:390–6. 14. Gould MK, Ananth L, Barnett PG; Veterans Affairs SNAP Cooperative Study Group. A clinical model to estimate the pretest probability of lung cancer in patients with solitary pulmonary nodules. Chest 2007;131:383–8.
GENERAL THORACIC
Ann Thorac Surg 2013;96:1761–8
1768
CHEN ET AL N2 METASTASIS IN PATIENTS WITH CT-DEFINED N0 NSCLC
GENERAL THORACIC
15. Takamochi K, Nagai K, Suzuki K, Yoshida J, Ohde Y, Nishiwaki Y. Clinical predictors of N2 disease in non-small cell lung cancer. Chest 2000;117:1577–82. 16. Yasufuku K, Pierre A, Darling G, et al. A prospective controlled trial of endobronchial ultrasound-guided transbronchial needle aspiration compared with mediastinoscopy for mediastinal lymph node staging of lung cancer. J Thorac Cardiovasc Surg 2011;142:1393–400. 17. Carnochan FM, Walker WS. Positron emission tomography may underestimate the extent of thoracic disease in lung cancer patients. Eur J Cardiothorac Surg 2009;35:781–5. 18. Nomori H, Watanabe K, Ohtsuka T, Naruke T, Suemasu K, Uno K. The size of metastatic foci and lymph nodes yielding false negative and false-positive lymph node staging with positron emission tomography in patients with lung cancer. J Thorac Cardiovasc Surg 2004;127:1087–92.
Ann Thorac Surg 2013;96:1761–8
19. Ramalingam S, Pawlish K, Gadgeel S, Demers R, Kalemkerian GP. Lung cancer in young patients: analysis of a Surveillance, Epidemiology, and End Results database. J Clin Oncol 1998;16:651–7. 20. Bourke W, Milstein D, Giura R, et al. Lung cancer in young adults. Chest 1992;102:1723–9. 21. Yang F, Chen H, Xiang J, et al. Relationship between tumor size and disease stage in non-small cell lung cancer. BMC Cancer 2010;10:474. 22. Suzuki K, Nagai K, Yoshida J, Nishimura M, Takahashi K, Nishiwaki Y. Clinical predictors of N2 disease in the setting of a negative computed tomographic scan in patients with lung cancer. J Thorac Cardiovasc Surg 1999;117:593–8. 23. Bleeker SE, Moll HA, Steyerberg EW, et al. External validation is necessary in prediction research: a clinical example. J Clin Epidemiol 2003;56:826–32.
Notice From the American Board of Thoracic Surgery The 2013 Part I (written) examination will be held on Monday, November 25, 2013, at multiple sites throughout the United States using an electronic format. The closing date for registration was August 15, 2013. Those wishing to be considered for examination must apply online at www.abts.org. To be admissible to the Part II (oral) examination, a candidate must have successfully completed the Part I (written) examination.
Ó 2013 by The Society of Thoracic Surgeons Published by Elsevier Inc
A candidate applying for admission to the certifying examination must fulfill all the requirements of the Board in force at the time the application is received. Please address all communications to the American Board of Thoracic Surgery, 633 N St. Clair St, Suite 2320, Chicago, IL 60611; telephone: (312) 202-5900; fax: (312) 202-5960; e-mail: [email protected].
Ann Thorac Surg 2013;96:1768
0003-4975/$36.00