View metadata, citation and similar papers at core.ac.uk brought to you by CORE provided by Elsevier - Publisher Connector General Thoracic Surgery Zhang et al A prediction model for N2 disease in T1 non–small cell lung cancer Yang Zhang, MD,a,b Yihua Sun, MD,a,b Jiaqing Xiang, MD,a,b Yawei Zhang, MD,a,b Hong Hu, MD,a,b and Haiquan Chen, MDa,b Objective: Controversy remains over the routine use of mediastinoscopy or positron emission tomography in T1 non–small cell lung cancer without lymph node enlargement on computed tomography because the risk of N2 GTS involvement is comparatively low. We aimed to develop a prediction model for N2 disease in cT1N0 non–small cell lung cancer to aid in the decision-making process. Methods: We reviewed the records of 530 patients with computed tomography–defined T1N0 non–small cell lung cancer who underwent surgical resection with systematic lymph node dissection. Correlations between N2 involvement and clinicopathologic parameters were assessed using univariate analysis and binary logistic regression analysis. A prediction model was built on the basis of logistic regression analysis and was internally validated using bootstrapping. Results: The incidence of N2 disease was 16.8%. Four independent predictors were identified in multivariate logistic regression analysis and included in the prediction model: younger age at diagnosis (odds ratio, 0.974; 95% confidence interval, 0.952-0.997), larger tumor size (odds ratio, 2.769; 95% confidence interval, 1.818-4.217), central tumor location (odds ratio, 3.204; 95% confidence interval, 1.512-6.790), and invasive ad- enocarcinoma histology (odds ratio, 3.537; 95% confidence interval, 1.740-7.191). This model shows good cal- ibration (Hosmer–Lemeshow test: P ¼ .784), reasonable discrimination (area under the receiver operating characteristic curve, 0.726; 95% confidence interval, 0.669-0.784), and minimal overfitting demonstrated by bootstrapping. Conclusions: We developed a 4-predictor model that can estimate the probability of N2 disease in computed tomography–defined T1N0 non–small cell lung cancer. This prediction model can help to determine the cost- effective use of mediastinal staging procedures. (J Thorac Cardiovasc Surg 2012;144:1360-4) The wide application of computed tomography (CT) has (CT-defined N0) because the risk of N2 disease is low in increased the detection of T1 (3 cm)1 lung cancers. Surgi- such patients.8,9 Although positron emission tomography cal resection is considered the optimal treatment for T1 (PET) showed superiority over CT in the mediastinal non–small cell lung cancer (NSCLC) without mediastinal staging of NSCLC10,11 and a promise to reduce the need lymph node (N2) involvement or distant metastasis. How- for invasive staging tools, its high expense is an obstacle ever, patients with T1 lung cancer with N2 involvement to the routine application in many countries.12 Moreover, should take definitive concurrent chemoradiation or induc- the benefit of PET for CT-defined clinical stage IA in pa- tion chemotherapy.2-4 Accurate mediastinal staging is a key tients also remains controversial.13 factor for the successful management of NSCLC. Some studies sought to determine the cost-effective strat- Mediastinoscopy is deemed the gold standard for medi- egies of applying invasive or expensive diagnostic proce- astinal lymph node staging.5,6 However, the risk of dures.14,15 To obtain pretest probability of N2 disease is morbidity and mortality (2% and 0.08%, respectively)7 essential for the cost-effectiveness measurement of subse- cannot be overlooked because of the invasive nature of me- quent diagnostic tests. Therefore, this study aimed to de- diastinoscopy. Controversy remains as to whether routine velop a risk prediction model of N2 disease in CT-defined mediastinoscopy should be performed in patients with T1 T1N0 NSCLC. lung cancer who have no nodal enlargement on CT scans PATIENTS AND METHODS From the Department of Thoracic Surgery,a Fudan University Shanghai Cancer Cen- Patients ter, Shanghai, China; and Department of Oncology,b Shanghai Medical College, From June 2007 to August 2011, we retrospectively reviewed our data- Fudan University, Shanghai, China. base of all patients who underwent resection with curative intention at the Disclosures: Authors have nothing to disclose with regard to commercial support. Department of Thoracic Surgery, Fudan University Shanghai Cancer Hos- Received for publication Feb 16, 2012; revisions received May 14, 2012; accepted for pital, Shanghai, China. We routinely performed contrast-enhanced chest publication June 18, 2012; available ahead of print July 23, 2012. CT scans at the Fudan University Shanghai Cancer Hospital before surgery, Address for reprints: Haiquan Chen, MD, Department of Thoracic Surgery, Fudan even for the patients who had received CT scans in other hospitals. Other University Shanghai Cancer Center, 270 Dong-An Rd, Shanghai 200032, China (E-mail: [email protected]). routine preoperative examinations included cardiopulmonary tests, brain 0022-5223/$36.00 magnetic resonance imaging or CT, bone scanning, and abdominal CT or Copyright Ó 2012 by The American Association for Thoracic Surgery ultrasonography. Lymph nodes were considered to be positive if the short http://dx.doi.org/10.1016/j.jtcvs.2012.06.050 axis exceeded 1 cm on chest CT images. Peripheral nodules were defined 1360 The Journal of Thoracic and Cardiovascular Surgery c December 2012 Zhang et al General Thoracic Surgery median age at diagnosis was 59 years (interquartile range, Abbreviations and Acronyms 13). Tumor size ranged from 0.4 to 3.0 cm (median, 2.0; in- CT ¼ computed tomography terquartile range, 1.0). The most common lobar location NSCLC ¼ non–small cell lung cancer was the left upper lobe (30.4%), followed by the right upper PET ¼ positron emission tomography lobe (28.7%). The majority (91.5%) of patients underwent ROC ¼ receiver operating characteristic lobectomy. The incidence of mediastinal lymph node metastasis was 16.8% (89/530). In univariate analysis (Table 1), N2-positive patients were significantly younger than as tumors with the center located within the outer one third of the lung on N2-negative patients (P ¼ .028). The mean tumor diameter GTS CT scans. Inclusion criteria included (1) NSCLC 3 cm or less in diameter of patients with mediastinal nodal involvement was signif- measured on CT scans without evidence of positive lymph nodes or distant metastasis (cT1N0M0) and (2) systematic lymph node dissection. Patients icantly larger than those without mediastinal nodal involve- who had adenocarcinoma in situ or minimally invasive adenocarcinoma ment (P <.001). The incidence of N2 disease in patients histology according to the new lung adenocarcinoma classification16 with invasive adenocarcinoma was 19.5% compared with were excluded (only invasive adenocarcinomas were included). Patients 9.6% in patients with other histology (P ¼ .006). Patients who received neoadjuvant chemotherapy or radiotherapy were excluded. with positive N2 nodes were more likely to be never- Patients with a history of malignant tumors were excluded. % % ¼ Clinicopathologic data on age, gender, smoking history, family history smokers (69.7 vs 58.0 ; P .041) and to have centrally of lung cancer, symptoms at presentation, tumor site, tumor size measured located tumors (18.0% vs 10.2%; P ¼ .036). on CT, type of surgery, histology, and lymph nodal status according to path- Finally, 4 independent predictors were included in the ologic reports were collected. prediction model after multivariate logistic regression. This study was conducted in accordance with the Helsinki Declaration Odds ratios, 95% confidence intervals, and P values of sig- and approved by the institutional review board of the Fudan University Shanghai Cancer Center, Shanghai, China. Informed consent was waived nificant predictors are listed in Table 2. Larger tumor size because this was a retrospective analysis. (P<.001), central tumor location (P ¼ .002), invasive ad- enocarcinoma histology (P<.001), and younger age at di- Statistical Analyses agnosis (P ¼ .025) were independent predictors of In univariate analyses, we used Pearson’s chi-square test or the Fisher mediastinal lymph node metastasis. exact test to evaluate the correlation between a mediastinal lymph node me- A formula was developed to estimate the probability tastasis and a categoric variable, and an independent sample t test to assess of having mediastinal lymph node metastasis on the the association between N2 involvement and a continuous variable. Vari- ables with a P value less than .2 were entered into a binary logistic regres- basis of the results of the binary logistic regression sion analysis that formed the basis of a prediction model. We used forward analysis. A score is calculated using tumor diameter, stepwise selection procedures with the likelihood-ratio test. Factors statis- tumor location, histology, and age at diagnosis: tically significant at the .05 level remained in the final model. score ¼3.449 þ (1.018 $ diameter) þ (1.164 $ location) Calibration (concordance between predicted and observed probabili- þ(1.263 $ histology)À(0.026 $ age). The units for diame- ties) of the final model was determined with the Hosmer–Lemeshow statis- tic and the calibration plot using 10 equal contiguous risk ranges showing ter and age are centimeter and year, respectively. If the observed versus predicted probabilities. The discriminative ability of the tumor is centrally located, location ¼ 1 (location ¼ 0 model was assessed with the area under the receiver operating characteris- if the tumor is peripherally located). If the tumor is ad- tic (ROC) curve, which ranges from 0.5 (no discrimination) to 1.0 (perfect enocarcinoma in histology, histology ¼ 1 (histology ¼ 0 discrimination). for other histology). The likelihood of N2 nodal A problem with a predictive model is that the performance is overesti- mated when assessed on the sample used to build the model.17 We inter- involvement is then calculated: likelihood of positive Score Score nally validated this model by bootstrapping, which is reported to be N2 nodes ¼ e /(1 þ e ).