Abstract: Objective: To estimate the probability of N2 lymph node metastasis and to assist physicians in making diagnosis and treatment decisions. Methods: We reviewed the medical records of 739 patients with computed tomography-defined stage Ⅰ non-small cell lung cancer (NSCLC) that had an exact tumor-node-metastasis stage after surgery. A random subset of three fourths of the patients (n=554) were selected to develop the prediction model. Logistic regression analysis of the clinical characteristics was used to estimate the independent predictors of N2 lymph node metastasis. A prediction model was then built and externally validated by the remaining one fourth (n=185) patients which made up the validation data set. 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=0.923) with an area under the receiver operating characteristic curve (AUC) of 0.748 (95% confidence interval, 0.710-0.784). When validated with all the patients of group B, the AUC of our model was 0.781 (95% CI: 0.715-0.839) and the VA model was 0.677 (95% CI: 0.604-0.744) (P =0.04). When validated with T1 patients of group B, the AUC of our model was 0.837 (95% CI: 0.760-0.897) and Fudan model was 0.766 (95% CI: 0.681-0.837) (P<0.01). Conclusion: Our prediction model estimated the pretest probability of N2 disease in computed tomography-defined stage Ⅰ 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.

Key words: Carcinoma, non-small-cell lung, Logistic models, ROC curve, Diagnosis