Aims: Lung cancer patients within the pN0 category have a significantly different outcome. The aim of this study was to develop a mathematical model to assist in predicting the prognosis of pN0 lung squamous cell carcinoma (SCC).
Methods and results: Twenty-three proteins were examined by immunohistochemical (IHC) analysis on primary tumour tissues from 319 lung SCC patients. In a training group, using IHC data, a recursive partitioning decision tree (RP-DT) was used to build a model for estimating the risk for lymphatic metastasis. This model was then validated in a test cohort. Of 23 proteins, 8 (matrix metallopeptidase 1, metalloproteinase inhibitor 1, Ras GTPase-activating-like protein IQGAP1, targeting protein for Xklp2, urokinase-type plasminogen activator, cathepsin D, fascin, polymeric immunoglobulin receptor/secretory component) were selected, and generated a tree model in a training group of 255 patients to classify them as at high or low risk of lymphatic invasion, with accuracy of 78.0% (compared to histopathological diagnosis), sensitivity of 83.0% and specificity of 70.3%. When the tree model was applied to the test group, the accuracy, sensitivity and specificity were 76.6%, 76.0% and 76.9%, respectively. The performance of this mathematical model was substantiated further in 34 'problematic' stage I/pN0 patients by survival analysis.
Conclusions: The RP-DT model, constructed with eight protein markers for estimating lymphatic metastasis risk in pN0 lung SCC, is clinically feasible and practical, using IHC data from the primary tumour.
© 2011 Blackwell Publishing Limited.