Probability of mediastinal involvement in non–small-cell lung cancer: a statistical definition of the clinical target volume for 3-dimensional conformal radiotherapy?

Abstract

Conformal irradiation (3D-CRT) of non-small-cell lung carcinoma (NSCLC) is largely based on precise definition of the nodal clinical target volume (CTVn). A reduction of the number of nodal stations to be irradiated would facilitate tumor dose escalation. The aim of this study was to design a mathematical tool based on documented data to predict the risk of metastatic involvement for each nodal station. We reviewed the large surgical series published in the literature to identify the main pretreatment parameters that modify the risk of nodal invasion. The probability of involvement for the 17 nodal stations described by the American Thoracic Society (ATS) was computed from all these publications. Starting with the primary site of the tumor as the main characteristic, we built a probabilistic tree for each nodal station representing the risk distribution as a function of each tumor feature. Statistical analysis used the inversion of probability trees method described by Weinstein and Feinberg. Validation of the software based on 134 patients from two different populations was performed by receiver operator characteristic (ROC) curves and multivariate logistic regression. Analysis of all of the various parameters of pretreatment staging relative to each level of the ATS map results in 20,000 different combinations. The first parameters included in the tree, depending on tumor site, were histologic classification, metastatic stage, nodal stage weighted as a function of the sensitivity and specificity of the diagnostic examination used (positron emission tomography scan, computed tomography scan), and tumor stage. Software is proposed to compute a predicted probability of involvement of each nodal station for any given clinical presentation. Double cross validation confirmed the methodology. A 10% cutoff point was calculated from ROC and logistic model giving the best prediction of mediastinal lymph node involvement. To more accurately define the CTVn in NSCLC three-dimensional conformal radiotherapy, we propose a software that evaluates the risk of mediastinal lymph node involvement from easily accessible individual pretreatment parameters.