A Prediction Model with a Combination of Variables for Diagnosis of Lung Cancer.

BACKGROUND Multivariate models with a combination of variables can predict disease more accurately than a single variable employed alone. We developed a logistic regression model with a combination of variables and evaluated its ability to predict lung cancer. MATERIAL AND METHODS The exhaled breath from 57 patients with lung cancer and 72 healthy controls without cancer was collected. The VOCs of exhaled breath were examined qualitatively and quantitatively by a novel electronic nose (Z-nose4200 equipment). The VOCs in the 2 groups were compared using the Mann-Whitney U test, and the baseline data were compared between the 2 groups using the chi-square test or ANOVA. Variables from VOCs and baseline data were selected by stepwise logistic regression and subjected to a prediction model for the diagnosis of lung cancer as combined factors. The receiver operating characteristic (ROC) curve was used to evaluate the predictive ability of this prediction model. RESULTS Nine VOCs in exhaled breath of lung cancer patients differed significantly from those of healthy controls. Four variables - age, hexane, 2,2,4,6,6-pentamethylheptane, and 1,2,6-trimethylnaphthalene - were entered into the prediction model, which could effectively separate the lung cancer samples from the control samples with an accuracy of 82.8%, a sensitivity of 76.0%, and a specificity of 94.0%. CONCLUSIONS The profile of VOCs in exhaled breath contained distinguishable biomarkers in the patients with lung cancers. The prediction model with 4 variables appears to provide a new technique for lung cancer detection.