Breath carbonyl compounds as biomarkers of lung cancer.

OBJECTIVE Lung cancer dysregulations impart oxidative stress which results in important metabolic products in the form of volatile organic compounds (VOCs) in exhaled breath. The objective of this work is to use statistical classification models to determine specific carbonyl VOCs in exhaled breath as biomarkers for detection of lung cancer. MATERIALS AND METHODS Exhaled breath samples from 85 patients with untreated lung cancer, 34 patients with benign pulmonary nodules and 85 healthy controls were collected. Carbonyl compounds in exhaled breath were captured by silicon microreactors and analyzed by Fourier transform ion cyclotron resonance mass spectrometry (FT-ICR-MS). The concentrations of carbonyl compounds were analyzed using a variety of statistical classification models to determine which compounds best differentiated between the patient sub-populations. Predictive accuracy of each of the models was assessed on a separate test data set. RESULTS Six carbonyl compounds (C(4)H(8)O, C(5)H(10)O, C(2)H(4)O(2), C(4)H(8)O(2), C(6)H(10)O(2), C(9)H(16)O(2)) had significantly elevated concentrations in lung cancer patients vs. CONTROLS A model based on counting the number of elevated compounds out of these six achieved an overall classification accuracy on the test data of 97% (95% CI 92%-100%), 95% (95% CI 88%-100%), and 89% (95% CI 79%-99%) for classifying lung cancer patients vs. non-smokers, current smokers, and patients with benign nodules, respectively. These results were comparable to benchmarking based on established statistical and machine-learning methods. The sensitivity in each case was 96% or higher, with specificity ranging from 64% for benign nodule patients to 86% for smokers and 100% for non-smokers. CONCLUSION A model based on elevated levels of the six carbonyl VOCs effectively discriminates lung cancer patients from healthy controls as well as patients with benign pulmonary nodules.