Key points are not available for this paper at this time.
The aim of the present study was to investigate the ability of breath analysis to distinguish lung cancer (LC) patients from patients with other respiratory diseases and healthy people. The population sample consisted of 51 patients with confirmed LC, 38 patients with pathological computed tomography (CT) findings not diagnosed with LC, and 53 healthy controls. The concentrations of 19 volatile organic compounds (VOCs) were quantified in the exhaled breath of study participants by solid phase microextraction (SPME) of the VOCs and subsequent gas chromatography-mass spectrometry (GC-MS) analysis. Kruskal-Wallis and Mann-Whitney tests were used to identify significant differences between subgroups. Machine learning methods were used to determine the discriminant power of the method. Several compounds were found to differ significantly between LC patients and healthy controls. Strong associations were identified for 2-propanol, 1-propanol, toluene, ethylbenzene, and styrene (p-values < 0.001-0.006). These associations remained significant when ambient air concentrations were subtracted from breath concentrations. VOC levels were found to be affected by ambient air concentrations and a few by smoking status. The random forest machine learning algorithm achieved a correct classification of patients of 88.5% (area under the curve-AUC 0.94). However, none of the methods used achieved adequate discrimination between LC patients and patients with abnormal computed tomography (CT) findings. Biomarker sets, consisting mainly of the exogenous monoaromatic compounds and 1- and 2- propanol, adequately discriminated LC patients from healthy controls. The breath concentrations of these compounds may reflect the alterations in patient's physiological and biochemical status and perhaps can be used as probes for the investigation of these statuses or normalization of patient-related factors in breath analysis.
Koureas et al. (Mon,) studied this question.