Do linear logistic model analyses of volatile biomarkers in exhaled breath of cystic fibrosis patients reliably indicate Pseudomonas aeruginosa infection?
Abstract
Non-invasive breath analysis has been used to search for volatile biomarkers of lungs and airways infection by Pseudomonas aeruginosa, PA, in cystic fibrosis patients. The exhaled breath of 20 PA-infected patients and 38 PA-negative patients was analysed using selected ion flow tube mass spectrometry, SIFT-MS.
Special attention was given to the positive identification and accurate quantification of 16 volatile compounds (VOCs) as assured by the detailed consideration of their analytical ion chemistry occurring in the SIFT-MS reactor. However, the diagnostic sensitivity and specificity of the concentrations of any of the 16 compounds taken individually were found to be low.
But when a linear combination of the concentrations of all 16 VOCs was used to construct an optimised receiver operating characteristics (ROC) curve using a linear logistic model, the diagnostic separation of PA-infected patients relative to the PA-negative patients was apparently good in terms of the derived sensitivity (89%), specificity (86%), and the area under the ROC curve is 0.91. Four compounds were revealed by the linear logistic model as significant, viz. malondialdehyde, isoprene, phenol and acetoin.
The implications of these results to PA detection in the airways are assessed. Whilst such a metabolomics approach to optimise the ROC curve is widely used in breath analysis, it can lead to misleading indications.
Therefore, we conclude that the results of the linear logistic model analyses are of limited immediate clinical value. The identified compounds should rather be considered as a stimulus for further independent studies involving larger patient cohorts.