Breath print for asthma phenotyping

Breath print for asthma phenotyping

Delphine Zanella1* ; Pierre-Hugues Stefanuto1 ; Romain Pesesse1 ; Florence Schleich2 ; Renaud Louis2 ; Jean-François Focant1

1 Organic and Biological Analytical Chemistry Group, University of Liège, Belgium 
2 Respiratory Medicine, GIGA I3, CHU Sart-Tilman, Belgium

Asthma is one of the most prevalent chronic disorder worldwide, affecting 235 million people. This represents a serious public health issue associated with high health costs, mainly due to the diagnosis and treatment. A European study has estimated the total cost of asthma to 19.3 billion euros/year. Asthma is characterized by an inflammation of the airways, involving several different underlying mechanisms. Current therapies remain ineffective in a large proportion of patients. Therefore, the characterization of the different inflammation phenotypes (i.e. eosinophilic, neutrophilic, paucigranulocytic, mixed-granulocytic asthma) is of great importance to provide personalized treatment.  Volatile organic compounds from breath of 245 asthmatic patients covering a range of four different asthma phenotypes were analyzed. The breath samples were collected into Tedlar® bags. Thermal desorption coupled with comprehensive two-dimensional gas chromatography – high resolution time-of-flight mass spectrometry was applied for the analysis. The data were split between training and test (60-40%). Random forest algorithm was used to investigate the ability of exhaled breath VOCs to distinguish between the inflammatory profiles.  The random forest algorithm was built on 7 significant features highlighted in a first discovery study. ROC curve were constructed to evaluate the classification performance in pair-classification scenario. The AUROC classifications reached 0.71, 0.68 and 0.70 with 70%, 60% and 64% of accuracy.   This first large-scale confirmatory study permitted the discrimination of patients according to their respective phenotypes. The present study shows that breath VOCs analysis is an efficient approach for asthma phenotyping and is going to lead to further development in clinical diagnosis.