Exhaled breath volatiles for asthma diagnosis: discovery and validation in untreated but symptomatic patients

Abstract Approximately 30% of people with a clinical diagnosis of asthma do not have the condition. Exhaled volatile organic compounds (VOCs) hold promise for non-invasive asthma testing; however, discovery of breath biomarkers is confounded by high biological and technical variability. We collected repeated breath samples from symptomatic patients being tested for asthma before starting treatment. Background samples were collected immediately before patient sampling to characterize VOCs present in the preconditioned inspired airstream. VOCs were measured by thermal desorption-gas chromatography-mass spectrometry. Data from two cohorts (n = 62, 53) were processed independently and used as training and validation datasets. VOCs were classified as breath-enriched or ambiguous based on comparison between breath and background abundance using Wilcoxon test and log fold change. Associations between breath VOCs and diagnosis were tested with univariate mixed-effect models and multivariate classification models. Here we show that background has significant effect on breath VOC abundance in 60% of identified compounds. Breath VOCs show high inter-session and inter-patient variability, reflected by increase in multivariate model classification error in cross-validation (37%) and validation (52%) cohorts. Among VOCs positively associated with asthma, 30% are breath-enriched. Ethyl butanoate, 2-methylfuran and 3-methylpentane show consistent discrimination performance across the cohorts. When used in conjunction with clinical tests, these breath VOCs have comparable contribution to asthma prediction accuracy to established clinical diagnostics such as fractional exhaled nitric oxide. Measuring exhaled VOCs could add value to diagnosis of asthma based on routine clinical tests. However, developing measures to limit technical variation in breath analysis is needed to support discovery and clinical translation of such biomarkers.