Breathomics in liver disease

Abstract

BACKGROUND: Non-alcoholic fatty liver disease (NAFLD) may progress to cirrhosis and end-stage liver disease. The prevalence is increasing in line with the global obesity epidemic. There is a need to develop a non-invasive diagnostic tool that could be used as a point of care. Exhaled breath contains a vast array of volatile organic compounds (VOCs) with potential for diagnostic exploitation.

METHODOLOGY: The study was a prospective single-centre cohort study (ClinicalTrials.gov: NCT02950610). In the present study, exhaled breath of 60 well-characterised NAFLD (cirrhotics and non-cirrhotic) patients was compared against that of self-declared healthy individuals. Gas chromatography - mass spectrometry and electronic signature of breath using a metal oxide semi-conductor sensor electronic nose was studied. Data were analysed using R studio (v 2.3.2) and SPSS 21. An unbiased machine learning clustering technique was applied. A 5-year longitudinal data was collected with endpoints of disease progression, liver disease-related complications and all-cause mortality.

RESULTS: Combined dimethyl sulphide (DMS) and D-limonene led to better discrimination of patients with NAFLD cirrhosis from healthy volunteers (AUROC 0.98; 95% CI 0.93–1.00; p <0.001) and patients with NAFLD cirrhosis from those with non-cirrhotic NAFLD (AUROC 0.91; 95% CI 0.82–1.00; p <0.001). Breath terpinene concentrations discriminated between patients with non-cirrhotic NAFLD and healthy volunteers (AUROC 0.84; 95% CI 0.68–0.99; p = 0.002). The eNose was able to differentiate between healthy from non-cirrhotic NAFLD (p<0.001, CVV 96.8%) and NAFLD cirrhotic (p<0.001, CVV 95.1%). An unbiased clustering technique further classified the patients into three distinct clusters. Cluster 1 consists of 23 patients, cluster 2 consists of 24 patients, and cluster 3 consists of 13 patients. The clusters were comparable in clinical phenotyping. Cluster 2 was identified as a higher risk group with significant differences in serum hyaluronic acid levels (p=0.001), portal hypertension(p=0.003) and dimethyl sulphide (p=0.041) and D-limonene (p=0.015). Cluster 2 was associated with a significant 5-year odds risk of 8.5 [95%CI 1.8 – 39.7] for disease progression. A 25% decompensation rate, 12.5% variceal bleeding and 12.5% all-cause mortality were noted in cluster 2 compared with 4.3% and 0% for decompensation. No variceal bleeding and 1% all-cause mortality was observed in cluster 1 and 3, respectively.

CONCLUSION: Electronic noses can differentiate between healthy and patient groups with high confidence. In addition, unbiased clustering within the NAFLD spectrum identifies three subtypes: mild, moderate, and severe disease phenotypes. These results warrant prospective studies on the potential of exhaled breath fingerprinting using eNose technology as point-of-care diagnostics and identifying high-risk disease progressors.