**Background:** Chronic liver disease affects over 30% of adults in some countries, yet more than 50% of cirrhosis cases are diagnosed at advanced stages when interventions are often ineffective. Current diagnostic tools—including liver biopsy, imaging, and serological markers—have limitations for population screening. Breath analysis of volatile organic compounds (VOCs) offers a non-invasive approach to assess hepatic function, as impaired liver clearance and protein synthesis alter the spectrum of compounds detectable in exhaled breath. Previous work by this group showed that limonene, an exogenous VOC metabolized by hepatic CYP enzymes, was elevated in cirrhosis and correlated with blood metrics of liver function. However, single-biomarker approaches may not capture the complexity of hepatic metabolic dysfunction.
**Methods:** This cross-sectional case-control study enrolled 46 patients with cirrhosis (Child-Pugh class A or B, 14 with HCC) and 42 controls with no known liver disease. Breath samples were collected using the ReCIVA Breath Sampler with purified inhaled air to minimize environmental contamination. Samples were analyzed via Breath Biopsy OMNI global VOC analysis using gas chromatography-mass spectrometry (GC-MS). Feature extraction was performed using MZmine 2.53 with the ADAP chromatogram builder. A total of 2,593 molecular features were initially detected; after filtering for missing values (<50%) and excluding environmental contaminants (by comparison with blank samples), 196 VOCs were retained for analysis. Univariate analysis (Mann-Whitney U-test) identified differentially abundant VOCs. Classification models were built using LASSO logistic regression with stability selection and validated by five-fold cross-validation (70%/30% train/test split). Correlations with blood biomarkers (bilirubin, albumin, INR) were assessed using Pearson correlation and canonical correlation analysis (CCA).
**Key Results:** Univariate analysis identified 29 VOCs significantly different between cirrhosis and controls (p<0.05), with 15 upregulated and 2 downregulated (log fold-change ≥2). As expected, limonene and 2-pentanone were elevated, and dimethyl selenide was reduced in cirrhosis. A classification model based on these 29 VOCs achieved an AUC of 0.99±0.00 in training sets and 0.95±0.04 in cross-validated test sets. The confusion matrix showed 3 false positives (7%) and 8 false negatives (17%); among false negatives, 6 were Child-Pugh class A, 2 were class B, and 3 had HCC with class A. Step-forward feature selection showed that 7 VOCs were sufficient to maximize classification performance: 2-pentanone (AUC 0.82), 4-methyl-1-pentene or 1-hexene, indole, dimethyl selenide (AUC 0.76), limonene (AUC 0.79), eucalyptol (AUC 0.80), and benzene-(1-propylnonyl)-. Within the cirrhosis group, 11 VOCs with Benjamini-Hochberg adjusted p<0.1 were correlated with blood metrics. Limonene showed positive correlation with bilirubin (r=0.516, p=0.01) and negative correlation with albumin (r=-0.487, p=0.01). Dimethyl selenide had positive correlation with albumin (r=0.492, p=0.01). CCA demonstrated a significant collective correlation between breath VOCs and blood metrics (R²=0.842). PCA using the 11 VOCs separated patients by Child-Pugh score, though approximately 50% of patients with CP score of 5 overlapped with controls. PC1 explained 10.1% and PC2 explained 4.4% of variance.
**Clinical Implications:** This study demonstrates that a panel of 7 exhaled VOCs can discriminate cirrhosis from controls with high accuracy (AUC 0.95), substantially outperforming single-biomarker approaches (limonene alone: AUC 0.78). The correlation of breath VOCs with blood biomarkers of hepatic clearance and protein synthesis (bilirubin, albumin, INR) suggests that these VOCs reflect functional liver impairment rather than just anatomical changes. The ability to separate patients by Child-Pugh severity score supports potential for disease monitoring. While there is no clinical need for an additional diagnostic test for established cirrhosis, current modalities fail to detect early-stage disease—over 50% of patients are diagnosed after decompensation. A breath-based test could serve as a non-invasive screening tool in primary care or for home self-testing. Limitations include the small sample size, lack of blood data for controls, unstandardized dietary intake, and the exploratory nature of the study. The exogenous origin of most VOCs (e.g., limonene from diet) means that standardizing exposure could improve performance for detecting earlier-stage liver diseases such as NASH. Further validation in larger cohorts is needed.