Unsupervised analysis of electronic nose data identifies patient clusters in fibrosing interstitial lung disease

Introduction: Fibrosing interstitial lung disease (fILD) is a heterogeneous group of diseases with varying degrees of pulmonary inflammation and fibrosis. Electronic nose (eNose) analysis is an emerging method for profiling exhaled volatile organic compounds. We aimed to evaluate whether different fILD phenotypes can be identified through unsupervised analysis of eNose data. Methods: In a cross-sectional single centre study, exhaled breath of fILD patients was analysed using an eNose (SpiroNose). Unsupervised partition-around-medoids cluster analysis was applied to eNose sensor data. Comparison of clinical parameters between clusters was performed using one-way ANOVA, Kruskal-Wallis and chi-square tests. Results: 304 patients were included and three distinct clusters were identified. In cluster 1 (n=112) connective tissue disease ILD was most prevalent, while in cluster 2 (n=145) and 3 (n=47) IPF predominated (Figure 1). Clusters also significantly differed regarding gender (p=0.024), immunosuppressant use (p=0.004), and DLCO (p=0.038). Conclusions: Three fILD clusters were identified through analysis of eNose driven exhaled breath profiling. The distribution of fILD diagnoses and immunosuppressant use within the clusters suggest that breath profiles are influenced by inflammation. Further research should elucidate whether eNose data can guide personalised therapy in fILD.