Diagnosis of respiratory conditions using exhaled breath condensate using Inflammacheck® and advanced analytics: insights from the VICTORY study

Lung cancer, the third leading cause of death in England, is challenging to diagnose early. Traditional methods are costly, time-consuming and uncomfortable. Exhaled breath condensate (EBC) analysis with the Inflammacheck® device offers a non-invasive alternative, employing advanced analytics like t-distributed stochastic neighbour embedding (t-SNE), Bhattacharyya distances and network maps to differentiate respiratory conditions. The VICTORY study recruited participants (age ⩾ 16) with physician-confirmed respiratory conditions (asthma, chronic obstructive pulmonary disease, bronchiectasis, interstitial lung disease, lung cancer, pneumonia or a breathing pattern disorder) from inpatient and outpatient settings at a single NHS university hospital. EBC was collected using the Inflammacheck® device, to assess seven parameters: H₂O₂levels, peak CO₂percentage, peak breath humidity, peak breath temperature, exhalation flow rate, exhalation duration and sample collection time. After standardisation of EBC data, t-SNE was employed, Bhattacharyya distances calculated on tSNE components, network maps generated, and hierarchical clustering performed to illustrate the distinct classifications of the respiratory conditions based on the EBC parameters. The study included 282 participants. Multinomial logistic regression revealed elevated exhaled H₂O₂increased the odds of pneumonia (25.7-fold) and lung cancer (3.6-fold). t-SNE analysis showed distinct disease clusters, with Bhattacharyya distances for lung cancer and pneumonia demonstrating good separability from other conditions. Hierarchical clustering confirmed clear group distinctions, as visualised in heatmaps and dendrograms. The integration of advanced dimensionality reduction techniques t-SNE, combined with Bhattacharyya distance-based network mapping to interpret the EBC results facilitated discrimination between respiratory diseases. These methods were chosen over standard machine-learning classifiers due to their ability to provide intuitive, interpretable visualisations of complex data relationships, complementing their strong discriminatory power. Harnessing these analytical tools facilitated disease discrimination, particularly for lung cancer and pneumonia, suggesting promise as a diagnostic aid, paving the way for improved clinical decision-making and patient care.