Harnessing exhaled breath for lung cancer early detection, results from the ExPeL study

Technical Deep Dive: Harnessing exhaled breath for lung cancer early detection

Overview

The ExPeL study evaluated a point-of-care device called Inflammacheck that uses exhaled breath condensate (EBC) analysis and machine learning to detect early-stage lung cancer in a real-world screening population. The key findings are:

• Inflammacheck accurately distinguished lung cancer patients from low-risk controls, achieving 85.7% accuracy, 80% sensitivity, 100% specificity, and 0.90 ROC AUC on a held-out test set.
• The model showed no false positives, which is crucial for primary care screening where low cancer prevalence makes false positives particularly problematic.
• Multivariate analysis of EBC variables (H2O2, end-tidal CO2, humidity, temperature, exhalation flow) revealed physiological heterogeneity in cancer patients missed by univariate analysis.
• Untargeted metabolomics identified 4 key metabolites that discriminated cancer cases with 0.969 AUC.

Problem & Context

Lung cancer is a major public health challenge, with poor outcomes due to late-stage diagnosis. Scalable, non-invasive screening tools are critically needed to improve early detection and optimize primary care referral pathways.

Methodology

The ExPeL study recruited participants from the UK Targeted Lung Health Check (TLHC) screening program, including individuals with suspected lung cancer and low-risk ever-smoker controls.

Exhaled breath condensate (EBC) was collected using the Inflammacheck device, which measured:

• H2O2 concentration
• End-tidal CO2
• Humidity
• Temperature
• Exhalation flow rate

Data & Experimental Setup

• 34 participants with valid EBC data were included in the analysis
• 83% of cancer cases were early-stage (I or II), reflecting a screening population

Results

Multivariate Analysis

• Principal component analysis (PCA), linear discriminant analysis (LDA), and Mahalanobis distance mapping clearly separated lung cancer patients from controls.

Machine Learning Model

• A voting ensemble model (stacked and voting ensembles) achieved:
  • Accuracy: 85.7%
  • Sensitivity: 80%
  • Specificity: 100%
  • Precision (PPV): 100%
  • ROC AUC: 0.90
  • Matthews Correlation Coefficient: 0.73
• Crucially, the model had no false positives.

Metabolomics

• Untargeted LC-MS metabolomics identified 2,132 features.
• 4 key metabolites yielded an AUC of 0.969 for cancer discrimination.

Interpretation

The Inflammacheck device effectively distinguished early-stage lung cancer patients from low-risk controls through a rapid, non-invasive breath test. This is highly relevant for primary care screening, where non-specific symptoms and low cancer prevalence pose challenges.

The multivariate analysis revealed physiological heterogeneity in cancer patients that was missed by univariate approaches, highlighting the value of the device's comprehensive EBC profiling.

Limitations & Uncertainties

• Small sample size (34 participants)
• Validation on a larger, independent dataset is needed to confirm the results.

What Comes Next

The authors suggest that the Inflammacheck device could be a valuable tool for primary care triage and screening, though further research is required to validate the findings in larger cohorts.

Sources:

• Harnessing exhaled breath for lung cancer early detection, results from the ExPeL study