Breath analysis of volatile organic compounds with mid-infrared laser spectroscopy in a cohort of lung cancer patients and control subjects.

Abstract

8029 Background: Exhaled volatile organic compounds (VOCs) represent potential non-invasive biomarkers for lung cancer detection. This study assessed a mid-infrared cavity ring-down spectroscopy (CRDS) platform for measuring key breath VOCs in a large clinical cohort including lung cancer patients across stages/histologies, carcinoma in situ, and non-cancer controls. Methods: This prospective single-center validation study enrolled 1,041 participants: approximately 500 with pathologically confirmed lung cancer (various stages and histologies), 25 with carcinoma in situ, healthy controls, patients with benign pulmonary diseases, and those with low- and high-risk pulmonary nodules. Breath samples were collected under standardized fasting and mouth-rinsing conditions and analyzed using a self-developed mid-infrared CRDS system targeting isoprene, acetone, and formaldehyde (detection limit < 10 ppb). The primary endpoint was diagnostic accuracy for distinguishing lung cancer (including carcinoma in situ) from healthy controls, measured by area under the receiver operating characteristic curve (AUC). Sample size was powered to detect an AUC difference of ≥0.15 from 0.50 (null hypothesis) with > 85% power at α = 0.05. Group comparisons used independent t-tests or Mann-Whitney U tests with Bonferroni correction for multiple testing. Predictive modeling employed logistic regression and machine learning (random forest and XGBoost) with 5-fold cross-validation; feature importance and SHAP values were assessed. Performance metrics included AUC, sensitivity, specificity, and 95% confidence intervals (CI). Results: Isoprene exhibited the most consistent group differences across comparisons (P < 0.001 in lung cancer vs healthy and carcinoma in situ vs healthy). Random forest and XGBoost models yielded AUC of 0.94 (95% CI, 0.92-0.96) for lung cancer vs healthy controls and 0.97 (95% CI, 0.95-0.99) for carcinoma in situ vs healthy controls. At optimal thresholds, sensitivity and specificity reached 90-96% and 92-95%, respectively, in primary analyses. Subgroup performance remained robust in non-smokers and stage I patients. Decision curve analysis demonstrated net clinical benefit across 1-20% risk thresholds compared with low-dose CT screening strategies. Acetone and formaldehyde provided supplementary contributions in pathological subtype and benign disease distinctions. Conclusions: Mid-infrared CRDS-based breath VOC analysis shows feasible diagnostic performance for lung cancer detection, including strong results in carcinoma in situ and early-stage disease, in this large cohort. Multi-center prospective validation is needed to confirm clinical utility and integration into screening strategies.