Exhaled breath carries volatile organic compounds (VOCs) that report on lung cancer metabolism, and many single-center case-control studies have produced discriminating VOC-based models, yet none has translated into clinical practice. The source study (Tsou et al., Cancers 2021) — a single-center selected-ion flow-tube mass-spectrometry (SIFT-MS) breath analysis of 148 lung cancer patients and 168 controls with an XGBoost classifier reaching AUC 0.95 (0.98 after adjustment for environmental VOCs) — explicitly identified what remains unresolved: there is no validated list of lung cancer VOC biomarkers, sampling and analysis methods are so inconsistent across studies that results cannot be compared, dietary confounding of breath VOCs has no established persistence time or elimination protocol, and accuracy in young, early-stage, and operable patients awaits prospective multicenter validation. This problem asks for exactly that completion: a chemically identified, quantitatively defined VOC biomarker panel that is reproducible under a standardized breath collection and analysis protocol, robust to behavioral confounders such as diet, and carried through to a fully specified design for its prospective multicenter validation in cohorts including early-stage disease. Meta-analyses pool accuracy near 0.93 with substantial heterogeneity; only three VOCs have replicated across published predictive models; no universally accepted collection protocol exists; and the largest chemically resolved studies defer generalizable multicenter validation to future work, while the largest prospective multicenter mass-spectrometry validation to date (2026) reports rule-out-level specificity (about 56%), positioning its test as a triage adjunct rather than a validated biomarker panel.

Background

Lung cancer is the leading cause of cancer death worldwide, largely because it is usually detected at an advanced stage. Low-dose computed tomography (LDCT) screening reduces mortality but involves radiation exposure, incidental findings, and cost, so a non-invasive, inexpensive adjunct test is attractive. Breathomics is the analysis of the molecular composition of exhaled breath; its main targets are volatile organic compounds (VOCs), small carbon-containing molecules (alcohols, ketones, aldehydes, hydrocarbons, aromatic compounds) that are produced by metabolism, exchanged from blood into alveolar air, and exhaled. Tumor-related changes in oxidative stress, lipid peroxidation, cytochrome P450 activity, and glycolysis are hypothesized to alter breath VOC profiles (Hakim et al., Chemical Reviews 2012), although the biochemical origin of most disease-associated VOCs remains unproven. Breath VOCs are measured by gas chromatography-mass spectrometry (GC-MS, the classical separation-plus-identification method), selected ion flow tube mass spectrometry (SIFT-MS, a soft-ionization technique that quantifies preselected compounds in real time), proton-transfer-reaction mass spectrometry (PTR-MS, a related real-time quantitative method), or electronic noses (eNose), which are cross-reactive chemosensor arrays that classify breath odor patterns without chemically identifying individual compounds. Predictive models, typically built with machine-learning classifiers such as XGBoost (a gradient-boosted decision-tree ensemble), are then trained to separate patients from controls, with performance reported as sensitivity, specificity, and the area under the receiver operating characteristic curve (AUC). The literature since the early 2000s contains dozens of such studies, but almost all are single-center case-control studies on predominantly elderly, advanced-stage patients, and the compound sets they report barely overlap: Saalberg and Wolff (Clinica Chimica Acta 2016) compared published VOC lists and found largely inconsonant candidate biomarkers. Against this backdrop, Tsou et al. (Cancers 2021) performed the then most comprehensive quantitative study: SIFT-MS quantification of 116 VOCs in breath from 148 lung cancer patients and 168 healthy volunteers in Taiwan, combined with an XGBoost classifier achieving AUC 0.95, improving to 0.98 after adjusting for environmental VOC contamination. Their Discussion states the field's unresolved issues: the study was single-center and case-control with age mismatch, so the model may fail in young, early-stage, or operable patients, and a multicenter study was planned to give a definitive answer on early-detection accuracy; breath sampling procedures, study designs, and analysis methods differ so much across the literature that results are inconsistent; food alters breath VOCs through metabolic, inflammatory, and redox pathways, yet the persistence time of dietary VOCs in breath is unknown and there is no consensus on eliminating these effects; and there is no validated list of lung cancer VOC biomarkers, nor a settled mechanistic account of the VOC changes. Later literature sharpens rather than settles this picture. A meta-analysis of 25 studies (Fan et al., BMC Cancer 2024) pooled sensitivity 85% and specificity 86% (AUC 0.93) but flagged marked variability in study quality and methods. A systematic review of 46 published VOC-based predictive models for distinguishing benign pulmonary nodules from lung cancer (Su et al., BMC Pulmonary Medicine 2024) found that only 2-butanone, 3-hydroxy-2-butanone, and 2-hydroxyacetaldehyde replicated with consistent direction across more than two studies, with substantial heterogeneity in sampling and analysis. A systematic review of breath collection methods across 89 lung cancer VOC studies (He et al., Journal of Thoracic Disease 2024) concluded that no universally accepted collection protocol exists and that standardization is urgently needed. On the validation front, a genuine multicenter, multi-device validation of exhaled breath analysis for non-small cell lung cancer was completed with an eNose device (Kort et al., Chest 2023; sensitivity 95%, specificity about 50%, negative predictive value 94%), but eNose output is a sensor pattern, not a chemically identified VOC panel, and its role is rule-out triage rather than screening confirmation. Large recent studies with chemical identification — a prospective single-center study of pulmonary nodule malignancy (Lu et al., Cancer Medicine 2025) and a 4515-participant cross-sectional study using proton-transfer-reaction time-of-flight mass spectrometry (PTR-TOF-MS, a PTR-MS variant that resolves ions by time-of-flight detection) (Huang et al., Respirology 2026) — both explicitly defer multicenter validation to future work. The largest prospective multicenter validation of a mass-spectrometry breath test to date (Qin et al., Chest 2026) analyzed 5,214 symptomatic patients with radiological lung abnormalities by PTR-TOF-MS at two campuses with a geographically independent external validation cohort: a locked machine-learning model integrating a VOC signature with clinical factors reached external AUC 0.85 with 93% sensitivity but only about 56% specificity at a rule-out threshold, and is positioned as a triage tool — not as a validated, universally accepted VOC biomarker panel — leaving protocol standardization and dietary confounding unaddressed. On confounders, Krilaviciute et al. (Journal of Breath Research 2019) showed in 1447 average-risk individuals that coffee, leeks, and garlic consumption are associated with common breath VOCs, and Hageman et al. (Molecular Nutrition and Food Research 2019) traced postprandial VOC changes over five hours with large inter- and intra-individual variation, but no consensus persistence time or dietary-elimination protocol for the lung-cancer screening setting has emerged. Infrastructure for standardization is beginning to appear — a 2024 analysis using Owlstone Medical's Breath Biopsy OMNI thermal-desorption gas chromatography-mass spectrometry (TD-GC-MS) workflow produced a list of 148 chemically confirmed on-breath VOCs as a cross-study reference (Arulvasan et al., Metabolomics 2024) — but this list is not disease-specific and is explicitly a starting point for, not a substitute for, clinical biomarker validation.

Problem Statement

Establish a validated panel of breath volatile organic compound (VOC) biomarkers for lung cancer detection, together with the standardized methodology that makes it reproducible. Concretely, solving this problem requires, as one linked translational objective: (i) a chemically identified and quantitatively defined candidate list of lung-cancer-associated breath VOCs, drawn from compounds reported to replicate across published studies and to be tested for cross-study, cross-platform, and cross-population reproducibility in the validation designed in component (iv), rather than the current situation in which only three compounds have replicated across published predictive models; (ii) a standardized breath collection and analysis protocol (patient preparation, environmental and background control, breath fraction, storage, and analytical platform) under which the panel's measurements are reproducible across centers, replacing the current absence of any universally accepted collection protocol; (iii) a characterized and controlled treatment of behavioral confounders, in particular dietary effects — the reported persistence time of food-derived VOCs in breath synthesized from the cited studies, and a dietary-elimination protocol grounded in that evidence, on which no consensus currently exists; and (iv) a fully specified prospective multicenter validation design that would demonstrate the panel's accuracy for detecting lung cancer in the clinically decisive population the source study could not assess — young, early-stage, and operable patients — against a reference standard of histological diagnosis or follow-up, with target performance sufficient to define the test's clinical role (for example, screening triage alongside low-dose CT or non-invasive assessment of pulmonary nodule malignancy). These four components form one linked objective: a breath VOC test for lung cancer specified to the point of definitive multicenter evaluation, in which the panel is inseparable from the protocol and confounder control that make it reproducible, and from the fully specified multicenter early-stage validation design that would test it. Consistently, the deliverable is analytical and design-based: components (i)–(iii) are assembled from the existing published data and the cited literature, and component (iv) is a fully specified validation design rather than an executed trial. Executing new prospective data collection or new human-subject measurements is out of scope. The panel's validated status and its cross-study, cross-platform, and cross-population reproducibility are hypotheses that the designed validation would test, not properties the assembled deliverable must already have established.

References

  1. Tsou PH, Lin ZL, Pan YC, Yang HC, Chang CJ, Liang SK, Wen YF, Chang CH, Chang LY, Yu KL, Liu CJ, Keng LT, Lee MR, Ko JC, Huang GH, Li YK. Exploring Volatile Organic Compounds in Breath for High-Accuracy Prediction of Lung Cancer. Cancers 2021;13(6):1431. doi:10.3390/cancers13061431. https://doi.org/10.3390/cancers13061431
  2. Kort S, Brusse-Keizer M, Schouwink H, Citgez E, de Jongh FHC, van Putten JWG, van den Borne B, Kastelijn EA, Stolz D, Schuurbiers M, van den Heuvel MM, van Geffen WH, van der Palen J. Diagnosing Non-Small Cell Lung Cancer by Exhaled Breath Profiling Using an Electronic Nose: A Multicenter Validation Study. Chest 2023;163(3):697-706. doi:10.1016/j.chest.2022.09.042. https://doi.org/10.1016/j.chest.2022.09.042
  3. Huang Y, Li W, Yue H, Yang Y, Tian C, Duan Y, Huang C, Pu Q, Yuan S, Tian P, Li W. Breathomics Analysis for Early Diagnosis of Lung Cancer Based on PTR-TOF MS: A Large Sample Size Cross-Sectional Study. Respirology 2026;31(6):610-620. doi:10.1002/resp.70223. https://doi.org/10.1002/resp.70223
  4. Lu G, Su Z, Yu X, He Y, Sha T, Yan K, Guo H, Tao Y, Liao L, Zhang Y, Lu G, Gong W. Differentiating Pulmonary Nodule Malignancy Using Exhaled Volatile Organic Compounds: A Prospective Observational Study. Cancer Medicine 2025;14(1):e70545. doi:10.1002/cam4.70545. https://doi.org/10.1002/cam4.70545
  5. Su Z, Yu X, He Y, Sha T, Guo H, Tao Y, Liao L, Zhang Y, Lu G, Lu G, Gong W. Inconsistencies in predictive models based on exhaled volatile organic compounds for distinguishing between benign pulmonary nodules and lung cancer: a systematic review. BMC Pulmonary Medicine 2024;24(1):551. doi:10.1186/s12890-024-03374-2. https://doi.org/10.1186/s12890-024-03374-2
  6. He Y, Su Z, Sha T, Yu X, Guo H, Tao Y, Liao L, Zhang Y, Lu G, Lu G, Gong W. Collection methods of exhaled volatile organic compounds for lung cancer screening and diagnosis: a systematic review. Journal of Thoracic Disease 2024;16(11):7978-7998. doi:10.21037/jtd-24-1001. https://doi.org/10.21037/jtd-24-1001
  7. Fan X, Zhong R, Liang H, Zhong Q, Huang H, He J, Chen Y, Wang Z, Xie S, Jiang Y, Lin Y, Chen S, Liang W, He J. Exhaled VOC detection in lung cancer screening: a comprehensive meta-analysis. BMC Cancer 2024;24(1):775. doi:10.1186/s12885-024-12537-7. https://doi.org/10.1186/s12885-024-12537-7
  8. Arulvasan W, Chou H, Greenwood J, Ball ML, Birch O, Coplowe S, Gordon P, Ratiu A, Lam E, Hatch A, Szkatulska M, Levett S, Mead E, Charlton-Peel C, Nicholson-Scott L, Swann S, van Schooten FJ, Boyle B, Allsworth M. High-quality identification of volatile organic compounds (VOCs) originating from breath. Metabolomics 2024;20(5):102. doi:10.1007/s11306-024-02163-6. https://doi.org/10.1007/s11306-024-02163-6
  9. Krilaviciute A, Leja M, Kopp-Schneider A, Barash O, Khatib S, Amal H, Broza YY, Polaka I, Parshutin S, Rudule A, Haick H, Brenner H. Associations of diet and lifestyle factors with common volatile organic compounds in exhaled breath of average-risk individuals. Journal of Breath Research 2019;13(2):026006. doi:10.1088/1752-7163/aaf3dc. https://doi.org/10.1088/1752-7163/aaf3dc
  10. Hageman JHJ, Nieuwenhuizen AG, van Ruth SM, Hageman JA, Keijer J. Application of Volatile Organic Compound Analysis in a Nutritional Intervention Study: Differential Responses during Five Hours Following Consumption of a High- and a Low-Fat Dairy Drink. Molecular Nutrition and Food Research 2019;63(20):e1900189. doi:10.1002/mnfr.201900189. https://doi.org/10.1002/mnfr.201900189
  11. Saalberg Y, Wolff M. VOC breath biomarkers in lung cancer. Clinica Chimica Acta 2016;459:5-9. doi:10.1016/j.cca.2016.05.013. https://doi.org/10.1016/j.cca.2016.05.013
  12. Hakim M, Broza YY, Barash O, Peled N, Phillips M, Amann A, Haick H. Volatile Organic Compounds of Lung Cancer and Possible Biochemical Pathways. Chemical Reviews 2012;112(11):5949-5966. doi:10.1021/cr300174a. https://doi.org/10.1021/cr300174a
  13. Jia Z, Patra A, Kutty VK, Venkatesan T. Critical Review of Volatile Organic Compound Analysis in Breath and In Vitro Cell Culture for Detection of Lung Cancer. Metabolites 2019;9(3):52. doi:10.3390/metabo9030052. https://doi.org/10.3390/metabo9030052
  14. Zhou M, Wang Q, Lu X, Zhang P, Yang R, Chen R, Xia J, Chen D. Exhaled breath and urinary volatile organic compounds (VOCs) for cancer diagnoses, and microbial-related VOC metabolic pathway analysis: a systematic review and meta-analysis. International Journal of Surgery 2024;110(3):1755-1769. doi:10.1097/js9.0000000000000999. https://doi.org/10.1097/js9.0000000000000999
  15. Qin J, Sun M, Li X, Duan X, Li Y, Ma H, Peng H, Jia W, Li G, Gu S, Zhao X, Cui F, Pei M, Sun D. Development and Multi-center Validation of a Breathomics-Based Triage Tool for Lung Cancer: A Prospective Study of 5,214 Participants. Chest 2026 (online ahead of print). doi:10.1016/j.chest.2026.06.067. https://doi.org/10.1016/j.chest.2026.06.067