How does breath analysis detect lung cancer?
Breath analysis works by detecting volatile organic compounds (VOCs) in exhaled breath. Cancer cells produce different metabolic byproducts than healthy cells, and these VOCs—like aldehydes, hydrocarbons, and ketones—can be measured in a person's breath. One study identified 16 specific VOCs that changed significantly after lung cancer surgery, confirming they come from the tumor [4]. Another study found elevated aldehyde levels in lung cancer patients' breath, likely due to overactive alcohol dehydrogenase pathways in cancerous tissues [10].
Several technologies are used to capture and analyze these VOCs. Mass spectrometry methods like PTR-TOF MS and GC-MS can identify individual compounds with high precision [2][4][6]. Electronic noses (eNoses) use sensor arrays to detect patterns of VOCs without identifying each one individually [3][5]. A newer approach uses a paper-based colorimetric sensor array with deep eutectic solvents, which achieved 93% accuracy in a small study [8]. Nanobiosensors are also being developed for real-time, point-of-care detection [12].
How accurate is it? The evidence is strong but mixed.
Several studies report very high accuracy. The largest study here, a cross-sectional study of 4,515 participants, used PTR-TOF MS with a machine learning model and achieved 95% sensitivity and 98% specificity for distinguishing lung cancer from healthy controls, and 97% sensitivity and 98% specificity for early-stage lung cancer vs. benign nodules [2]. A multi-center prospective external validation of an eNose in 364 participants found 94% sensitivity and 63% specificity across all participants [3]. Another study using perioperative breathomics in 525 participants (157 lung cancer, 368 healthy) achieved 89% sensitivity and 89% specificity with a 16-VOC model [4].
However, the largest multi-center prospective case-control study of breath VOCs for lung cancer diagnosis—the LuCID study with 1,844 participants—found that a 10-VOC panel performed poorly, with an AUC of only 0.54 for early-stage disease and 0.58 for all cases, which did not significantly outperform an epidemiological risk model [7]. This study highlights a critical challenge: many promising biomarkers from smaller studies fail to replicate in larger, more rigorous trials. The discrepancy likely stems from differences in study design, patient populations, and the difficulty of controlling for confounders like smoking, age, and other diseases [7][10].
What are the main challenges and what does the future look like?
The main challenges are standardization, validation, and real-world performance. Many studies are small and use different technologies, making it hard to compare results. The LuCID study, the largest and most rigorous here, showed that even well-conducted research can fail to validate earlier findings [7]. Water vapor in breath can interfere with sensors [11], and individual metabolism variations can affect VOC profiles [10]. One study found that an eNose could not distinguish lung cancer from healthy controls at all, while GC-MS could [11], showing that technology choice matters.
Despite these challenges, the field is advancing rapidly. Combining breath analysis with other data, like CT scan reports, improved accuracy in one study (87.7% vs. 68.1% for radiologists alone) [1]. Machine learning is being used to build more robust models [2][5][6]. New sensor technologies, like yolk-shell gold-MOF nanostructures, can detect VOCs at parts-per-billion levels [9]. The consensus from the most rigorous studies is that breath analysis has real potential as a non-invasive, low-cost screening tool, but it is not yet ready for widespread clinical use. Larger, multi-center trials with standardized protocols are needed to confirm which biomarkers and technologies truly work [7][12].
About These Sources
This answer is built on 12 peer-reviewed studies — published from 2021 to 2026, 9 from 2024 or later, 6 in Q1 journals, collectively cited 326 times — selected as the most relevant from 15 studies that passed quality screening, drawn from 81 papers retrieved from a database of over 500 million.
Sources used in this answer
Enhancing diagnosis of benign lesions and lung cancer through ensemble text and breath analysis: a retrospective cohort study
Combining breath analysis with CT report text analysis in 231 participants improved accuracy to 87.7% vs. 68.1% for radiologists alone, suggesting a multi-modal approach may be best.
Breathomics Analysis for Early Diagnosis of Lung Cancer Based on <scp>PTR</scp> ‐ <scp>TOF MS</scp> : A Large Sample Size Cross‐Sectional Study
In the largest study here (4,515 participants), PTR-TOF MS breath analysis with machine learning achieved 97% sensitivity and 98% specificity for early-stage lung cancer vs. benign nodules.
Lung cancer detection by electronic nose analysis of exhaled breath: a multicentre prospective external validation study
A multi-center prospective external validation of an eNose in 364 participants found 94% sensitivity and 63% specificity for lung cancer detection across all participants.
Identification of lung cancer breath biomarkers based on perioperative breathomics testing: A prospective observational study
Perioperative breathomics in 525 participants identified 16 VOCs that changed after surgery, with a diagnostic model achieving 89% sensitivity and 89% specificity.
Detection of lung cancer and stages via breath analysis using a self-made electronic nose device
A self-made eNose with 5 sensors in 261 participants achieved 83% sensitivity and 86% specificity for lung cancer detection in a validation phase.
Determination of lung cancer exhaled breath biomarkers using machine learning-a new analysis framework
GC-MS analysis of breath from lung cancer, TB, and control groups identified 10 VOCs; a PLS-DA model achieved 82% sensitivity and 80% accuracy for lung cancer vs. controls.
Multi-centre discovery and validation study evaluating breath biomarkers for the diagnosis of lung cancer – the LuCID study
The largest multi-center breath VOC study (1,844 participants) found that a 10-VOC panel performed poorly (AUC 0.54 for early-stage), not outperforming an epidemiological risk model.
Sniffing Out Lung Cancer: Biomimetic Breath Analysis via a Deep Eutectic Solvent-Driven Colorimetric Sensor Array.
A paper-based colorimetric sensor array using deep eutectic solvents in 91 participants achieved 93% accuracy, 98% sensitivity, and 98% specificity for lung cancer detection.
Yolk–Shell Hierarchical Pore Au@MOF Nanostructures: Efficient Gas Capture and Enrichment for Advanced Breath Analysis
A novel gold-MOF nanostructure sensor detected gaseous benzaldehyde at 0.32 ppb, demonstrating potential for ultra-sensitive breath analysis.
Breath Analysis for Lung Cancer Early Detection—A Clinical Study
A clinical study of 31 participants found that a combination of 8 VOCs (mainly aldehydes) achieved an AUC of 0.85 for lung cancer detection, with elevated levels linked to ADH pathways.
Exhaled breath analysis using GC-MS and an electronic nose for lung cancer diagnostics
GC-MS distinguished lung cancer from healthy controls with up to 96.5% sensitivity, but an eNose failed to differentiate groups, highlighting technology-dependent performance.
Nose-on-Chip Nanobiosensors for Early Detection of Lung Cancer Breath Biomarkers
A comprehensive review of nanobiosensor-based breath analysis concludes that integrating AI and IoT could enable real-time, point-of-care lung cancer screening.