Lung cancer detection by electronic nose: ready for prime time?

Introduction

Exhaled breath consists of a complex mixture of gasses, of which volatile organic compounds (VOCs) are formed metabolically both the in- and outside of the lungs as the result of a variety of biological processes (1). Metabolic changes resulting from a disease can therefore be measured in the exhaled breath and provide useful information for clinical diagnosis (1). Although the “smell” of breath as a key disease diagnostic analyte is already known in medicine for more than three thousand years, it has not been used very often so far (2). To date, the most well-known clinical breath tests applied are the exhaled nitric oxide test to detect eosinophilic airway inflammation in patients with asthma, the urea breath test to identify Helicobacter pylori infections, and the hydrogen breath test to diagnose lactose intolerance (1). For most diseases, however, measuring a single VOC has shown to insufficiently capture the characteristic breath signatures associated with the disease (2). Furthermore, individual VOC detection has shown to be expensive, time-consuming, and requires intensive equipment and expertise, thereby limiting its application in daily clinical practice (1).

In the last three decades, electronic nose (eNose) technologies have been developed that comprise of an array of electronic chemical sensors and an appropriate pattern-recognition system (3). Various sensors, such as metal-oxide semiconductor (MOS), quartz crystal microbalance (QCM), surface acoustic wave, electrochemical, and optical sensors, have been introduced in the hardware of the various eNose platforms, each having distinct properties resulting in specific advantages (4). Detection occurs when VOCs react with these specific sensors. For MOS sensors, for instance, this will cause conductivity changes, which are detected by transducers and converted into electrical signals that create specific VOC signatures (3,4). This way, the ultimate signal from the eNose platforms can be used to identify complex gas mixtures by comparing measured exhaled breath patterns with previously learned breath patterns; the so-called “breath prints” (2-4). Next to being non-invasive, eNoses also have the advantage of being easy to integrate, inexpensive, and easy-to-use, thereby fitting the profile of an ideal diagnostic biomarker (5). While in the food and environmental industry eNose technologies have already been implemented, in health care we are still struggling. Besides high standards that need to be met, the exact indication of this technology in the diagnostic pathways and positioning in relation to already applied diagnostic technologies has to be determined.

In the current health care system, more costly technologies are acceptable given specific indications (6). However, costs will become more relevant in the era where we redirect our health care system towards prevention. In this setting, eNose technologies might provide new opportunities. Screening programs will specifically require high throughput and low-cost technologies and here the eNose could start playing a role to screen populations at risk for the development of certain diseases. Moreover, also in the hospital setting there is an increased need for non-invasive diagnostic technologies that can be used next to radiological imaging to support clinical decision-making. In this editorial commentary, we will briefly summarize current evidence on lung cancer detection by eNose and discuss new evidence published by Rocco et al. on the use of eNose in early-stage lung cancer (7). Furthermore, we will elaborate on the specific requirements for implementing lung cancer detection by eNose technology in daily clinical practice, both in the hospital and screening setting.

Current technology and evidence

Up to now, various eNose devices have been tested in lung cancer (5). Especially with newer devices that integrate machine learning algorithms, high diagnostic accuracy rates (with sensitivity and specificity rates ranging from 71–96% to 33–100%, respectively) have been reported (5,8). Recent findings additionally suggest that lung cancer may be even accurately detected prior to diagnosis in high-risk groups (9). To determine the applicability of eNose analysis both in the hospital and screening setting, however, data specifically obtained in early-stage lung cancer are crucial.

The recently published paper by Rocco et al. addresses this issue in individuals who were referred to the outpatient clinic by various sources, including individuals who were recruited through a lung cancer screening program and individuals who showed incidental findings on radiological imaging performed for other indications (7). An eNose device based on QCM sensors was applied, utilizing anthocyanin-coated gold electrodes to detect mass changes induced by VOCs on the sensor surfaces. It was found that also in stage I disease, the diagnostic accuracy of lung cancer detection by eNose is high; eNose technology was able to identify 86 out of 88 (98%) patients who resulted to have lung cancer. Importantly, lung cancer detection was independent of tumor metabolism as assessed by positron emission tomography-computed tomography (PET-CT) imaging, tumor size (including 0–10 mm) and solidity, anatomic location, and histopathologic features (including tumor histology, lymphovascular invasion, and spread through airspaces), highlighting that eNose could accurately identify lung cancer irrespective of various intrinsic biological features. As a result, its false-negative rate (2%) compared favorably to commonly used radiological lung cancer risk prediction models (e.g., the Swenson and Brock model) (9–11%), first low-dose computed tomography (CT)-scans performed in lung cancer screening (6–72%), plasma biomarker assessments (40–78%), and tissue-acquisition methods (13–14%). In individuals who resulted to have benign pulmonary lesions, though, the eNose tended to overdiagnose lung cancer; 12 out of 12 (100%) individuals were classified as having an increased risk of having lung cancer. In a population with a high pre-test probability of having lung cancer (88%), the eNose thus effectively detected early-stage lung cancer, although this was accompanied by a relatively high rate of false-positive test results.

When used for clinical decision-making, the eNose showed a significant advantage over the Swenson and Brock model in individuals who resulted to have lung cancer; less patients would have been referred to observation and more patients would directly have been referred to surgery without the need for biopsy. In individuals who resulted to have benign pulmonary lesions, on the other hand, no significant advantage of eNose analysis over the radiological lung cancer risk prediction models was observed. For this subgroup it is important to note, however, that also the Swenson and Brock models did not perform well in preventing false-positive test results; both models classified the majority of the individuals as having an increased risk of having lung cancer (83% and 92%, respectively). The clear advantage of applying eNose analysis over the radiological lung cancer risk prediction models was therefore particularly seen in the subgroup of individuals who resulted to have lung cancer.

Test requirements according to diagnostic pathway

To establish the test requirements, it is first necessary to determine the indication and position of the technology in the diagnostic pathway.

The hospital setting

Individuals who are being referred with a suspicion of lung cancer have a high pre-test probability of having lung cancer; the majority is diagnosed with lung cancer. Histo- or cytopathology is generally recognized as the gold standard for the diagnosis of lung cancer (10). However, obtaining sufficient and representative sampling material has shown to be often challenging (11). A non-invasive technology that could be integrated in the diagnostic pathway to confirm the suspected diagnosis would therefore be of added value and additionally help decrease the time to start treatment. In this setting, the negative predictive value (NPV) of the diagnostic technology should thus be very high (≥95%) to minimize the rate of false-negative test results (7). The positive predictive value (PPV) is less critical, though it should at least be better than current practice to prevent a higher percentage of unnecessary invasive diagnostic procedures being performed in individuals with benign pulmonary lesions.

The screening setting

In lung cancer screening programs, the pre-test probability of having lung cancer is much lower (<2%) (12). In this setting, it is thus more relevant to accurately detect lung cancer (e.g., minimize the rate of false-positive test results) to prevent the performance of unnecessary invasive diagnostic procedures (7). Although lung cancer screening by low-dose computed tomography (LDCT) has shown to reduce lung cancer mortality, it is also associated with a relatively high rate of false-positive test results (13). A technology that could be used as second screen to support nodule management irrespective of nodule size to decrease the diagnostic burden would therefore be of added value. Risk stratification of pulmonary lesions is currently based on the British Thoracic Society (BTS), National Comprehensive Cancer Network (NCCN), American College of Chest Physicians (ACCP), and Fleischner guidelines, together with PET-CT imaging (14-17). This approach allows us to also take tumor growth and metabolism into account, thereby limiting the percentage of false-positive test results. When positioning the eNose following LDCT imaging, the PPV should thus be higher compared to current risk stratification methods in order to increase the effectiveness of the screening process, while the NPV is less relevant and will be determined by LDCT imaging.

When positioning the eNose prior to LDCT imaging to reduce unnecessary radiation exposure and potentially high costs, on the other hand, no preselection of screened individuals would be performed. This would result in an even lower pre-test probability of having lung cancer than when this technology is used as second screen (13). The PPV of the technology will therefore automatically be lower, while the NPV will automatically increase. Both values, however, should at least be as high as those of LDCT imaging to prevent a less effective screening process.

Is the diagnostic accuracy of current eNose platforms sufficient?

The diagnostic accuracy is not only dependent on the eNose technology used, but also on the tested population. Consequently, the established PPV and NPV can differ depending on the number of participants and prevalence of the disease in the tested population. Based on the high prevalence of lung cancer in the study by Rocco et al., accurate detection of early-stage lung cancer by eNose seems feasible in the hospital setting, although it is important to note that the results were obtained in a relatively small study population (7). The PPVs and NPVs reported by other studies that investigated the use of eNose technologies for the detection of lung cancer in relatively larger study populations (100–600 participants) with a high pre-test probability of having lung cancer (>40%), however, predominantly range from 54–78% and 83–92%, respectively. This suggests that current eNose technologies may not be accurate enough to be used as a stand-alone diagnostic test when applied for the general detection of lung cancer in the hospital setting (18-21). Furthermore, these studies and the study by Rocco et al. only included lung cancer patients and healthy controls. In the hospital setting, however, a significant proportion of individuals who have been referred due to suspicion of lung cancer already have a pre-existing cancer and/or will ultimately have another type of cancer than lung cancer. Since diagnostic accuracy rates obtained with eNose technology have shown to differ across cancer types, the PPV and NPV associated with the detection of (lung) cancer might result to be lower when including these individuals in the analyses (8). Confirmatory clinical utility studies should therefore first be performed in the hospital setting where the eNose technology is planned to be used, before implementation can be pursued.

Two studies investigating the use of eNose technology in high-risk groups found a higher true-negative rate (specificity) for detecting lung cancer with eNose compared to LDCT, suggesting that eNose analysis could be used as second screen to increase the effectiveness of lung cancer screening by LDCT (22,23). The reported PPVs and NPVs were 83–88% and 88–96%, respectively, however obtained in individuals with a high pre-test probability of having lung cancer and obtained in relatively small study populations (<120 participants) (22,23). It therefore first needs to be explored what the actual diagnostic accuracy of eNose analysis as first and second screen in larger study populations with a much lower pre-test probability of having lung cancer (<2%) is, before its clinical utility can be determined.

When assessing the clinical utility of eNose analysis for decision-making in both the hospital and screening setting, factors influencing the false-positive rate associated with radiological imaging [e.g., the presence of chronic obstructive pulmonary disease (COPD) and emphysema, which are both associated with an increased risk of developing lung cancer] should also be considered (13). For LDCT, the false-positive rate has shown to decrease if individuals who developed lung cancer within two years after the first moment of screening were also considered true-positives (13). This suggests that (a part of) the false-positive test results might potentially be the result of (a) precancerous lesion(s) present at the moment of the diagnostic assessment. Recently, de Vries et al. has shown that eNose technology could also be used to accurately identify patients with COPD in whom lung cancer becomes clinically manifest within 2 years [area under the curve (AUC) of 0.90, sensitivity of 86%, and specificity of 89%] (9). Clinical utility assessments of eNose analysis should therefore also include the potential benefit of detecting those cases who ultimately develop (lung) cancer within 2 years after the first diagnostic assessment.

Both in the hospital and screening setting, the majority of assessed diagnostic biomarkers (e.g., cell-free tumor DNA, blood proteins, DNA methylation) have shown to be suboptimal for detecting cancer at early disease stages (24,25). Results obtained by Rocco et al., however, suggest that eNose technology could also be used to detect lung cancer even in pulmonary lesions <8 mm in size, for which observation or no follow-up is currently indicated according to the number, size, and solidity of the lesion(s), and the presence of risk factors (7). Clinical utility analyses performed both in the hospital and screening setting should therefore also determine the potential benefit of eNose analysis in accelerating the diagnostic process for those cases with small lesions where current diagnostic technology is imminent and challenging.

Conceptual thinking: what is needed for acceptance of “black box” technology?

For a diagnostic technology to become accepted, clear data on what we measure is considered necessary. In the case of eNose technology, however, this is not straightforward. We do not have a full understanding of the biological processes underlying the development of a malignancy in the first place; e.g., there are many interactions between cancer cells, the inflammatory system, and the microbiome. Accumulations of minor changes in this complex biological interplay can finally result in the development of cancer and alter the composition of the exhaled breath (26). ENose analysis is, at present, not the technology to elucidate these details. However, it can provide us the likelihood of a malignancy or another disease being present or to develop in the near future. To attempt to determine which VOCs drive lung cancer detection by eNose, studies should also characterize the individual VOCs alongside with the eNose profiles. Once sufficient combined data is gathered and the eNose has shown to be adequately accurate for lung cancer detection in confirmatory clinical utility studies, then we expect the technology to become accepted for use in health care.

A high probability exists, though, that no individual VOC will be found that exactly reflects the breath signature as measured by eNose. To date, studies using methods for individual VOC detection have shown that no single individual compound was able to discriminate lung cancer patients from healthy controls (9,27). Furthermore, a lack of conformity in the list of detected individual compounds exists across the different studies (27). Since most of the VOCs altered in cancer patients have shown to derive from specific subgroups of molecules (e.g., alkanes, alcohols, aldehydes, ketones, nitriles, and aromatic compounds), focusing on molecular subgroups instead of individual VOCs might more reliably reflect the complexity of changes in the exhaled breath of (lung) cancer patients as detected by eNose (27,28).

Conclusions

Increasing evidence supports the potential of eNose technology for the accurate detection of lung cancer (Table 1). In the hospital setting, however, confirmatory clinical utility studies performed in the setting where the eNose technology is planned to be used, are lacking. In the screening setting, studies exploring the diagnostic performance of the technology as first and second screen in populations with a low pre-test probability of having lung cancer (<2%) should first be performed before determining the potential benefit of the technology over and in combination with currently used diagnostic risk stratification methods. Performed studies should both characterize the individual VOCs alongside with the eNose profiles to further enable implementation of lung cancer detection by eNose technology in daily clinical practice.

Acknowledgments

None.

Footnote

Provenance and Peer Review: This article was commissioned by the editorial office, AME Clinical Trials Review. The article has undergone external peer review.

Peer Review File: Available at https://actr.amegroups.com/article/view/10.21037/actr-24-257/prf

Funding: None.

Conflicts of Interest: Both authors have completed the ICMJE uniform disclosure form (available at https://actr.amegroups.com/article/view/10.21037/actr-24-257/coif). M.M.V.D.H. reports research funding from Roche Genentech, AstraZeneca, BMS, Boehringer Ingelheim, Merck, Novartis, Pamgene, Pfizer, Roche diagnostics, Stichting Treatmeds, and Takeda; speaker fees from Amgen, Astrazeneca, BMS, Janssen Pharmaceutica, Merck, MSD, Novartis, Pfizer, and Roche; serves on advisory boards for Abbvie, Amgen, AstraZeneca, Bayer, BMS, Boehringer Ingelheim, Janssen, Lilly, Merck, MSD, Novartis, Pfizer, Roche, Sanofi, and Takeda; serves as the chair of NVALT studies foundation and the chair of section oncology NVALT; and is the local PI for clinical trials sponsored by AstraZeneca, BMS, GSK, Novartis, Merck, Roche, Takeda, Mirati, Abbvie, MSD, Merck, Amgen, Boehringer Ingelheim, and Pfizer. The other author has no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Chen T, Liu T, Li T, et al. Exhaled breath analysis in disease detection. Clin Chim Acta 2021;515:61-72. [Crossref] [PubMed]
2. Sharma A, Kumar R, Varadwaj P. Smelling the Disease: Diagnostic Potential of Breath Analysis. Mol Diagn Ther 2023;27:321-47. [Crossref] [PubMed]
3. Licht JC, Grasemann H. Potential of the Electronic Nose for the Detection of Respiratory Diseases with and without Infection. Int J Mol Sci 2020;21:9416. [Crossref] [PubMed]
4. Li Y, Wei X, Zhou Y, et al. Research progress of electronic nose technology in exhaled breath disease analysis. Microsyst Nanoeng 2023;9:129. [Crossref] [PubMed]
5. van Geffen WH, Lamote K, Costantini A, et al. The electronic nose: emerging biomarkers in lung cancer diagnostics. Breathe (Sheff) 2019;15:e135-41. [Crossref] [PubMed]
6. Ben-Aharon O, Iskrov G, Sagy I, et al. Willingness to pay for cancer prevention, screening, diagnosis, and treatment: a systematic review. Expert Rev Pharmacoecon Outcomes Res 2023;23:281-95. [Crossref] [PubMed]
7. Rocco G, Pennazza G, Tan KS, et al. A Real-World Assessment of Stage I Lung Cancer Through Electronic Nose Technology. J Thorac Oncol 2024;19:1272-83. [Crossref] [PubMed]
8. Scheepers MHMC, Al-Difaie Z, Brandts L, et al. Diagnostic Performance of Electronic Noses in Cancer Diagnoses Using Exhaled Breath: A Systematic Review and Meta-analysis. JAMA Netw Open 2022;5:e2219372. [Crossref] [PubMed]
9. de Vries R, Farzan N, Fabius T, et al. Prospective Detection of Early Lung Cancer in Patients With COPD in Regular Care by Electronic Nose Analysis of Exhaled Breath. Chest 2023;164:1315-24. [Crossref] [PubMed]
10. Tseng LJ, Matsuyama A, MacDonald-Dickinson V. Histology: The gold standard for diagnosis? Can Vet J 2023;64:389-91. [PubMed]
11. van den Heuvel G, Mandos B, Schuurbiers O, et al. Stage migration in resectable NSCLC. Ann Oncol 2024;35:S775-S93.
12. Krilaviciute A, Brenner H. Low positive predictive value of computed tomography screening for lung cancer irrespective of commonly employed definitions of target population. Int J Cancer 2021;149:58-65. [Crossref] [PubMed]
13. Hammer MM, Byrne SC, Kong CY. Factors Influencing the False Positive Rate in CT Lung Cancer Screening. Acad Radiol 2022;29:S18-22. [Crossref] [PubMed]
14. Callister ME, Baldwin DR, Akram AR, et al. British Thoracic Society guidelines for the investigation and management of pulmonary nodules. Thorax 2015;70:ii1-ii54. [Crossref] [PubMed]
15. Wood DE, Kazerooni EA, Aberle D, et al. NCCN Guidelines® Insights: Lung Cancer Screening, Version 1.2022. J Natl Compr Canc Netw 2022;20:754-64. [Crossref] [PubMed]
16. Expert Panel on Thoracic Imaging. ACR Appropriateness Criteria® Incidentally Detected Indeterminate Pulmonary Nodule. J Am Coll Radiol 2023;20:S455-70. [Crossref] [PubMed]
17. Farjah F, Monsell SE, Smith-Bindman R, et al. Fleischner Society Guideline recommendation for incidentally detected pulmonary nodules and the probability of lung cancer. J Am Coll Radiol 2022;19:1226-35. [Crossref] [PubMed]
18. van de Goor R, van Hooren M, Dingemans AM, et al. Training and Validating a Portable Electronic Nose for Lung Cancer Screening. J Thorac Oncol 2018;13:676-81. [Crossref] [PubMed]
19. Chen KC, Kuo SW, Shie RH, et al. Advancing accuracy in breath testing for lung cancer: strategies for improving diagnostic precision in imbalanced data. Respir Res 2024;25:32. [Crossref] [PubMed]
20. Kort S, Tiggeloven MM, Brusse-Keizer M, et al. Multi-centre prospective study on diagnosing subtypes of lung cancer by exhaled-breath analysis. Lung Cancer 2018;125:223-9. [Crossref] [PubMed]
21. Kort S, Brusse-Keizer M, Schouwink H, et al. Diagnosing Non-Small Cell Lung Cancer by Exhaled Breath Profiling Using an Electronic Nose: A Multicenter Validation Study. Chest 2023;163:697-706. [Crossref] [PubMed]
22. Rocco R, Incalzi RA, Pennazza G, et al. BIONOTE e-nose technology may reduce false positives in lung cancer screening programmes. Eur J Cardiothorac Surg 2016;49:1112-7. [Crossref] [PubMed]
23. Shlomi D, Abud M, Liran O, et al. Detection of Lung Cancer and EGFR Mutation by Electronic Nose System. J Thorac Oncol 2017;12:1544-51. [Crossref] [PubMed]
24. Dama E, Colangelo T, Fina E, et al. Biomarkers and Lung Cancer Early Detection: State of the Art. Cancers (Basel) 2021;13:3919. [Crossref] [PubMed]
25. Marmor HN, Zorn JT, Deppen SA, et al. Biomarkers in Lung Cancer Screening: a Narrative Review. Curr Chall Thorac Surg 2023;5:5. [Crossref] [PubMed]
26. Hanahan D. Hallmarks of Cancer: New Dimensions. Cancer Discov 2022;12:31-46. [Crossref] [PubMed]
27. Dent AG, Sutedja TG, Zimmerman PV. Exhaled breath analysis for lung cancer. J Thorac Dis 2013;5:S540-50. [PubMed]
28. Einoch Amor R, Nakhleh MK, Barash O, et al. Breath analysis of cancer in the present and the future. Eur Respir Rev 2019;28:190002. [Crossref] [PubMed]