Out of the ivory tower: transitioning breathomics to clinical care

Extract

The clinical applicability of breathomics profiling (the study of volatile organic compounds (VOCs) in exhaled breath) in the early diagnosis of lung cancer remains tantalisingly out of reach. The lack of validation of breath analysis in studies with large patient populations through multicentric collaboration has impeded the standardisation of this approach in clinical practice. Prof. Lamote and his group should be acknowledged for their painstakingly decisive contributions to the literature on breathomics during the last decade. The article by Zwijsen et al. [1] in this issue of ERJ Open Research is no exception to the customary rigorous analyses and excellent insights from this group on the differentiation, through analysis of exhaled breath, of benign and malignant solitary pulmonary nodules. The aim of their study was to substantiate – once again – the value of breath analysis in the selection of candidates for lung screening, with the obvious relevance to cost reduction and efficiency. Despite some hurdles in the application of breath analytic technology [2], the accuracy of breath analysis in predicting lung malignancy has unquestionably been established [3]. The validated experiences of several groups in the detection of lung cancer with electronic nose (E-nose) technology [4] consistently demonstrate the ability of both quantitative and qualitative exhalate analytic methods to achieve high accuracy.

Shareable abstract

The time has come to transition from the academic stage of breathomics to the demonstration of its clinical applicability in the management of patients with lung cancer https://bit.ly/4hs8nk8

The clinical applicability of breathomics profiling (the study of volatile organic compounds (VOCs) in exhaled breath) in the early diagnosis of lung cancer remains tantalisingly out of reach. The lack of validation of breath analysis in studies with large patient populations through multicentric collaboration has impeded the standardisation of this approach in clinical practice. Prof. Lamote and his group should be acknowledged for their painstakingly decisive contributions to the literature on breathomics during the last decade. The article by Zwijsen et al. [1] in this issue of ERJ Open Research is no exception to the customary rigorous analyses and excellent insights from this group on the differentiation, through analysis of exhaled breath, of benign and malignant solitary pulmonary nodules. The aim of their study was to substantiate – once again – the value of breath analysis in the selection of candidates for lung screening, with the obvious relevance to cost reduction and efficiency. Despite some hurdles in the application of breath analytic technology [2], the accuracy of breath analysis in predicting lung malignancy has unquestionably been established [3]. The validated experiences of several groups in the detection of lung cancer with electronic nose (E-nose) technology [4] consistently demonstrate the ability of both quantitative and qualitative exhalate analytic methods to achieve high accuracy.

And yet, despite these encouraging results, the implementation of exhalate analysis into clinical practice still seems far away and it can feel as though we are only scraping the surface of what breathomics can do. There is no doubt that the breathomics community needs to transition from an academic research setting to the clinical use of the technology, in the name of standardisation and reproducibility in a real-world scenario [3, 5]. To achieve this transition, standardisation of exhalate sampling – of the analytic techniques used, and of blinding of investigators to patient demographic and clinical details – is necessary [3, 5]. In addition, data analysis and reporting must evolve. The different prevalences of disease (i.e. lung cancer) among different patient populations and the effect of prevalence on sensitivity and specificity must be acknowledged and described [5]. In this setting, whether we will need to rely on additional parameters to define the performance of breathomics methodologies remains to be clarified [5]. As an example, F₁ score and critical success index, which are currently used in artificial intelligence models, represent measurements that can correct imbalance between precision and recall among populations with different disease prevalences [5, 6].

Another step toward standardisation is to define criteria for the generation of disease-specific VOC libraries that can be used in both quantitative and qualitative analyses [7]. Once again, the selection of a subset of disease-associated VOCs specific to the studied patient population reveals the potential issue of different VOC profiles in different patient populations. Far from being a fixed, one-size-fits-all methodology, breath analysis should adapt to the demographic and epidemiological variations in different cohorts of patients with the same disease. Irrespective of the technology used for breath analysis – whether quantitative (based on multicapillary column/ion mobility spectrometry [1] as an evolution of gas chromatography/mass spectrometry) or qualitative sensor-based (i.e. E-nose) – a preliminary identification of population-specific, disease-associated VOC libraries is necessary before embarking on the clinical assessment of this technology. As Zwijsen et al. [1] correctly point out, quantitative analysis should move away from measuring all VOCs and, rather, focus on preidentified disease-associated VOCs, to increase predictivity. Ideally, the combination of quantitative and qualitative breath analysis will represent a standard [8] to provide lung cancer diagnostic accuracy, especially for ground-glass lesions, which are increasingly detected in patients with multifocal disease.

If breath analysis is to gain a prominent role in the clinical assessment of thoracic cancers, its reproducibility must be confirmed. Technologies such as multicapillary column/ion mobility spectrometry, used in the study by Zwijsen et al. [1], must be demonstrated to overcome the well-known flaws of the classic gas chromatography/mass spectrometry system, such as lack of user-friendliness, lack of portability and requirement for expensive equipment [7]. Another potential adverse factor complicating the achievement of reproducibility is the complexity involved in the use of machine learning as an analytic method in breathomics. Ensuring transparency of the machine-learning processes, to avoid creating a sense of arbitrariness, and sharing data and know-how among collaborating institutions may facilitate standardisation [3]. In this setting, any technical interference that potentially affects outcome analysis (i.e. sensor drift in E-nose systems) should be corrected [3].

Last, a renewed interest in nonvolatile organic compounds that are detectable in breath condensates or in filtered exhalates has been advocated in the recent literature to help define new biomarkers of clinical relevance [9]. However, the analysis of large molecules, such as host and pathogen proteins, lipids, metabolites and nucleic acids, can pose significant technical challenges and can be undermined by instrument limitations unless refined methods of metabolite profiling through ultrahigh-resolution separation and molecule identification are used (i.e. direct injection Fourier transform ion cyclotron resonance mass spectrometry) [9]. Longitudinal studies of the ability of E-nose technology to assess treatment response in patients with mesothelioma (www.clinicaltrials.gov identifier number NCT06037941) and locally advanced lung cancer (NCT04734145) are currently underway. The information resulting from the use of breathomics methodologies in these trials may pave the way for their extended use in lung cancer management, thereby easing the transition from the ivory tower of pure academic research to the high street of clinical practice.