Volatilome and machine learning in ischemic heart disease: Current challenges and future perspectives

Abstract

Integrating exhaled breath analysis into the diagnosis of cardiovascular diseases holds significant promise as a valuable tool for future clinical use, particularly for ischemic heart disease (IHD). However, current research on the volatilome (exhaled breath composition) in heart disease remains underexplored and lacks sufficient evidence to confirm its clinical validity. Key challenges hindering the application of breath analysis in diagnosing IHD include the scarcity of studies (only three published papers to date), substantial methodological bias in two of these studies, and the absence of standardized protocols for clinical implementation. Additionally, inconsistencies in methodologies—such as sample collection, analytical techniques, machine learning (ML) approaches, and result interpretation—vary widely across studies, further complicating their reproducibility and comparability. To address these gaps, there is an urgent need to establish unified guidelines that define best practices for breath sample collection, data analysis, ML integration, and biomarker annotation. Until these challenges are systematically resolved, the widespread adoption of exhaled breath analysis as a reliable diagnostic tool for IHD remains a distant goal rather than an imminent reality.

Key Words: Volatilome; Breathome; Ischemic heart disease; Mass spectrometer; Machine learning

Core Tip: Exhaled breath analysis offers a non-invasive, cost-effective alternative to traditional ischemic heart disease diagnostics, with superior accuracy (84% vs 60%–70% for stress electrocardiography). To enhance reliability, standardized protocols for breath collection and the integration of machine learning are essential. Collaborative efforts among clinicians, chemists, and data scientists are key to unlocking its full clinical potential.

TO THE EDITOR

Ischemic heart disease (IHD) and its associated sequelae remain the leading cause of global mortality and morbidity, posing a critical public health challenge despite advances in diagnostic and therapeutic strategies[1-4]. While emerging technologies, such as machine learning (ML)-enhanced medical applications, hold promise for refining diagnostic accuracy and treatment efficacy, significant gaps persist. A key barrier to the effective management of IHD is the limited availability of highly sensitive, widely accessible diagnostic tools, particularly in resource-constrained settings.

For the diagnosis of IHD, current guidelines from the European Society of Cardiology prioritize coronary computed tomography angiography and stress testing (e.g., pharmacological vasodilation with adenosine triphosphate) for moderate-to-high-risk patients[5]. However, these modalities face practical limitations: (1) They are costly, especially in poor countries; (2) They require specialized infrastructure; and (3) They depend on highly trained personnel to mitigate risks such as arrhythmias during vasodilator administration. 12-lead electrocardiography (ECG) with physical stress testing is a simple and cost-effective alternative but exhibits poor diagnostic accuracy, particularly in postmenopausal women, with high false-positive and false-negative rates.

To address these challenges, integrating exhaled breath analysis to identify disease-specific volatile organic compounds (VOCs) presents a novel, non-invasive strategy. This approach could complement existing methods by detecting metabolic biomarkers linked to ischemic pathophysiology, offering potential for an earlier and more precise diagnosis (Figure 1).

Figure 1 Integration of exhaled breath analysis into the current diagnostic pathway for ischemic heart disease. Clinical decision-making typically begins with an assessment of pre-test probability. Patients with < 5% risk undergo accessible but low-accuracy methods such as 12-lead electrocardiography physical stress tests and cardiovascular risk-scoring systems (SCORE2, SCORE2-OP, SMART). Patients with ≥ 5% risk are directed to high-accuracy modalities such as coronary computed tomography angiography, with or without pharmacological stress testing. However, these tests are resource-intensive, require specialized personnel, and are not universally available. Introducing exhaled breath analysis as an adjunct diagnostic tool can serve as a cost-effective alternative to improve the area under the receiver operating characteristic curve, thereby enhancing diagnostic and prognostic accuracy for ischemic heart disease. AUC: Area under the receiver operating characteristic curve; CCTA: Coronary computed tomography angiography; IHD: Ischemic heart disease.

CURRENT ACHIEVEMENTS AND FUTURE PERSPECTIVES

Current findings advocate advancing the use of exhaled breath analysis to diagnose IHD[6]. Although few studies have explored exhaled breath in this context, existing research demonstrates promising results, identifying VOCs linked to IHD, such as 2-pentanone or 3-methyl-2-butanone (C₄H₈O⁺), C₂H₇NO₃H⁺, and ions at m/z 87.9337 and 144.9178[7-9]. Notably, a recent prospective study highlights an 84% diagnostic accuracy (area under the curve) for breath-based detection, outperforming traditional 12-lead ECG stress tests (60%–70%)[10,12]. The study included 80 participants, divided into two groups according to the results of the stress computed tomography with myocardial perfusion (CTP) imaging with vasodilation test with adenosine triphosphate[10]. The first group (n = 31) included patients with IHD, confirmed by positive post-stress myocardial perfusion defect on the CTP. The second group (n = 49) included participants without IHD, confirmed by the negative post-stress myocardial perfusion defect on the CTP[10]. Participants provided breath samples using PTR-TOF-MS-1000 at three time points: at rest, at peak physical stress, and 3 min after the second breath sample. ML methods were employed to analyze the obtained concentrations of the compounds based on the mass/charge ratio, using an internal validation method[10].

Current evidence suggests these VOCs primarily reflect myocardial membrane breakdown rather than atherosclerotic plaque metabolism[13]. The temporal pattern of VOC release (peaking during exercise-induced ischemia) supports direct cardiac origin over systemic inflammation[8]. However, controlled studies using isotope-labeled lipid tracers are needed to confirm the biochemical pathways[14].

Breathomics presents a compelling, cost-effective alternative to invasive diagnostics such as coronary computed tomography angiography and stress ECG, particularly in resource-limited settings, by offering a non-invasive, low-risk approach that eliminates radiation exposure and procedural complications while enhancing patient compliance through simplicity and comfort[15–17]. Its portability and minimal infrastructure requirements—operability with basic training and without the need for expensive imaging equipment or reliable electricity—reduce both upfront and operational costs, making it scalable for rural or remote areas. Rapid results and screening potential enable timely triage and early intervention, supported by synergies with mobile health technologies for cloud-based data analysis and telemedicine integration. While further validation of diagnostic accuracy and addressing of regulatory and cultural adoption barriers remain critical, breathomics prioritizes affordability, accessibility, and sustainability, addressing systemic gaps in healthcare delivery for underserved populations.

Challenges in clinical implementation persist, including the absence of standardized protocols—from patient preparation (diet, medication, exhalation technique) to analytical methods (mass spectrometry workflows, ML models)[13,18–20]. The following critical questions remain unresolved: (1) Should only the final exhaled breath (reflecting lower bronchial regions) be analyzed? and (2) Which ML algorithms (e.g., support vector machine, neural networks) are optimal? Additionally, offline methods risk sample degradation during storage, whereas real-time (online) mass spectrometry enhances accuracy by preserving biochemical integrity.

The integration of ML in breathomics, such as random forest (RF) algorithms, has demonstrated proven utility in analyzing high-dimensional breathomics data, such as VOC profiles, owing to their inherent advantages in handling noisy, multi-dimensional datasets[21-24]. RF operates by constructing an ensemble of decision trees, each trained on random subsets of features and data, which reduces overfitting—a critical concern in breathomics where datasets are often small but rich in features (e.g., thousands of VOCs)[25-28]. Its ability to rank feature importance (e.g., identifying key biomarkers such as aldehydes or ketones linked to diseases such as lung cancer or chronic obstructive pulmonary disease) enables interpretable insights, even with limited clinical samples[29–32]. For instance, RF has been successfully applied in studies distinguishing asthma subtypes or early-stage cancers by prioritizing robust VOC signatures amid confounding variables (e.g., environmental VOCs)[33,34]. Unlike “black-box” models (e.g., deep neural networks), the computational efficiency and tolerance to missing data make RF particularly suitable for resource-limited settings, where rapid, low-power analysis is essential[35-38]. However, its efficacy hinges on standardized data collection (to minimize batch effects) and validation across diverse cohorts to ensure generalizability—a gap that must be addressed through collaborative benchmarking initiatives[29,31,39].

The clinical translation of breathomics faces critical challenges that must be addressed. Key hurdles include methodological bias in biomarker identification due to variability in sample collection (e.g., breath sampling techniques, environmental contaminants) and analytical platforms (e.g., mass spectrometry vs. sensor-based devices), which can skew results and limit reproducibility[40–43]. A lack of standardization across studies—in protocols for breath capture, storage, and data processing—further complicates cross-validation and comparability of findings[44]. Additionally, the influence of confounding factors such as diet, medications, and comorbidities on VOC profiles raises concerns about diagnostic specificity, potentially leading to false positives or negatives[45]. While portable devices and artificial intelligence-driven analytics offer scalability, the limited validation against gold-standard diagnostics (e.g., coronary computed tomography angiography) in diverse populations undermines clinical confidence, and regulatory frameworks lag in adapting to these novel technologies[46]. Finally, the integration of breathomics into existing healthcare workflows requires addressing data interpretation complexities (e.g., distinguishing disease-specific biomarkers from noise) and ensuring equitable access to avoid exacerbating health disparities[47]. Overcoming these challenges demands collaborative efforts to establish standardized protocols, robust validation studies, and adaptive regulatory pathways to realize the potential of breathomics as a transformative diagnostic tool.

To address methodological variability and accelerate the clinical translation of breathomics, a standardized minimal breath collection checklist—inspired by the European Respiratory Society and American Thoracic Society guidelines—should be implemented[48-50]. Key elements of patient preparation include fasting for ≥ 6 h, avoiding exercise/smoking, and documenting current medications. The exhalation technique should involve the use of a nose clip (or nasal inhalation followed by oral exhalation), with a slow vital capacity exhalation at 200–300 mL/s, discarding dead-space air. Sample capture should involve the use of inert containers such as Tedlar^® bags to collect mid-exhalation alveolar air at a standardized volume while controlling ambient VOCs and recording environmental metadata such as temperature and humidity. Post-collection protocols must prioritize immediate analysis or storage at −80 °C, include procedural blanks to detect contamination, and require rigorous documentation of patient demographics (e.g., body mass index, smoking status) and technical parameters (e.g., device type, operator ID). To ensure utility, this framework requires validation through multicenter trials assessing cross-platform reproducibility (e.g., gas chromatography–mass spectrometry vs e-noses vs PTR-TOF-MS-1000) and open-access protocol sharing via platforms such as European Multicenter Bronchiectasis Audit and Research Collaboration, coupled with regulatory collaboration to harmonize standards and address confounders, such as diet and comorbidities, that threaten diagnostic specificity.

A 2018 chronic obstructive pulmonary disease/asthma study using eNose technology demonstrated that breathomics could identify distinct inflammatory profiles independent of traditional diagnoses[51]. This shows that exhaled VOCs reflect systemic metabolic changes rather than just local lung pathology[52]. The same principle likely applies to cardiac conditions where systemic inflammation and oxidative stress drive pathology.

Research on 893 patients with chronic obstructive pulmonary disease revealed that breath analysis predicted the development of lung cancer within 2 years with 87% cross-validated accuracy (area under the receiver operating characteristic curve = 0.90)[53]. This proves that breath biomarkers can detect subclinical systemic metabolic shifts–a critical capability needed for early detection of IHD, where traditional tests often miss pre-symptomatic stages.

POTENTIAL ORIGIN OF THE OBTAINED MASS/CHARGE

The pathophysiological mechanisms underlying IHD provide a foundation for hypothesizing the origins of detected VOCs, such as C₄H₈O⁺ and ions at m/z 87.9337 and 144.9178, with three proposed hypotheses[32,54–56]. The first posits that myocardial ischemia triggers mitochondrial lipid peroxidation, generating reactive breakdown products such as C₄H₈O⁺ as byproducts of oxidative cellular damage. The second implicates atherosclerotic plaque formation in coronary arteries, where microbial colonization of vascular lesions may release distinctive metabolites detectable as exhaled VOCs. The third links gut microbiota dysbiosis to bacterial enzymatic activity, producing gases or metabolic byproducts that translocate systemically and exit via breath. While these hypotheses connect IHD pathophysiology to VOC signatures, rigorous validation is needed to confirm their biological sources and establish their specific roles in disease mechanisms.

The biomarker m/z 87.9337 (C₄H₈O⁺), likely reflecting VOCs such as butyraldehyde or fragmented ketone derivatives, may arise from oxidative stress and lipid peroxidation in IHD[57–59]. During ischemia-reperfusion injury, reactive oxygen species surge via mitochondrial dysfunction and enzymatic activation (e.g., xanthine oxidase), oxidizing polyunsaturated fatty acids in cell membranes to generate reactive aldehydes such as malondialdehyde or 4-hydroxynonenal, whose breakdown could yield smaller fragments such as C₄H₈O⁺. These aldehydes disrupt mitochondrial integrity, impair adenosine triphosphate production, and promote apoptosis while also modifying proteins/DNA, exacerbating cardiomyocyte dysfunction. Concurrently, lipid peroxidation products activate inflammatory pathways (e.g., nucleotide-binding domain, leucine-rich repeat, and pyrin domain-containing protein 3 inflammasomes) and endothelial dysfunction by scavenging nitric oxide, worsening ischemia. Clinically, detecting such VOCs in breath or blood could offer real-time insights into oxidative damage, while therapies targeting aldehyde detoxification (e.g., aldehyde dehydrogenase 2 activators) or antioxidant strategies may mitigate injury, bridging metabolic dysregulation to IHD pathophysiology for improved diagnostics and treatment[60].

CONCLUSION

The integration of exhaled breath analysis into clinical practice holds significant promise for revolutionizing the diagnosis, prognosis, and personalized treatment of IHD. Encouragingly, clinical trials have demonstrated its potential, achieving 84% diagnostic accuracy—surpassing traditional 12-lead ECG stress tests—and identifying key biomarkers such as C₄H₈O⁺ and ions at m/z 87.9337 and 144.9178. While challenges such as protocol standardization, ML model variability, and pre-analytical variables (e.g., breath collection methods) remain, the scientific community has begun to address these barriers through collaborative efforts in order to establish robust guidelines. As these hurdles are systematically resolved, exhaled breath analysis is poised to transition from experimental validation to routine clinical use, offering a non-invasive and cost-effective tool for early detection and tailored therapeutic strategies. The future of breath-based diagnostics hinges on multidisciplinary collaboration, bridging gaps between clinicians, chemists, and data scientists to unlock its full potential in improving cardiovascular care.