This systematised review evaluated the evidence for using breath analysis to detect disturbances in systemic glucose metabolism. The review targeted human studies published between 2010 and 2025 and identified 30 eligible studies. The clinical endpoints considered were diabetes, prediabetes, insulin resistance, gestational diabetes, hypoglycaemia, and the estimation of blood glucose concentration.
The authors framed breath analysis as a non-invasive diagnostic approach with several analytical pathways: laboratory-based chemical profiling, portable sensor arrays, single-analyte sensors, and isotope-labelled substrate tests. Across the included literature, study aims ranged from proof-of-concept biomarker discovery to early diagnostic performance assessments.
Three broad technological categories were reported.
Laboratory chemical profiling: Most studies analysed volatile organic compounds (VOCs) using laboratory platforms such as gas chromatography–mass spectrometry (GC-MS) or proton transfer reaction–mass spectrometry (PTR-MS). These platforms characterise complex breath mixtures and enable identification and quantification of multiple compounds.
Portable sensor-based systems: An increasing number of studies used portable devices, including electronic noses (sensor arrays with pattern recognition) and single-gas sensors designed to detect individual compounds. These systems prioritize point-of-care applicability and rapid readout but often provide less chemical specificity than laboratory methods.
Isotope-based breath tests: A smaller group of studies employed isotope-based approaches that measure labelled carbon dioxide in breath after ingestion of a labelled substrate. These tests probe metabolic pathways directly and were applied to discriminate metabolic states such as prediabetes and diabetes subtypes.
The reviewed studies covered multiple clinical targets:
Diabetes and prediabetes: Several studies sought to distinguish people with normoglycaemia from those with prediabetes or diabetes using breath VOC signatures or isotope tests.
Insulin resistance: A subset of investigations explored breath markers associated with insulin resistance, often overlapping with efforts to detect prediabetes.
Gestational diabetes: Breath analysis was evaluated in screening contexts for gestational diabetes in a smaller number of reports.
Hypoglycaemia: Several studies specifically targeted hypoglycaemia detection in type 1 diabetes mellitus, an important acute clinical application.
Blood glucose estimation: Some work attempted quantitative blood glucose prediction from breath measurements, a challenging translational target requiring accurate, continuous or near-continuous estimation.
Reported performance varied by clinical target and platform.
Hypoglycaemia detection in type 1 diabetes showed the most consistent high performance, with reported sensitivity and specificity approximately 90%–95% in the studies cited.
Isotope-based breath tests reported high discrimination for prediabetes and diabetes subtypes (approximately 90%–100% accuracy in reported studies). The review notes that these isotope-based results originate from a single research group and therefore require independent replication before wider acceptance.
Qualitative detection of diabetes (binary classification) and quantitative blood glucose prediction were more variable across studies and platforms. Performance was inconsistent when studies relied on single-compound markers rather than multivariable or spectral pattern analyses.
Overall, approaches that accounted for signal complexity—multivariable models and spectral/pattern recognition methods—outperformed single-compound strategies.
The review identified several recurring limitations that impede clinical translation:
Small cohorts: Most studies were conducted in limited sample sizes, restricting statistical power and the ability to generalise findings.
Internal validation only: The majority of studies used internal cross-validation without independent external validation cohorts, increasing the risk of optimistic performance estimates.
Poor standardisation: Sampling procedures, analytical workflows, and reporting standards were heterogeneous and often inadequately described, complicating comparison across studies.
Platform heterogeneity: Differences between laboratory-grade mass spectrometry platforms, portable sensor arrays, single-gas sensors, and isotope tests produce varying signal types and analytical challenges; not all platforms were matched to the intended clinical role in the studies reviewed.
Concentration on single research groups: Notably, promising isotope-based findings stemmed from a single group, underscoring the need for replication by independent teams.
The authors conclude that breath-based diagnostics show promising potential for specific clinical roles—most notably hypoglycaemia detection in type 1 diabetes and isotope-based metabolic discrimination—but significant steps remain before routine clinical implementation. Key requirements include:
Development and adoption of standardised sampling protocols, analytical workflows, and reporting guidelines to improve comparability and reproducibility across studies.
Larger cohorts with prospectively defined outcomes and independent external validation to confirm diagnostic performance and generalisability.
Careful selection of analytical platforms aligned with the intended clinical role (e.g., point-of-care detection vs laboratory-based metabolic phenotyping).
Independent replication of promising results, particularly those from single research groups (for example the isotope-based discrimination studies).
Until these requirements are met, breath analysis for detecting disturbances in systemic glucose metabolism remains an investigational approach with specific promising applications but insufficient evidence for broad clinical adoption.