|
Abstract INTRODUCTION:
Type 2 diabetes mellitus (T2DM) is an escalating global health concern, with over 463 million people affected worldwide as of 2019 — a number projected to reach 700 million by 2045. In Nepal, the burden is rising rapidly, with recent estimates suggesting that approximately 10% of the adult population now lives with T2DM. Factors such as urbanization, changing diets, sedentary lifestyles, and limited access to preventive care are contributing to the steep rise in prevalence across South Asia. Current diagnostic strategies—such as fasting blood glucose and HbA1c tests—require blood sampling and are often underutilized in low-resource settings due to infrastructure, cost, and patient discomfort. These methods also fail to detect early metabolic changes before clinical thresholds are crossed.
Exhaled breath contains volatile organic compounds (VOCs) that reflect underlying metabolic processes, and several studies have shown altered VOC profiles in individuals with diabetes. Acetone, for example, was first identified in diabetic breath as early as the 19th century and remains one of the most studied VOCs linked to glucose metabolism. Breath analysis represents a non-invasive, painless, and potentially real-time method of screening for metabolic disease. This study aims to evaluate the utility of breath VOC profiling for detecting pre-diabetes and T2DM in a South Asian cohort, using thermal desorption–gas chromatography–time-of-flight mass spectrometry (TD-GC-TOF-MS). The goal is to support the development of a portable, clinically relevant, and cost-effective diagnostic tool suitable for widespread screening in resource-constrained settings.
METHODS:
This cross-sectional validation study is set to recruit 240 participants from clinical sites in Nepal, including 80 healthy controls, 80 individuals with pre-diabetes, and 80 individuals with T2DM. This study is ensuring statistical power through a sample size calculation with a 95% confidence and accounting for 5% type 1 errors alongside the 50% prevalence, a total of 146 participants would be required. At least 73 diabetic participants and 73 controls would need to be recruited to ensure the power of the sensitivity and specificity of the study. Each participant will provide two exhaled breath samples, one fasting and one postprandial, collected using a BIO-VOC2 breath sampler. For each sampling, five full exhalations will be overlaid onto a single thermal desorption (TD) tube to enhance compound detection. After collection, TD tubes will be sealed and stored in a cooler at 4 °C, then transported under refrigerated conditions to the Biomedical Sensors Laboratory at the University of Warwick (UK) for analysis.
Breath VOC samples were analysed using a TD-GC-TOF-MS system. Chromatographic peaks were deconvoluted and integrated using Chrome Compare™ software. The resulting VOC data were analysed using multivariate statistical analysis and machine learning (ML) techniques to identify compound features capable of differentiating between glycaemic and healthy classifications. ML models included Random Forest, XGboost, and Logistic Regression. Feature selection was based on the Mann–Whitney U test, with Benjamini–Hochberg false discovery rate (FDR-BH) correction applied to control for multiple comparisons.
Ethical approval for this study was granted by the Nepal Health Research Council (NHRC), reference number 3833, and the Manmohan Memorial Medical College Research Committee (MMMC-RC), reference number 1471079180. Written informed consent was obtained from all participants prior to enrolment.
RESULTS:
Preliminary multivariate analysis was conducted on a subset of 132 participants: 51 healthy controls, 35 individuals with pre-diabetes, and 46 individuals with T2DM. These preliminary analyses revealed several VOCs with statistically significant differences between glycaemic classifications. In the Healthy vs pre-diabetic group, Longifolene achieved significance after multiple testing correction (adjusted p = 0.028). In the pre-diabetic vs T2DM group, Nonadecane (adjusted p = 0.017), Octadecane (p = 0.019), and Silanediol, dimethyl- (p = 0.025) also remained significant following FDR correction. While no VOCs reached statistical significance in the Healthy vs Diabetic group after FDR correction, several compounds showed strong combined p-values in other comparisons. Notably, Cyclohexane (corrected p = 0.061) in the Healthy vs pre-diabetic group and 2-Pentanone (corrected p = 0.073) in the pre-diabetic vs T2DM group demonstrated significant raw and combined p-values. As these are preliminary results, it is expected further discriminatory markers may be identified as the further data is collected.
CONCLUSION:
Preliminary results from this validation study indicate that breath VOC profiling using TD-GC-TOF-MS can differentiate between glycaemic states in a South Asian population. Several compounds, including Longifolene, Nonadecane, and Octadecane, showed statistically significant differentiation between classification groups. These preliminary findings support the potential of breath analysis as a non-invasive, low-burden tool for early identification of individuals at risk of pre-diabetes and type 2 diabetes, particularly in resource-limited settings. As data collection continues, it is anticipated that further discriminatory markers will be identified, strengthening the clinical utility of this approach.
|
= Discovery stage. (53.14%, 2025)
= Translation stage. (22.33%, 2025)
= Clinically available. (24.53%, 2025)
