Human breath gas analysis in the screening of gestational diabetes mellitus

Abstract

Background: We present a pilot study on the feasibility of the application and advantages of online, noninvasive breath gas analysis (BGA) by proton transfer reaction quadrupole mass spectrometry for the screening of gestational diabetes mellitus (GDM) in 52 pregnant women by means of an oral glucose tolerance test (OGTT).

Subjects and methods: We collected and identified samples of end-tidal breath gas from patients during OGTT. Time evolution parameters of challenge-responsive volatile organic compounds (VOCs) in human breath gas were estimated. Multivariate analysis of variance and permutation analysis were used to assess feasibility of BGA as a diagnostic tool for GDM.

Results: Standard OGTT diagnosis identified pregnant women as having GDM (n = 8), impaired glucose tolerance (n = 12), and normal glucose tolerance (n = 32); a part of this latter group was further subdivided into a "marginal" group (n = 9) because of a marginal high 1-h or 2-h OGTT value. We observed that OGTT diagnosis (four metabolic groups) could be mapped into breath gas data. The time evolution of oxidation products of glucose and lipids, acetone metabolites, and thiols in breath gas after a glucose challenge was correlated with GDM diagnosis (P = 0.035). Furthermore, basal (fasting) values of dimethyl sulfide and values of methanol in breath gas were inversely correlated with phenotype characteristics such as homeostasis model assessment of insulin resistance index (R = -0.538; P = 0.0002, P(corrected) = 0.0034) and pregestational body mass index (R = -0.433; P = 0.0013, P(corrected) = 0.022).

Conclusions: Noninvasive BGA in challenge response studies was successfully applied to GDM diagnosis and offered an insight into metabolic pathways involved. We propose a new approach to the identification of diagnosis thresholds for GDM screening.

FIG. 1.

Flow chart of optimization process for the analysis of volatile organic compounds (VOCs). GDM, gestational diabetes mellitus; MANOVA, multivariate analysis of variance; OGTT, oral glucose tolerance test.

FIG. 2.

Examples of (A) kinetic (mass-to-charge [m/z] 59, 61, 79) and (B) linear (m/z 55, 18, 45) fitting for the determination of the time evolution parameters based on two kinetic models.

FIG. 3.

Comparison between the results of multivariate analysis of variance and permutation analysis on three different datasets associated with the oral glucose tolerance test patients: (B) point measurements at zero-time (i.e., the signal intensities before the glucose challenge, corresponding to the basal values) and (C) point measurements at time 15 min (the moment when most kinetic-fitted mass-to-charge signals reach their peak). Notice how the high dimensionality of the problem affects data analysis: although multivariate analysis of variance is able to produce an acceptable separation with any of the three datasets, the associated permutation analysis shows that the results corresponding to the time point measurements (B) and (C) are not statistically significant (way too high P values), meaning that the data are so noisy/diverse that a separation of comparable quality in randomly chosen groups is very much probable. This demonstrates the advantage of the proposed method of focusing on time evolution parameters (A), which leads to an excellent differentiation of the four metabolic groups with only one misclassification. GDM, gestational diabetes mellitus; IGT, impaired glucose tolerance.

FIG. 4.

Correlation of (A) the volatile organic compound dimethyl sulfide (mass-to-charge 63) and homeostasis model assessment of insulin resistance (HOMA-IR) and (B) methanol (mass-to-charge 33) and pregestational body mass index (BMI) in pregnant women participating in gestational diabetes mellitus screening. Dimethyl sulfide and methanol are measured before the oral glucose tolerance test (0 min, basal value).