PTR-MS breath gas analysis during oral glucose tolerance test in gestational diabetes screening .

The authors study the feasibility of human breath gas analysis via proton transfer reaction mass spectrometry (PTR-MS) as a diagnostic technique for gestational diabetes mellitus (GDM) screening. PTR-MS breath gas analysis allows a real-time, non invasive diagnostic procedure which can integrate many different screenings into a single test, saving time, cost and patient stress. Standard medical practice includes a pregnancy screening where blood samples are taken during an oral glucose tolerance test (OGTT). This allows the measurement of various molecule concentrations in blood including glucose, which correlates very well to fetal as well as maternal complications. The investigators collected two-hour long sequences of breath gas samples from 53 consecutive pregnant women undergoing GDM screening.  The samples were analyzed on-the-fly by a proton transfer reaction mass spectrometry (PTR-MS) device connected to a buffered end-tidal (BET) sampler [Herbig 2008], allowing the measurements of the volatile organic compounds (VOCs) in the alveolar breath gas alone. Feature selection was performed by identifying a subset of PTR-MS mass signal sequences with an identifiable time evolution. Selected mass signal sequences were fitted either with a linear function representing long-time evolution, or with an exponential function  representing short-time evolution. The following statistical analysis concentrated on the time evolution effects. Only a subset of the fitting parameters was therefore considered, namely the estimated speed parameters µ and β for each mass signal and patient. Such reduced data was organized into four groups by mean of the information provided by the screening diagnoses: GDM, IGT, healthy and borderline healthy donors. A multivariate analysis of variance (MANOVA) was performed as a feature extraction step, identifying linear classifiers and computing a properly defined separation score. A randomized permutation analysis was performed to obtain an indication of significance on the distribution of separating scores. The authors obtained a coincidence of classification with standard diagnoses, the linear classifiers identifying the 4 subgroups without mismatches. The low permutation p-value of 12% suggests that this result is significant. Current work covers, among other topics, the identification of the most important mass signals involved by means of the weights of the linear classifiers and the bio-kinetic interpretation of the findings