K-means clustering analysis of C/C, O/O and H/H isotope ratios for identification of alcohol origin.

Using a single qualitative and quantitative test for ethanol consumption sometimes fails to reflect the objective facts at the beginning of the incident for a variety of reasons, such as metabolism, diffusion, collection contamination, storage losses, and decomposition. Stable isotope signatures, because they are natural properties of the substance itself, have the ability to describe the origin of the substance and therefore have the potential to directly trace ethanol. Previous studies have shown that discrimination of ethanol origins can be achieved to a certain extent by combining the measurement of δC values with the calculation of likelihood ratios. In this study, we analyzed 99 alcohol consumption samples (group A), 77 contamination samples (group B), and 14 decomposition samples (group C) by a gas chromatography-isotope ratio mass spectrometry method. We obtained δC values, δO values, and δH values for all three sample types and analyzed the data by a likelihood ratio calculation or a k-means clustering analysis-ROC curve method and compared them to 5 authentic samples to determine whether they were derived from alcohol consumption. The accuracy of the likelihood ratio method was 80 %, and the accuracy of the k-means cluster analysis-ROC curve system was 100 %. These findings demonstrated that increasing the dimensionality of the data can improve the accuracy of the likelihood rate method and that the k-means cluster analysis-ROC curve system was able to distinguish with good accuracy whether ethanol in blood samples was derived from alcohol consumption.