Application of Skin Gas GC/MS Analysis for Prediction of the Severity Scale of Parkinson’s Disease

Abstract

Parkinson's disease (PD) is diagnosed by neurological examinations, as well as by scintigraphy for the dopamine transporter and metaiodobenzylguanidine (MIBG). We studied possible application of the skin gas in diagnosis of PD. We analyzed chemical substances emanated from the skin by gas chromatograph/mass spectrometer (GC/MS) after on-line-pre-concentrator. We analyzed the skin gas in 61 PD patients and 61 controls. The GC/MS chromatograms were sectionalized every 30 sec. The retention time drift was shifted every 5 sec, and a similarity coefficient (Z score) between a reference chromatogram and each shifted chromatogram was calculated. Chromatograms with high Z scores were excluded from our analysis. Models were made with partial least square (PLS), support vector machine (SVM), and support vector regression (SVR) analyses. PLS modeling to predict the Unified Parkinson's Disease Rating Scale part 3 (UPDRS3), representing motor deficits in PD, with all the detected mass numbers yielded 50 mass numbers with high PLS coefficients. Among the 50 mass numbers, nine mass numbers (m/e 48, 63, 67, 70, 81, 93, 96, 104, and 105) had dependable signal-to-noise ratios. We then generated an SVM model to differentiate PD and controls. Our SVM model had a sensitivity of 90.2% and a specificity of 85.2% by leave-one-out cross-validation (LOOCV) analysis. We next generated an SVR model to predict UPDRS3 with the nine mass numbers, and obtained a Pearson's correlation coefficient of 0.834. LOOCV analysis of our SVR model similarly gave rise to a correlation coefficient of 0.710. We propose that chemical substances in the skin gas potentially serve as biomarkers for PD.