Diagnosis of Type II Diabetes based on Non-glucose Regions of 1H NMR Spectra of Urine A metabonornic approach
Nicolescu, Alina; Dolenko, Brion; Bezabeh, Tedros; Stefan, Lorena-Ivona; Ciurtin, Coziana; Kovacs, Eugenia; Smith, Ian C. R.; Simionescu, Bogdan C.; Deleanu, Calin
NMR spectroscopy; Urinary metabolites; diabetes statistical classification chemometry
A NMR dataset with non-buffered urine samples consisting of 73 controls and 94 type II diabetes was subject to an in-house statistical classifier. A model was developed based on two glucose-free regions of the spectrum and those maximally discriminatory subregions selected most often by the algorithm were noted. The final classifier achieved 83.0% sensitivity and 83.6% specificity, with 83.2% overall accuracy. There were five spectral subregions selected by the algorithm as most relevant for discrimination. The protocol works well with non-buffered samples and has the potential for an automated clinical diagnosis of diabetes.