Direct diagnosis of tuberculosis by computer assisted pattern recognition gas chromatographic analysis of sputum.

Chemical detection of tuberculosis (TB) products in sputum was attempted by using gas chromatographic analysis in conjunction with different pattern recognition computer models. For the chromatographic separations, we used a 2 mm x 1.8 m packed column and a 0.25 mm x 30 m fused silica capillary column to analyse the methylated glycosides and fatty acid methyl ester derivatives. Three computer pattern recognition methods were applied: error score, TB score and discriminant analysis. These methods predicted the presence of active TB most often in sputa of active TB patients and less so in those from inactive, suspected and non-TB patients, in that order. Although the best true positive of 75% was obtained from the TB score method and best true negative of 98% from discriminant analysis, the accompanying false positive and false negative results (36% and 50%, respectively) were unacceptable. The use of capillary column or fatty acid methyl ester derivatives of the samples did not improve on the predictive values of chromatograms obtained from the packed column on trimethylsilylglycosidic derivatives. Additional work is needed before this method can have a direct clinical application.