Yusuf Yilmaz1, Fatih Eren. 1. Department of Gastroenterology, Marmara University, School of Medicine, Istanbul, Turkey; Institute of Gastroenterology, Marmara University, Istanbul, Turkey. dryusufyilmaz@gmail.com
Abstract
BACKGROUND: Although liver biopsy remains the best diagnostic standard for nonalcoholic steatohepatitis (NASH), non-invasive tests are eagerly awaited. In this study, we sought to develop a support vector machine (SVM) algorithm to discriminate with high accuracy between subjects with NASH and controls using a blood-based biomarker panel. METHODS: A total of 17 biomarkers were measured by commercially available enzyme-linked immunosorbent assays in 136 serum samples from patients with biopsy-proven NASH (n = 60) and subjects with normal ALT and no evidence of fatty liver on ultrasound (n = 76). The database was randomly divided (1:1 fashion) into a discovery set for classification training and in a validation set of the chosen biomarkers in blinded samples. Multivariate analysis was performed by means of SVM. RESULTS: After the identification of a group of three most discriminative biomarkers (osteoprotegerin, fibroblast growth factor 21, and M30) in the discovery set, the application of SVM to the validation test resulted in a 92.5% sensitivity and 84.1% specificity for distinguishing subjects with NASH from controls. CONCLUSIONS: A targeted biomarker profiling combined with a SVM-based pattern identification approach may allow the identification of patients with NASH with clinically relevant accuracy and validity.
BACKGROUND: Although liver biopsy remains the best diagnostic standard for nonalcoholic steatohepatitis (NASH), non-invasive tests are eagerly awaited. In this study, we sought to develop a support vector machine (SVM) algorithm to discriminate with high accuracy between subjects with NASH and controls using a blood-based biomarker panel. METHODS: A total of 17 biomarkers were measured by commercially available enzyme-linked immunosorbent assays in 136 serum samples from patients with biopsy-proven NASH (n = 60) and subjects with normal ALT and no evidence of fatty liver on ultrasound (n = 76). The database was randomly divided (1:1 fashion) into a discovery set for classification training and in a validation set of the chosen biomarkers in blinded samples. Multivariate analysis was performed by means of SVM. RESULTS: After the identification of a group of three most discriminative biomarkers (osteoprotegerin, fibroblast growth factor 21, and M30) in the discovery set, the application of SVM to the validation test resulted in a 92.5% sensitivity and 84.1% specificity for distinguishing subjects with NASH from controls. CONCLUSIONS: A targeted biomarker profiling combined with a SVM-based pattern identification approach may allow the identification of patients with NASH with clinically relevant accuracy and validity.