The degree of liver fibrosis in non-alcoholic fatty liver disease (NAFLD) is a critical predictive factor for patient prognosis. This study was intended to perform external validation of the various fibrosis prediction models for assessing advanced fibrosis in Korean NAFLD patients.Methods:
A retrospective study of 412 patients with NAFLD confirmed by liver biopsy in hospitals affiliated with the Koran NAFLD study group was conducted and the predictive ability of existing liver fibrosis prediction models including NAFLD fibrosis score (NFS), BARD, and fibrosis-4 were compared.Results:
Among 412 samples, 328 liver slides were suitable for evaluation. Advanced fibrosis was present in 60 (18.3%) of the patient samples. Univariate analysis found that the group with advanced fibrosis showed low alanine aminotransferase values and high aspartate aminotransferase/alanine aminotransferase ratios as well as a high incidence of diabetes. However, multivariate analysis showed that only the presence of diabetes and triglycerides was independent risk factors. The receiver operating characteristic was 0.64 in NFS, 0.58 in fibrosis-4, and 0.594 in the BARD model. The NFS was found to be the best at predicting advanced fibrosis among the three prediction models. The negative predictive value which predicts advanced fibrosis using the low cutoff (<−1.455) was high (86.6%). However, the positive predictive value which predicts advanced fibrosis using the high cutoff (>0.676) was 50.0% when we applied the NFS.Conclusion:
Negative predictive value using the low cutoff value was high, but positive predictive value using the high cutoff value was low in a Korean NAFLD cohort using NFS.