Evaluating the Discriminative Performance of Noninvasive Biomarkers in Chronic Hepatitis B/C, Alcoholic Cirrhosis, and Nonalcoholic Cirrhosis: A Comparative Analysis

Introduction: The clinical implementation of noninvasive tests for liver fibrosis assessment has attracted increasing attention, particularly for diagnosing advanced fibrosis (>= F3). This observational study aimed to evaluate the stratification accuracy of nine direct and seven indirect biomarkers across four etiologies: chronic hepatitis B (CHB), chronic hepatitis C (CHC), alcoholic liver cirrhosis (ALC), and nonalcoholic liver cirrhosis (NALC). Materials and Methods: Our study was conducted on 116 participants, including 96 with chronic liver disease (16 CHB, 15 CHC, 49 ALC, and 16 NALC) and 20 healthy controls. The values of direct (aspartate aminotransferase, alanine aminotransferase, total bilirubin, serum albumin, platelet count, international normalized ratio, gamma-glutamyl transpeptidase, CD5 antigen-like, and transforming growth factor-beta 1) and indirect non-serological biomarkers (De Ritis ratio, albumin-bilirubin score, gamma-glutamyl transpeptidase-to-platelet ratio, aspartate aminotransferase-to-platelet-ratio index, fibrosis-4 index, INR-to-platelet ratio, and fibrosis quotient) were analyzed for their discriminative power in fibrosis stratification. Results: Statistical analyses revealed a significant correlation (0.05 level; two-tailed), and AUC 95% CI ranged within 0.50-1.00 between the direct and indirect biomarker values across all etiologies. Among the evaluated biomarkers, the recorded AUC was 0.998 in CHB for APRI, 0.981 in CHC for FIB-4, and 1.000 in ALC and NALC for APRI and AST, respectively, while CD5L consistently achieved an AUC of 1.000 across all etiologies. Conclusions: These findings suggest that applying a multifactorial approach in liver pathology may improve diagnosis accuracy compared to the use of individual biomarkers and can provide data that may inform the development of clinically applicable mathematical models.