A Stepwise Algorithmic Approach and External Validation Study for Noninvasive Prediction of Advanced Fibrosis in Nonalcoholic Fatty Liver Disease.

Journal Article (Journal Article)

BACKGROUND AND AIMS: Advanced F3-4 fibrosis predicts liver-related mortality in nonalcoholic fatty liver disease (NAFLD). Noninvasive tests, designed to rule in/out advanced fibrosis, are limited by indeterminates, necessitating biopsy. We aimed to determine whether stepwise combinations of noninvasive serum-based tests and elastography (VCTE) could predict F3-4, reduce indeterminates, and decrease liver biopsies. METHODS AND RESULTS: Five hundred forty-one biopsy-proven NAFLD cases were identified between 2010 and 2018 from two Canadian centers. Characteristics of training (n = 407)/validation (n = 134) cohorts included: males 54%/59%; mean age 48.5/52.5 years; mean body mass index 32.3/33.6 kg/m2; diabetes mellitus 30%/34%; and F3-4 48%/43%. For training/validation cohorts, area under the receiver operating curve (AUROC) for FIB-4, AST-platelet ratio index (APRI), NAFLD fibrosis score (NFS), BARD score, and AST/ALT ratio ranged from 0.70 to 0.83/0.68 to 0.81, with indeterminates 25-39%/34-45%, for F3-4. In the training cohort, parallel FIB-4 + NFS had good accuracy (AUROC = 0.81) but was limited by 38% indeterminates and 16% misclassified. Sequential FIB-4 → NFS reduced indeterminates to 10%, and FIB-4 → VCTE to 0%, misclassified 20-22%, while maintaining high specificity (0.88-0.92) and accuracy (AUROC 0.75-0.78) for combined cohorts. Liver biopsy could have been avoided in 27-29% of patients using sequential algorithms. CONCLUSIONS: Sequential FIB-4 ➔ NFS/VCTE predicts F3-4 with high specificity and good accuracy, while reducing indeterminates and need for biopsy. Parallel algorithms are limited by high indeterminates. Sequential FIB-4 ➔ NFS had similar accuracy to VCTE-containing algorithms. Validation in low-prevalence cohorts may allow for potential use in community or resource-limited areas for risk stratification.

Full Text

Duke Authors

Cited Authors

  • Kosick, HM-K; Keyrouz, A; Adeyi, O; Sebastiani, G; Patel, K

Published Date

  • November 2021

Published In

Volume / Issue

  • 66 / 11

Start / End Page

  • 4046 - 4057

PubMed ID

  • 33389416

Electronic International Standard Serial Number (EISSN)

  • 1573-2568

Digital Object Identifier (DOI)

  • 10.1007/s10620-020-06748-8


  • eng

Conference Location

  • United States