Precision in Diagnosis of Liver Fibrosis in MASLD : Machine Learning‐Based Scores May Be More Accurate Than Conventional NITs

We read with great interest the study by Vali et al. Precision in Liver Diagnosis: Varied Accuracy Across Subgroups and the Need for Variable Thresholds in Diagnosis of MASLD [1]. Vali et al. robustly evaluated three widely used non-invasive liver tests (NITs)—fibrosis-4 index (FIB-4), enhanced liver fibrosis (ELF) test and vibration-controlled transient elastography (VCTE) across diverse patient subgroups, reinforcing the clinical utility of personalised diagnostic thresholds in a well-characterised cohort of 586 histologically confirmed MASLD patients. The study underscores the necessity of adjusting NIT thresholds based on individual patient characteristics, aligning with the principles of personalised medicine in MASLD management. Moreover, it validates the ‘stepwise’ approach recommended by the European Association for the Study of the Liver (EASL), reinforcing its clinical relevance [2]. However, we had some comments. First, its focus on secondary and tertiary care settings may restrict generalisability to primary care populations. Second, the lack of centralised histological scoring introduces potential variability, emphasising the need for expert pathologist review in future studies. Third, incomplete data from certain centres could affect robustness, underscoring the importance of comprehensive data collection. Fourth, reliance on single-variable analyses limits the understanding of interactions between clinical factors and NIT performance, highlighting the need for multivariable approaches. Finally, the study demonstrates the influence of disease prevalence on predictive values, advocating for the development and validation of multivariable thresholds in independent datasets. We have developed a simple machine-learning (ML) FIB-6 score that integrates six readily available clinical parameters: age, serum albumin, aspartate aminotransferase (AST), alkaline phosphatase (ALP), alanine aminotransferase (ALT), and platelet count. FIB-6 underwent extensive multicentre internal and external validation in patients with biopsy-proven chronic hepatitis C (n = 7238, from 8 countries) [3], chronic hepatitis B (n = 603, from 7 countries) [4], autoimmune hepatitis (AIH, n = 217 patient from 3 countries) [5] and patients with MASLD (n = 2472 from 6 countries) [6]. FIB-6 has demonstrated high diagnostic accuracy for cirrhosis (F4), significant (≥ F2) and advanced fibrosis (≥ F3) in MASLD [6]. FIB-6 score offers several advantages. It demonstrates superior area under the curve (AUC) values compared to FIB-4, particularly in ruling out advanced fibrosis. The integration of ML algorithms enhances its predictive power, addressing some limitations of conventional NITs. Moreover, FIB-6 has been validated in diverse multiethnic populations across various liver diseases, including MASLD, viral hepatitis B and C, and autoimmune hepatitis, reinforcing its generalisability. Unlike VCTE, which requires specialised equipment, the FIB-6 score relies solely on routine laboratory parameters, making it more accessible and cost-effective, particularly in resource-limited settings. Another important advantage of the FIB-6 score is its ability to provide a more comprehensive assessment of fibrosis staging. While the study by Vali et al. focuses primarily on detecting advanced fibrosis, FIB-6 offers a more detailed classification by establishing cut-offs for cirrhosis (F4) and significant fibrosis (≥ F2). This enhanced staging capability, along with its improved diagnostic accuracy and broader clinical applicability, makes FIB-6 a valuable complement to conventional non-invasive tests. Furthermore, direct comparisons between FIB-6 and the NITs evaluated in this study could clarify the most effective non-invasive fibrosis assessment strategy. Our multicentre study by Anushiravani et al. demonstrated that FIB-6 outperformed FIB-4, aspartate aminotransferase-to-platelet ratio index (APRI) and aspartate aminotransferase-to-alanine aminotransferase ratio (AAR) in sensitivity and negative predictive value for cirrhosis exclusion (97.0% and 94.7% for FIB-6, 71.6% and 94.7% for FIB-4, 36.4% and 90.7% for APRI and61.2% and 90.9% for AAR) [6]. Additionally, our FIB-6 exhibited a superior overall AUC compared to FIB-4 (0.74) [6], further supporting FIB-6's potential as a highly accurate fibrosis assessment tool in MASLD. Given the increasing role of ML-driven approaches in hepatology, validating FIB-6 in MASLD, particularly across diverse subgroups, is a logical extension of the current research. Incorporating ML-based NITs like FIB-6 into future studies could refine diagnostic accuracy, improve risk stratification and facilitate early intervention for patients at risk of advanced fibrosis. In conclusion, while the study provides valuable insights into NIT performance, future research should integrate and validate ML-based models such as FIB-6 to enhance the non-invasive diagnosis of fibrosis in MASLD. By leveraging ML, personalised diagnostic thresholds and broader validation cohorts, the field can advance towards more precise and accessible fibrosis assessment strategies. The authors declare no conflicts of interest. The authors have nothing to report.