Development and clinical validation of a novel protein biomarkers-based algorithm for risk prediction and diagnosis of advanced liver fibrosis: a multi-centre study

Abstract

Introduction

Type 2 diabetes (T2D) and obesity contribute significantly to the elevated risk of liver fibrosis in metabolic dysfunction-associated steatotic liver disease (MASLD). However, there is a lack of reliable and cost-effective non-invasive test (NIT) for detecting liver fibrosis in T2D/obese individuals.

Objectives

This study aimed to develop a simple biomarker-based algorithm for detecting advanced liver fibrosis among T2D/obesity subjects with MASLD and to validate its diagnostic performance in both clinic- and community-based cohorts.

Methods

Diagnostic performances of circulating thrombospondin-2 (TSP2), a novel fibrosis marker, and the three individual components of Enhanced Liver Fibrosis (ELF) test were evaluated in three independent cohorts. These included a clinic-based derivation cohort (N = 846) and a community-based validation cohort (N = 803), both comprising of T2D patients with vibration-controlled transient elastography (VCTE)-diagnosed MASLD. Additionally, a clinic-based validation cohort of morbidly-obese patients with biopsy-proven MASLD (N = 223) was included. An algorithm (TaP score) based on TSP2 and procollagen 3 N-terminal peptide (PIIINP), a component of ELF, was constructed from the multivariate logistic regression model and compared with existing NITs, including ELF, fibrosis-4 index (FIB-4) and NAFLD fibrosis score (NFS). The dual-cut-off approach was used to define the rule-in and rule-out cut-offs.

Results

Circulating TSP2 (AUC[95 %CI]:0.844[0.810–0.878]) and PIIINP (AUC[95 %CI]:0.843(0.807–0.875]) showed excellent diagnostic performance and were used to construct the biomarker-based algorithm. The TaP score (AUC[95 %CI]:0.900[0.874–0.925]) significantly outperformed ELF (AUC[95 %CI]:0.809[0.773–0.843]), FIB-4 (AUC[95 %CI]:0.597[0.544–0.647]) and NFS (AUC[95 %CI]:0.585[0.528–0.639]) (all DeLong P < 0.001), showing high specificity (85.16 %), sensitivity (78.62 %), and negative predictive value (NPV) (95.08 %) at the optimal cut-off. This algorithm resulted in fewer patients with indeterminate results compared to ELF. Its diagnostic performance in the two external validation cohorts was comparable to that in the derivation cohort.

Conclusions

The TaP score demonstrated good diagnostic ability with generally better performance compared to ELF, and had the potential to be developed as a novel NIT.