Machine Learning Prediction of 90-Day Mortality in HBV-Related ACLF Using Olink-Derived Inflammatory Protein
Yi Zhang1, Lina Sun1, Huimin Liu2
1Center of Integrative Medicine, Beijing Ditan Hospital, Capital Medical University, Beijing, China.
Background:
Hepatitis B virus-induced acute-on-chronic liver failure (HBV-ACLF) has extremely high mortality rates and is closely linked to inflammatory responses. Early interventions in high-risk patients can significantly improve survival rates.
Methods:
In this prospective cohort study, we used the Olink proximity extension assay to compare 96 inflammation-related proteins in survivors and non-survivors of ACLF. Functional enrichment analysis of differentially expressed proteins (DEPs) was performed using KEGG to analyse their biological characteristics and correlation with disease severity. The key proteins identified in the discovery cohort (n = 32) were quantified by enzyme-linked immunosorbent assay in the modelling cohort (n = 100) to develop a machine-learning model, which was then validated in an independent cohort (n = 52).
Result:
The analysis revealed that the expression of 26 proteins was significantly elevated in non-survivors, mainly enriched in the cytokine-cytokine receptor interaction pathway. Three key proteins were highly correlated with mortality: IL-6, MMP10, and CX3CL1. The machine learning model based on these proteins accurately predicted the 90-day mortality (AUC = 0.98). Modelling and validation in both cohorts confirmed that these three indicators effectively identified high-risk patients with ACLF.
Conclusion:
IL-6, MMP10, and CX3CL1 can be used as effective biomarkers for predicting the 90-day mortality in patients with HBV-ACLF. The progression of liver failure seems to be closely related to the abnormal activation of the cytokine-cytokine receptor interaction pathway. These findings provide new directions for the study of ACLF pathogenesis and identifying potential drug targets, which are conducive to early clinical decision-making.
