MELD-S prognostic model: development and external validation for 90-day mortality prediction in acute-on-chronic liver failure

Accurate prognostic assessment of acute-on-chronic liver failure (ACLF) remains a major clinical challenge. Existing models, such as the Model for End-Stage Liver Disease (MELD), rely solely on clinical indicators and do not incorporate molecular biomarkers potentially associated with hepatic regenerative processes. This study aimed to identify key regeneration-associated genes through bioinformatics, integrate the most promising biomarker with clinical parameters to construct a novel 90-day mortality prediction model, and validate its performance for improving risk stratification in ACLF.

Transcriptomic data from an ACLF cohort (GSE139602) and a liver regeneration dataset (GSE15239) were obtained from the Gene Expression Omnibus (GEO) database. Differentially expressed genes (DEGs) were screened, followed by functional enrichment analysis and protein-protein interaction (PPI) network construction. From August 2023 to July 2024, 64 patients with ACLF were prospectively enrolled and categorized based on 90-day outcomes. After 1:1 propensity score matching for age, sex, MELD score, and complications, 50 patients (25 survivors, 25 non-survivors) were selected as the development cohort. Separately, from August 2024 to July 2025, 52 patients with ACLF were enrolled, and after identical matching, 40 patients (20 survivors, 20 non-survivors) formed the independent external validation cohort. The matched development cohort was used for exploratory model development, while the independent external validation cohort was used to evaluate reproducibility. Serum levels of proteins encoded by core genes (SGK1, GADD45β, FOXO1) were measured using enzyme-linked immunosorbent assay (ELISA). The MELD-S model, integrating MELD score and SGK1, was developed via multivariate logistic regression. Its performance was evaluated through ROC analysis, calibration curves, Hosmer-Lemeshow test, and decision curve analysis (DCA) in both the development and external validation cohorts, with a nomogram created for visualization.

Bioinformatics analysis identified 20 key genes commonly dysregulated in ACLF and liver regeneration, enriched in transcription-related functions and the FOXO signaling pathway. SGK1, GADD45β, and FOXO1 were identified as candidate regeneration-associated genes, with significant protein interactions. In the development cohort, serum levels of SGK1, GADD45β, FOXO1, PT, and INR were significantly higher in non-survivors compared with survivors (all P < 0.05). Serum SGK1 levels showed a positive correlation with MELD score (r = 0.316, P = 0.027) and SGK1 was identified as an independent risk factor for 90-day mortality (OR = 3.955, 95% CI: 1.486 - 10.525, P = 0.006). The integrated MELD-S model yielded an AUC of 0.865 (95% CI: 0.767-0.963), demonstrating superior discriminative ability to MELD alone (AUC = 0.735) or SGK1 alone (AUC = 0.832). In the independent external validation cohort, the MELD-S model achieved an AUC of 0.725 (95% CI: 0.563-0.887, P = 0.015), compared with an apparent AUC of 0.865 in the development-cohort. Bootstrap internal validation of the development cohort yielded an optimism-adjusted AUC of 0.842.

SGK1 is an independent prognostic biomarker associated with disease severity and biological stress responses in ACLF. The MELD-S model shows promising prognostic performance and may have potential clinical utility, offering a practical tool for risk assessment and individualized risk stratification. The biological mechanisms linking SGK1 to ACLF outcomes, including the potential involvement of the SGK1-FOXO1 pathway, remain hypothetical and require direct mechanistic validation in future studies.