Machine Learning Identifies Three Distinct Ascites Phenotypes in Cirrhosis: Liver Transplantation | August 2026

Introduction:

Ascites marks a major transition in cirrhosis, yet patients with apparently similar ascites can have markedly different clinical trajectories. This multicenter study used machine learning-based latent class analysis to determine whether routinely available clinical variables could identify distinct ascites phenotypes and predict complications among liver transplant candidates.

Why was this study needed?

Ascites substantially worsens prognosis in cirrhosis.

Conventional grading does not capture the biological heterogeneity of decompensation.

Patients with similar ascites severity can have very different risks of renal dysfunction and clinical deterioration.

Phenotype-based classification could improve individualized management beyond MELD 3.0.

Results:

Three reproducible phenotypes emerged: CKD-Metabolic, Vasodilatory-Synthetic Dysfunction, and PVT-Intermediate.

The Vasodilatory-Synthetic Dysfunction phenotype, characterized by higher bilirubin and lower blood pressure, had approximately a threefold greater risk of acute kidney injury (AKI).

Importantly, this increased AKI risk remained significant independent of MELD 3.0 and was confirmed in external validation cohorts.

Waitlist mortality was similar across phenotypes, although the highest-risk phenotype underwent transplantation more frequently.

Clinical Impact:

This study suggests that ascites should not be viewed as a single clinical phenotype. Simple routinely available variables can identify patients with distinct pathophysiology and complication risk. In particular, recognizing the vasodilatory–synthetic dysfunction phenotype may identify patients requiring closer renal monitoring and earlier intervention, even when MELD 3.0 alone does not fully reflect their risk.

Bottom Line:

Machine learning identifies three clinically distinct ascites phenotypes in cirrhosis. A vasodilatory–synthetic dysfunction phenotype carries a threefold higher risk of AKI independent of MELD 3.0, supporting a more personalized approach to patients with decompensated cirrhosis.