A Predictive Framework for Acute Liver Failure Outcomes: Insights from Nonbiological Artificial Liver Treatment

The aim of this study was to evaluate the efficacy of nonbiological artificial liver (NBAL) in the treatment of acute liver failure (ALF) and to screen prognostic predictors.

This study was a single-center retrospective observational study that included 186 ALF patients treated with NBAL (plasma exchange alone, DPMAS alone, or combination therapy). Laboratory parameters (TBil, INR, Cr, etc.) before and after 24 hours of treatment and 28-day and 90-day survival data were collected. Prognostic features were screened by XGBoost and Random Forest, independent risk factors were analyzed by Cox regression, and column-line plots were constructed.

The 90-day poor prognosis group (mortality/liver transplantation) presented a significant high-risk phenotype as evidenced by a higher MELD score (34.46 vs. 29.84, p < 0.001), greater severity of hepatic encephalopathy (64.6% vs. 38.8% share of grades III - IV, p < 0.001), and concomitant overall deterioration in the indicators of hepatic and renal impairment (TBil, INR, and Cr) (all p < 0.001). The machine learning model further validated the predictive value of the dynamic metrics, with TBil% on day 1 of treatment with a decision-tree significance of 26.3%, significantly outperforming traditional baseline metrics (e.g., MELD score and TBil). Risk stratification based on median TBil% (14.26%) showed a 28-day all-cause mortality rate of 46.24% (vs. 10.75% in the low-risk group, p < 0.001) in the high-risk group and a 90-day survival of only 36.6% (vs. 75.3% in the low-risk group, p < 0.0001). Each 1% decrease in 24-hour TBil% increased the risk of mortality by 7% (HR = 0.93, p < 0.001), while baseline MELD score (HR = 1.07, p < 0.001) and 24-hour INR (HR = 1.70, p < 0.001) were independent risk factors.

Early TBil clearance in ALF treated with NBAL is a strong predictor of 90-day survival, and baseline MELD score and post-treatment INR are also independent risk factors.