APASL-ACLF Research Consortium-Artificial Intelligence (AARC-AI) model precisely predicts outcomes in acute-on-chronic liver failure patients.

BACKGROUND AND AIMS: We hypothesized that artificial intelligence (AI) models are more precise than standard models for predicting outcomes in acute-on-chronic liver failure (ACLF). METHODS: We recruited ACLF patients between 2009 and 2020 from APASL-ACLF Research Consortium (AARC). Their clinical data, investigations and organ involvement were serially noted for 90-days and utilized for AI modelling. Data were split randomly into train and validation sets. Multiple AI models, MELD and AARC-Model, were created/optimized on train set. Outcome prediction abilities were evaluated on validation sets through area under the curve (AUC), accuracy, sensitivity, specificity and class precision. RESULTS: Among 2481 ACLF patients, 1501 in train set and 980 in validation set, the extreme gradient boost-cross-validated model (XGB-CV) demonstrated the highest AUC in train (0.999), validation (0.907) and overall sets (0.976) for predicting 30-day outcomes. The AUC and accuracy of the XGB-CV model (%Δ) were 7.0% and 6.9% higher than the standard day-7 AARC model (p < .001) and 12.8% and 10.6% higher than the day 7 MELD for 30-day predictions in validation set (p < .001). The XGB model had the highest AUC for 7- and 90-day predictions as well (p < .001). Day-7 creatinine, international normalized ratio (INR), circulatory failure, leucocyte count and day-4 sepsis were top features determining the 30-day outcomes. A simple decision tree incorporating creatinine, INR and circulatory failure was able to classify patients into high (~90%), intermediate (~60%) and low risk (~20%) of mortality. A web-based AARC-AI model was developed and validated twice with optimal performance for 30-day predictions. CONCLUSIONS: The performance of the AARC-AI model exceeds the standard models for outcome predictions in ACLF. An AI-based decision tree can reliably undertake severity-based stratification of patients for timely interventions.

Dynamic assessments of hepatic encephalopathy and ammonia levels predict mortality in acute-on-chronic liver failure.

BACKGROUND: We evaluated the dynamics of hepatic encephalopathy (HE) and ammonia estimation in acute-on-chronic liver failure (ACLF) patients due to a paucity of evidence. METHODS: ACLF patients recruited from the APASL-ACLF Research Consortium (AARC) were followed up till 30 days, death or transplantation, whichever earlier. Clinical details, including dynamic grades of HE and laboratory data, including ammonia levels, were serially noted. RESULTS: Of the 3009 ACLF patients, 1315 (43.7%) had HE at presentation; grades I-II in 981 (74.6%) and grades III-IV in 334 (25.4%) patients. The independent predictors of HE at baseline were higher age, systemic inflammatory response, elevated ammonia levels, serum protein, sepsis and MELD score (p < 0.05; each). The progressive course of HE was noted in 10.0% of patients without HE and 8.2% of patients with HE at baseline, respectively. Independent predictors of progressive course of HE were AARC score (≥ 9) and ammonia levels (≥ 85 µmol/L) (p < 0.05; each) at baseline. A final grade of HE was achieved within 7 days in 70% of patients and those with final grades III-IV had the worst survival (8.9%). Ammonia levels were a significant predictor of HE occurrence, higher HE grades and 30-day mortality (p < 0.05; each). The dynamic increase in the ammonia levels over 7 days could predict nonsurvivors and progression of HE (p < 0.05; each). Ammonia, HE grade, SIRS, bilirubin, INR, creatinine, lactate and age were the independent predictors of 30-day mortality in ACLF patients. CONCLUSIONS: HE in ACLF is common and is associated with systemic inflammation, poor liver functions and high disease severity. Ammonia levels are associated with the presence, severity, progression of HE and mortality in ACLF patients.

Comparative accuracy of prognostic models for short-term mortality in acute-on-chronic liver failure patients: CAP-ACLF.

BACKGROUND: Multiple predictive models of mortality exist for acute-on-chronic liver failure (ACLF) patients that often create confusion during decision-making. We studied the natural history and evaluated the performance of prognostic models in ACLF patients. METHODS: Prospectively collected data of ACLF patients from APASL-ACLF Research Consortium (AARC) was analyzed for 30-day outcomes. The models evaluated at days 0, 4, and 7 of presentation for 30-day mortality were: AARC (model and score), CLIF-C (ACLF score, and OF score), NACSELD-ACLF (model and binary), SOFA, APACHE-II, MELD, MELD-Lactate, and CTP. Evaluation parameters were discrimination (c-indices), calibration [accuracy, sensitivity, specificity, and positive/negative predictive values (PPV/NPV)], Akaike/Bayesian Information Criteria (AIC/BIC), Nagelkerke-R2, relative prediction errors, and odds ratios. RESULTS: Thirty-day survival of the cohort (n = 2864) was 64.9% and was lowest for final-AARC-grade-III (32.8%) ACLF. Performance parameters of all models were best at day 7 than at day 4 or day 0 (p < 0.05 for C-indices of all models except NACSELD-ACLF). On comparison, day-7 AARC model had the numerically highest c-index 0.872, best accuracy 84.0%, PPV 87.8%, R2 0.609 and lower prediction errors by 10-50%. Day-7 NACSELD-ACLF-binary was the simple model (minimum AIC/BIC 12/17) with the highest odds (8.859) and sensitivity (100%) but with a lower PPV (70%) for mortality. Patients with day-7 AARC score > 12 had the lowest 30-day survival (5.7%). CONCLUSIONS: APASL-ACLF is often a progressive disease, and models assessed up to day 7 of presentation reliably predict 30-day mortality. Day-7 AARC model is a statistically robust tool for classifying risk of death and accurately predicting 30-day outcomes with relatively lower prediction errors. Day-7 AARC score > 12 may be used as a futility criterion in APASL-ACLF patients.