Abstract

Background: Patients with acute-on-chronic liver failure grades 2 and 3 (severe ACLF) have significantly high short-term mortality. Though liver transplantation (LT) provides survival benefit for this population, scarcity of donor organs and potentially high post-LT mortality may cause uncertainty in proceeding with LT. Our aim was to develop a tool to determine if critically ill patients with severe ACLF have a high likelihood of mortality within 1-year after LT.

Methods: We collected data from 15 LT centers in the United States of patients transplanted from the intensive care unit (ICU), years 2014-2019. ACLF was defined according to the EASL-CLIF criteria. Predictors were selected based on clinical utility and statistical significance (p<0.10) from univariate logistic regression analysis. We randomly split the sample 70:30 for discovery and internal validation, with bootstrapping of 1,000 patients. We examined linear and nonlinear association for continuous variables with the LOESS method. Following the TRIPOD statement, we estimated overall performance, discrimination, and calibration. We selected the final model based on performance and simplicity.

Results: Of 735 patients transplanted from the ICU, 521 patients (70.8%) had severe ACLF, including 237 patients with ACLF-2 and 284 patients with ACLF-3. Median age was 55 years, median MELD-Na score at LT was 40, and 104 patients with severe ACLF (19.9%) died within 1-year. Using stepwise logistic regression, we created the Multi-Organ Dysfunction and Evaluation for Liver transplantation (MODEL) score, which ranges from 0.08 to 1.14. The following variables were included: age >50 years (OR=2.24, p=0.012), one vasopressor at LT (OR=1.80, p=0.09), more than one vasopressor at LT (OR=4.05, p<0.001), respiratory failure (OR=2.03, p=0.019) and multi-drug resistant bacterial infection before LT (OR=1.90, p=0.02). The combined c-statistic of the training and validation sets for the MODEL score was 0.71, indicating good discrimination. Observed and expected probability revealed good calibration (Brier score of 0.14). Hosmer-Lemeshow Goodness-of-Fit Test of Deciles of Mortality Risk showed excellent performance (p=0.8). A MODEL score of >0.43 predicted mortality ≥30% within 1-year after LT with a specificity above 87%, while a score >0.66 predicted ≥40% mortality in 1 year after LT with a specificity above 94%. Discrimination of the MODEL score significantly outperformed the MELD-Na, MELD-lactate, donor (D)-meld, delta meld, and balance of risk (BAR) scores at predicting 1-year post-LT mortality (p<0.001). (Figure)

Conclusion: Derived from the largest granular dataset of patients transplanted with severe ACLF, the MODEL score provides a prognostic tool using easily available information to predict high probability of mortality within 1-year post-LT. Such information can guide patients and providers as to whether LT may be futile. We are piloting an online calculator.