A model for individualized prediction of liver-related death in outpatients with alcohol-associated cirrhosis

Abstract

Introduction: In alcohol-associated cirrhosis, an accurate estimate of the risk of death is essential for patient care. We developed individualized prediction charts for 5-year liver-related mortality among outpatients with alcohol-associated cirrhosis that take into account the impact of abstinence.

Methods: We collected data on outpatients with alcohol-associated cirrhosis in a prospective registry. The model was derived, internally and externally validated, and compared with the Child-Pugh and the Model For End-Stage Liver Disease (MELD) scores.

Results: A total of 527 and 127 patients were included in the derivation and validation data sets, respectively. A model was developed based on the 3 variables independently associated with liver-related mortality in multivariate analyses (age, Child-Pugh score, and abstinence). In the derivation data set, the model combining age, Child-Pugh score, and abstinence outperformed the Child-Pugh and the MELD scores. In the validation data set, the Brier score was lower for the model (0.166) compared with the Child-Pugh score (0.196, p = 0.008) and numerically lower compared with the MELD score (0.190) (p = 0.06). The model had the greatest AUC (0.77; 95% CI 0.68-0.85) compared with the Child-Pugh score (AUC = 0.66; 95% CI 0.56-0.76, p = 0.01) and was numerically higher than that of the MELD score (AUC = 0.66; 95% CI 0.56-0.78, p = 0.06). Also, the Akaike and Bayesian information criterion scores were lower for the model (2163; 2172) compared with the Child-Pugh (2213; 2216) or the MELD score (2205; 2208).

Conclusion: A model combining age, Child-Pugh score, and abstinence accurately predicts liver-related death at 5 years among outpatients with alcohol-associated cirrhosis. In this study, the model outperformed the Child-Pugh and the MELD scores, although the AUC and the Brier score of the model were not statically different from the MELD score in the validation data set.

FIGURE 1

Flowchart of the derivation data set.

FIGURE 2

Chart predicting the risk of liver-related death at 5 years in patients who abstained from alcohol (A) and who did not abstain from alcohol (B). Legend:☆, hypothetical 60-year-old patient with a Child-Pugh score of 9 at baseline who abstained from alcohol: 5-year liver-related mortality rate is 23%. Legend:☆, hypothetical 60-year-old patient with a Child-Pugh score of 9 at baseline who did not abstain from alcohol: 5-year liver-related mortality rate is 51%.

FIGURE 3

Receiver operating characteristic curves of the different models in the derivation data set.

FIGURE 4

Calibration plots for the models combining (A) age, Child-Pugh score, and abstinence, (B) MELD score alone, (C) Child-Pugh score alone, and (D) age, MELD score, Child-Pugh score, and abstinence in the derivation data set. The calibration slope evaluates the spread of the estimated risks and has a target value of 1. A slope < 1 suggests that estimated risks are too extreme, that is, too high for patients who are at high risk (‘overfitting’). A slope > 1 suggests the opposite (‘underfitting’). The calibration intercept has a target value of 0, negative values suggest overestimation, whereas positive values suggest underestimation. Abbreviation: MELD, Model For End-Stage Liver Disease