MELD remains the best predictor of mortality in outpatients with cirrhosis and severe ascites.

Background The Model for Endstage Liver Disease (MELD) score may put patients with severe ascites at a disadvantage because they often have a poor quality of life and high mortality despite a favourable MELD score. Aim To develop a model that is better than the MELD score at predicting 1-year mortality among patients with cirrhosis, severe ascites and MELD <= 18. Methods We used data from a randomised trial (SPARe-1) of patients with cirrhosis and severe ascites to develop a model to predict 1-year mortality. We used stepwise backward elimination and Cox regression to identify the strongest predictors. Performance was assessed with the C index and the Brier score. We examined performance in an external cohort of trial participants with cirrhosis and severe ascites (SPARe-2 participants). Results We included 308 patients with a 1-year mortality of 20.4%. The final prediction model (Severe Ascites Mortality score, "SAM score") included four variables: serum bilirubin, serum sodium, history of SBP (yes or no) and diabetes (yes or no). No indicators of quality of life were included. After correction for optimism bias, the SAM and MELD scores had nearly identical predictive ability. The external validation cohort included 149 patients whose 1-year mortality was 22.4%. The MELD score performed marginally better in this cohort, partly because the effects of SBP and diabetes on mortality were much smaller in this cohort. Conclusion We did not succeed in developing a prediction model that was superior to the MELD score among patients with cirrhosis and severe ascites.