Prognostic performance of a series of model for end-stage liver disease and respective Δ scores in patients with hepatitis B acute-on-chronic liver failure

The present study aimed to compare the short-term prognostic performance of a series of model for end-stage liver disease (MELD) and respective delta (Δ) scores scoring systems in a population with acute-on-chronic hepatitis B liver failure (ACHBLF), and to investigate the potential effects from antivirals. A total of 77 patients with ACHBLF of mean age 46 years, 82% male, with 58.4% receiving antivirals, were recruited for this study. The Δ scores for MELDs were defined as the changes one week after admission. Thirty-eight (49%) patients (22 treated with antivirals) died within three months. The mean MELD and ΔMELD scores of the survival group were 19.5±4.4 and 0.2±3.7 respectively, and those of the mortality group were 23.5±5.5 and 7.9±6, respectively. The area under the receiver operating characteristic curve (AUC) for MELD, integrated MELD (iMELD), MELD with the addition of serum sodium (MELD-Na), updated MELD (upMELD), MELD excluding the international normalized ratio (INR; MELD-XI), United Kingdom MELD (UKMELD) and their Δ scores were 0.72, 0.81, 0.77, 0.69, 0.65, 0.77 and 0.86, 0.83, 0.83, 0.82, 0.79 and 0.79, respectively. iMELD and MELD-Na significantly improved the accuracy of MELD (P<0.05). A cut-off value of 41.5 for the iMELD score can prognose 71% of mortalities with a specificity of 85%. In each pair of models, the Δ score was superior to its counterpart, particularly when applied to patients with MELD ≤30. Decreased accuracy was observed for all models in the subset of patients treated with antivirals, although their baseline characteristics were comparable to those of untreated patients, while iMELD, MELD-Na and respective Δ models remained superior with regard to the predictability. The iMELD and MELD-Na models predicted three-month mortality more accurately, while the Δ models were superior to their counterparts when MELD ≤30; however, their performance was altered by antivirals, and thus requires optimization.