OBJECTIVETo compare the acute kidney injury classification systems of RIFLE,AKIN,KDIGO and conventional criteria for determining prognosis of acute-on-chronic liver failure (ACLF) patients.

METHODSPatients with ACLF admitted to our hospital between July 2008 and March 2014 were enrolled in the study. The incidence, stages, and outcomes of acute kidney injury were determined according to the RIFLE, AKIN,KDIGO and conventional criteria.ROC curves were generated to compare the predictive ability for 30-day mortality of the four systems.Chi-square test and Fisher's exact test were used for statistical analyses, as well.

RESULTSAll four classification systems detected acute kidney injury among the patients in the study population (n =358), but the detection rates were not consistent (expressed as % of total): KDIGO criteria: 45.0%, AKIN: 38.8%, rIFLE: 35.5%, conventional criterion: 20.4%. The KDIGO and AKIN criteria showed higher sensitivity (72%), especially to early kidney injury, but the conventional criterion showed higher specificity (92%). The AUC for 30-day mortality was highest for the conventional criteria (0.75), followed by AKIN (0.72), rIFLE (0.70) and KDIGO (0.69) (all, P less than 0.05). In-hospital mortality increased with severity of AKI in a stepwise manner.

CONCLUSIONAmong the four common evaluation systems for acute kidney injury, the conventional criteria has the highest specificity for predicting short-term prognosis of patients with ACLF, while the AKIN and KDIGO criteria have the highest sensitivity for the presence of acute kidney injury, especially at the early stage.