Background: Severity scores are useful to guarantee similar disease severity among groups in clinical trials and to enable comparison between different studies. The aim of this study was to assess the performance of the third generation models of severity scoring systems [simplified acute physiology score (SAPS) 3, acute physiology and chronic health evaluation (APACHE) IV and mortality probability model (MPM)-III] in acute kidney injury (AKI) patients in the intensive care unit (ICU).
Methods: Three hundred and sixty-six consecutive AKI critically ill patients were prospectively assessed in six ICUs of an academic tertiary care center. Scores were applied on AKI diagnosis day (DD) and on the day of nephrology consultation (NCD). Discrimination was assessed by area under the receiver operating characteristic curve (AUCROC) and calibration by Hosmer-Lemeshow (HL) goodness-of-fit test.
Results: Hospital mortality rate was 67.8%. SAPS 3 general and Central and South America (CSA) customized equations presented identical good discrimination (AUCROC curve: 0.80 on NCD) and satisfactory HL tests on both analyzed days (P > 0.100). CSA SAPS 3 equation predicted mortality more accurately [standardized mortality ratio (SMR) on NCD = 1.00 (95% confidence interval (CI) 0.84-1.34)]. APACHE IV and MPM-III scores presented similar discrimination compared to SAPS 3 on both analyzed days (P > 0.05). APACHE IV presented satisfactory HL tests over time (P > 0.100) but underestimated mortality [SMR on DD = 1.92 (95% CI 1.61-2.23); SMR on NCD = 1.46 (95% CI 1.48-1.96)]. MPM-III showed unsatisfactory HL test results (P = 0.027 on DD; P = 0.045 on NCD) and underestimated mortality [SMR on NCD = 2.09 (95% CI 1.48-1.96)].
Conclusions: SAPS 3, especially the geographical customized equation, presented good discrimination and calibration performances, accurately predicting mortality in this group of AKI critically ill patients.