Predicting patient outcome from acute renal failure comparing three general severity of illness scoring systems

A major problem of studies on acute renal failure (ARF) arises from a lack of prognostic tools able to express the medical complexity of the syndrome adequately and to predict patient outcome accurately. Our study was thus aimed at evaluating the predictive ability of three general prognostic models [version II of the Acute Physiology and Chronic Health Evaluation (APACHE II), version II of the Simplified Acute Physiology Score (SAPS II), and version II of the Mortality Probability Model at 24 hours (MPM24 II)] in a prospective, single-center cohort of patients with ARF in an intermediate nephrology care unit.

Four hundred twenty-five patients consecutively admitted for ARF to the Nephrology and Internal Medicine Department over a five-year period were studied (272 males and 153 females, median age 71 years, interquartile range 61 to 78, median APACHE II score 23, interquartile range 18 to 28). Acute tubular necrosis (ATN) accounted for 68.7% (292 out of 425) of patients. Renal replacement therapies (hemodialysis or continuous hemofiltration) were used in 64% (272 out of 425) of ARF patients.

Observed mortality was 39.1% (166 out of 425). The mean predicted mortality was 36.2% with APACHE II (P = 0.571 vs. observed mortality), 39.3% with SAPS II (P = 0.232), and 45.1% with MPM24 II (P < 0.0001). Lemeshow-Hosmer goodness-of-fit C and H statistics were 15.67 (P = 0.047) and 12.05 (P = 0.15) with APACHE II, 32.53 (P = 0.0001), 39.8 (P = 0.0001) with SAPS II, 21.86 (P = 0.005), and 20. 24 (P = 0.009) with MPM24 II, respectively. Areas under the receiver operating characteristic (ROC) curve were 0.75, 0.77, and 0.85, respectively.

The APACHE II model was a slightly better calibrated predictor of group outcome in ARF patients, as compared with the SAPS II and MPM24 II outcome prediction models. The MPM24 II model showed the best discrimination capacity, in comparison with both APACHE II and SAPS II models, but it constantly and significantly overestimated mean predicted mortality in ARF patients. None of the models provided sufficient confidence for the prediction of outcome in individual patients. A high degree of caution must be exerted in the application of existing general prognostic models for outcome prediction in ARF patients.