Acute renal failure needing dialysis in the intensive care unit and prognostic scores

Generic prognostic scores used in acute renal failure (ARF) give imprecise results; disease-specific indices applied to distinct populations or intensive care practices becomes inaccurate. The current study evaluates the adequacy of prognostic scores, in patients with severe ARF needing dialysis.

Known generic (APACHE II) and disease-specific (ATN-ISS) indices were applied to a cohort (n = 280) with ARF needing dialysis, under intensive care. Possible risk factors as causal factors, organ dysfunctions and clinical variables were examined, and a local index assembled by multivariate logistic regression analysis. Area under the receiver operating characteristics (ROC) curves evaluated the indices discriminating capacity. Goodness-of-fit testing and linear regression analysis appraised calibration. Validation was accomplished by the bootstrapping technique. The end-point was hospital mortality.

Overall mortality was 85%. Female gender < 44 years (OR: 0.29; 95% CI: 0.10-0.84), liver/obstructive biliary disease (OR: 6.03; 95% CI: 1.65-22.08), being conscious (OR: 0.49; 95% CI: 0.21-1.14), use of vasoactive drug (OR: 3.13; 95% CI: 1.25-7.83), respiratory dysfunction (OR: 5.20; 95% CI: 1.25-7.83) or sepsis (OR: 2.62; 95% CI: 1.14-6.02) were associated with outcome. Areas under the ROC curve of 0.815, 0.652 and 0.814; Goodness-of-fit test P = 0.593, P < 0.001 and P = 0.002; and linear regression R2 = 0.973, R2 = 0.526 and R2 = 0.919 for the local index, APACHE II and ATN-ISS, respectively, indicate better performance by the local index. The local index median area under the ROC curve, by bootstrapping, was 0.820 (95% CI: 0.741-0.907).

APACHE II score was inaccurate, and ATN-ISS poorly calibrated. When mortality or intensive care practices significantly deviate, local scores may better evaluate prognosis in severe ARF.