Severely ill patients in acute renal failure and prognostic indexes

Background

Several predictive scores have been developed to estimate the mortality risk in acute renal failure (ARF). Yet the accuracy of scores has been shown to vary when applied to foreign populations, and locally developed indexes may become better predictors. The aim of this study was to compare the performances of three working scores – one locally developed.

Patients and methods

A total of 194 consecutive patients with ARF under intensive care were enrolled at initiation of renal replacement therapy maintained beyond 24 hours. Liaño's index (ATN-ISS), Mehta's index (PICARD), and a locally tested index (IRA-PUC) were properly applied. The discrimination and calibration of the indexes were evaluated (area under the ROC curves, goodness of fit and linear regression analysis), and were compared.

Results

The population's observed mortality was high (87.0%), with a median of four (95% CI: 3–4) failing organs at inclusion. Predicted mean mortalities were 86.8, 84.9 and 87.9 (ATN-ISS, PICARD and IRA-PUC, respectively). The areas under the ROC curves were 0.702, 0.695 and 0.755 (ATN-ISS, PICARD and IRA-PUC, respectively). Goodness-of-fit determination was acceptable for all, and the linear regression analysis R² value was 0.954, 0.448 and 0.947 (P = 0.015; P = 0.088; P = 0.001) for ATN-ISS, PICARD and IRA-PUC, respectively. Discrimination was less adequate than previously stated for all tested indexes. Contrasting with ATN-ISS and IRA-PUC, PICARD was not so well calibrated.

Conclusion

Discrepancies in characteristics and in mortality rates among populations at risk and the original one may decrease the accuracy of ARF predictive indexes, precluding their extensive use.