Predicting mortality in intensive care patients with acute renal failure treated with dialysis.

Existing prognostic methods were compared in their ability to predict mortality in intensive care unit (ICU) patients on dialysis for acute renal failure (ARF). The clinical goal of this study was to determine whether these models could iden- tify a group of patients where dialysis would provide no benefit because of a near 100% certainty of death even with dialysis treatment. This retrospective cohort study included 238 adult patients who received a first dialysis treatment for ARF in the ICU. This study examined the performance of seven general ICU mortality prediction models and four mortality prediction models developed for patients with ARF. These models were assessed for their ability to discriminate mortality from sur- vival and for their ability to calibrate the observed mortality rate with the expected mortality rate. The observed inhospital mortality was 76% for our patient group. Areas under the receiver operating characteristic curve ranged from 0.50 to 0.78. With the Acute Physiology and Chronic Health Evalua- tion (APACHE) III and the Liano models, the observed mor- tality in the highest quintiles of risk were 97% and 98%. In conclusion, although none of the models examined in this study showed excellent discrimination between those patients who died in hospital and those who did not, some models (APACHE III, Liano) were able to identify a group of patients with a near 100% chance of mortality. This indicates that these models may have some use in supporting the decision not to initiate dialysis in a subgroup of patients. (1 Am Soc Nephrol 8: 111-117, 1997)