Background. Early detection of acute tubular necrosis (ATN) could permit implementation of salvage therapies and improve patient outcomes in acute renal failure (ARF). The utility of single and combined measurements of urinary tubular enzymes in predicting ARF in critically ill patients has not been evaluated using the receiver‐operating characteristic (ROC) plot method.

Methods. In this prospective pilot study, 26 consecutive critically ill adult patients admitted to the intensive‐care unit were studied. Urine samples were collected twice daily for up to 7 days. ARF was defined as an increase in plasma creatinine of ≥50% and ≥0.15 mmol/l. ROC plot analysis was applied to the tubular marker data to derive optimum cut‐offs for ARF.

Results. Four of the 26 study subjects (15.4%) developed ARF. Indexed to urinary creatinine concentration, γ glutamyl transpeptidase (γGT), alkaline phosphatase (AP), N‐acetyl‐glucosaminidase (NAG), and α‐ and π‐glutathione S‐transferase (α‐ and π‐GST) but not lactate dehydrogenase (LDH) were higher in the ARF group on admission (P<0.05). γGT, and α‐ and π‐GST remained elevated at 24 h. The onset of ARF based on changes in plasma creatinine varied from 12 h to 4 days (median 36 h). ROC plot analysis showed that γGT, π‐GST, α‐GST, AP and NAG had excellent discriminating power for ARF (AUC 0.950, 0.929, 0.893, 0.863 and 0.845, respectively). The discriminating strength of creatinine clearance, while lower, was still significant (AUC 0.796). Positive and negative predictive values for ARF on admission were 67/100% for γGT, 67/90% for AP, 60/95% for α‐GST, and 67/100% for π‐GST indices. Positive and negative predictive values for ARF for creatinine clearance ≤23 ml/min were 50 and 91%, respectively. Creatinine clearances tended to be lower in ARF than in non‐ARF patients on admission (P=0.06) and were significantly lower (P=0.008) after 12 h. Plasma urea and fractional sodium excretion were unhelpful.

Conclusions. Tubular enzymuria on admission to the ICU is useful in predicting ARF. The cheapness and wide availability of automated assays for γGT and AP suggests that estimation of these enzymes in random urine samples may be particularly useful for identifying patients at high risk of ARF.