64-0.85). On balance, oliguria of four or more hours provided best any other enquiries overall combination of sensitivity and specificity (Sensitivity 52%, 95% CI 0.31-0.73; Specificity 86%, 95% CI 0.84-0.8; positive likelihood ratio 3.8, 95% CI 2.2-5.6; P < 0.0001). However, even with this degree of oliguria, the positive predictive value for AKI-Cr was only 11% (95% CI 6-19%; Table Table3).3). For all levels of oliguria, there were strong negative predictive values (all > 97%) reflecting the low daily incidence of AKI-Cr. When analyzing medical and surgical ICU admissions separately, AKI-Cr in the ICU was more common in surgical patients (Table (Table2).2). However, oliguria appeared no better at predicting AKI-Cr in surgical admissions (ROCAUC = 0.73, 95% CI 0.59-0.87, Figure Figure2)2) than in medical patients (ROCAUC = 0.

79, 95% CI 0.63-0.94, Figure Figure3).3). Similarly, there was no significant difference in predictive ability for admissions with sepsis (ROCAUC = 0.78, 95% CI 0.64-0.90, Figure Figure4).4). Estimated baseline creatinine was used to define AKI-CR in 40% of patients, this may have led to an over- or under-estimate of the incidence of AKI-Cr in these patients. However, exclusion of all individuals with estimated baseline creatinine did not improve the diagnostic ability of oliguria to predict AKI-Cr in ROC analysis with a ROCAUC of 0.71 (0.54-0.87) when considering only patients with documented baseline creatinine.Figure 1Receiver-operator characteristic analysis of the ability of varying durations of oliguria to predict RIFLE Injury (I) or more the next day.

Receiver-operator characteristic (ROC) area under the curve = 0.75, 95% confidence interval (CI) 0.64-0.85.Figure 2Receiver-operator characteristic analysis of surgical patients. Receiver-operator characteristic (ROC) area under the curve = 0.73, 95% confidence interval (CI) 0.59-0.87.Figure 3Receiver-operator characteristic analysis of medical patients. Receiver-operator characteristic (ROC) area under the curve = 0.79, 95% confidence interval (CI) 0.63-0.94).Figure 4Receiver-operator characteristic analysis of patients with a diagnosis of sepsis. Receiver-operator characteristic (ROC) area under the curve = 0.78, 95% confidence interval (CI) 0.64-0.90).Extension of the time period during which the occurrence of oliguria could predict AKI-Cr from the day preceding to the two days preceding AKI-Cr did not improve the ability of oliguria to discriminate AKI-Cr (Table (Table4)4) and the ROCAUC for this comparison was 0.

72 (0.63-0.81).Table 4Relation between length of longest episode of oliguria during an ICU day at risk (patient day without a diagnosis of RIFLE I[Cr]) and AKI-Cr in the next two daysMost AKI-Cr in the ICU occurred in the first few days of critical illness. Inclusion of repeated Batimastat measures of urine output over many days of lengthy ICU admission might lead to some stable patients at low risk of AKI being over-represented in the dataset.