Results
During the study period, 9,049 patients were consecutively admitted to one of 30 ICUs. In total, 5,942 patients were excluded; of these patients, 110 were younger than 18 years old, one received renal transplantation during the past 3 months, and 95 patients had received RRT before admission to the ICU. A further 5,725 patients were excluded because their length of stay in the ICU was less than 24 hours, and 11 were excluded because of insufficient clinical recordings. Finally, 3,107 patients were enrolled. The characteristics of the whole cohort are shown in Table 1.
Comparison of Incidence of Acute Kidney Injury
AKI was diagnosed in 1,458 (46.9%) patients by using the RIFLE classification: 20.8% with Risk, 12.4% with Injury, and 13.8% with Failure. According to AKIN criteria, AKI occurred in 1,193 (38.4%) patients: 19% with stage 1, 6.6% with stage 2, and 12.8% with stage 3. When KDIGO criteria were used, AKI occurred in 1,584 (51%) patients: 23.1% with stage 1, 11.8% with stage 2, and 16% with stage 3. The KDIGO criteria were more sensitive than RIFLE (51% versus 46.9%, P <0.01) and AKIN (51% versus 38.4%, P <0.001).
A total of 259 patients received RRT within 10 days after ICU admission. According to the KDIGO and AKIN criteria, 247 of them were identified as AKI with stage 3; the other 12 patients without AKI received RRT for a number of reasons, including sepsis and drug overdose. On the basis of the RIFLE criteria, 245 patients were diagnosed with AKI: 14 with Risk, 33 with Injury, and 198 with Failure.
The KDIGO criteria identified 126 more patients with AKI than the RIFLE criteria did: 106 with stage 1, 12 with stage 2, and 8 with stage 3 (Table 2). Among them, 124 patients were identified by an increase in creatinine alone, and the other two patients received RRT. Seventy patients were defined by KDIGO as stage 3 but not as failure by RIFLE (19 with Risk, 44 with Injury, and 8 without AKI), and 49 of them received RRT.
Compared with the AKIN criteria, KDIGO diagnosed 391 more patients as having AKI; 270 of them were categorized as stage 1, 84 as stage 2, and 37 as stage 3 (Table 3). Among 391 patients, only 25 patients had chronic kidney disease. However, the median creatinine of these 391 patients on the first day of ICU admission was 118.6 μmol/L (IQR 78 to 159.7), which was much higher than the baseline: 118.6 (IQR 78 to 159.7) versus 70 (IQR 49 to 86), P <0.001.
Comparison of Outcomes
In-hospital Mortality. Crude in-hospital mortality was significantly higher for AKI patients than for non-AKI patients, regardless of the definition used: the RIFLE (27.8% versus 7%, P < 0.0001), AKIN (32.2% versus 7.1%, P < 0.0001) and KDIGO (27.4% versus 5.6%, P < 0.0001) criteria. Mortality rate of patients identified as AKI by AKIN was higher than by KDIGO or RIFLE (32.2% versus 27.4%, P = 0.006, and 32.2% versus 27.8%, P = 0.013; respectively) but did not differ significantly between RIFLE and KDIGO (27.8% versus 27.4%, P = 0.82) (Table 4).
We also compared the in-hospital mortality of patients without AKI according different criteria and found that the patients identified by KDIGO but missed by AKIN or RIFLE had higher mortality than patients with no-AKI based on KDIGO (12.8% versus 5.6%, P < 0.01; 23% versus 5.6%, P < 0.001).
The mortality rates of patients missed by the RIFLE criteria but identified by KDIGO as stage 1, stage 2, and stage 3 were 20.8%, 33.3%, and 37.5%, respectively. The mortality rates of those missed by the AKIN criteria but identified by KDIGO as stage 1, stage 2, and stage 3 were 9.6%, 19%, and 21.6%, respectively.
Length of Intensive Care Unit Stays (Alive)
In our study, length of ICU stay was longer in patients with AKI than in those without AKI, no matter which criteria were used: the RIFLE (5 versus 3,P < 0.001), AKIN (5 versus 3,P < 0.001), and KDIGO (5 versus 3,P < 0.001) criteria. For patients missed by RIFLE or AKIN but identified by KDIGO, length of ICU stay was also longer than that of patients with no-AKI based on KDIGO (5 versus 3,P < 0.01; versus 3,P < 0.01; respectively).
Predictive Ability for Mortality
Irrespectively of which definition was used, AKI was independently associated with in-hospital mortality even after adjustment for age, gender, diabetes, hypertension, chronic kidney disease, chronic heart failure, and SOFA score (without renal component) (Table 5).
For patients diagnosed as AKI by KDIGO but not by RIFLE, AKI was also an independent risk factor of in-hospital mortality (odds ratio (OR) 4.498, 95% confidence interval (CI) 3.727 to 5.429, P < 0.001) even after adjustment for age, gender, diabetes, hypertension, chronic kidney disease, chronic heart failure, and SOFA score (without renal component). Similarly, for patients identified as AKI by KDGIO but not by AKIN, AKI was an independent risk factor for mortality (OR 1.963, 95% CI 1.139 to 2.898, P < 0.01).
The area-under-ROC curves for in-hospital mortality for RIFLE, AKIN, and KDIGO criteria were 0.738 (P < 0.001), 0.746 (P < 0.001), and 0.757 (P < 0.001), respectively. Compared with the RIFLE criteria, KDIGO had greater predictive ability for in-hospital mortality (P < 0.001) (Figure 1 and Table 6). But there was no significant difference between AKIN and KDIGO (P = 0.38).
(Enlarge Image)
Figure 1.
Area under the curves for RIFLE, AKIN, and KDIGO classification schemes comparing the predictive ability of RIFLE, AKIN, and KDIGO classification schemes for in-hospital mortality. AKIN, Acute Kidney Injury Network; KDIGO, Kidney Disease: Improving Global Outcomes; RIFLE, Risk, Injury, Failure, Loss of Kidney Function, and End-stage Kidney Disease; ROC, receiver operating characteristic. RIFLE: Area Under the Curve 0.738 (95% CI 0.713–0.762, P < 0.001). AKIN: Area Under the Curve 0.746 (95% CI 0.721-0.770, P < 0.001). KDIGO: Area Under the Curve 0.757 (95% CI 0.733-0.780, P < 0.001.
Patients With Known Baseline
For patients with known baseline (n = 2,353), the rates of incidence of AKI according to RIFLE, AKIN, and KDIGO were 45.5%, 39%, and 50.6%, respectively. The KDIGO criteria were more sensitive than RIFLE (50.6% versus 45.5%, P < 0.01) and AKIN (50.6% versus 39%, P < 0.001). Compared with patients without AKI, in-hospital mortality was significantly higher for those diagnosed as AKI by the RIFLE (27.8% versus 7.3%, P < 0.001), AKIN (31.7% versus 7%, P < 0.001), and KDIGO (27.4% versus 5.7%, P < 0.001) criteria. There was no difference in AKI-related mortality between RIFLE and KDIGO (P = 0.82), but there was significant difference between AKIN and KDIGO (31.7% versus 27.4%, P =0.031). These results were identical to that of the whole study cohort.