Incidence and mortality of postoperative acute kidney injury in non-dialysis patients: comparison between the AKIN and KDIGO criteria

Abstract

Objectives This retrospective study determines whether the kidney disease: improving global outcomes (KDIGO) criteria are superior to acute kidney injury network (AKIN) criteria in detecting non-dialysis AKI events and predicting mortality in chronic kidney disease (CKD) patients after surgery. Methods Surgical patients who were admitted to the intensive care unit were enrolled. Non-dialysis AKI cases were defined using either KDIGO or AKIN creatinine criteria and stratified by CKD stages. The adjusted hazard ratios (AHRs) for in-hospital mortality are compared to those without AKI. The cumulative survival curves and the predictability for mortality are accessed by Kaplan–Meier method and calculating the area under the curve (AUC) for the receiver operating characteristic (ROC) curve, respectively. Results From a total of 826 postoperative patients, the overall in-hospital mortality rate was 11.6% (96 cases) and that for AKI according to KDIGO and AKIN criteria was 30.0% (248 cases) and 31.0% (256 cases). The cumulative survival curve stratified by CKD and AKI stages were comparable between KDIGO and AKIN criteria. The discriminative power for mortality stratified by CKD stages for KDIGO and AKIN criteria are as followed: all subjects: 0.678 versus 0.670 (both ps <0.001); non-CKD: 0.800 versus 0.809 (both ps <0.001); early-stage CKD: 0.676 versus 0.676 (both ps <0.001); late-stage CKD: 0.674 versus 0.660 (ps were <0.001 and 0.003). Conclusion The KDIGO criteria are superior to AKIN criteria in predicting mortality after surgery, especially in those with advanced CKD.

Keywords:

• Acute kidney injury
• acute kidney injury network criteria
• chronic kidney disease
• kidney disease: improving global outcomes criteria
• mortality

Introduction

Approximately 18% of patients undergoing cardiac surgery experience acute kidney injury (AKI).1^,2 AKI increases the burden of health insurance, the length of hospitalization and in-hospital, and long-term mortality.3^,4 The reported in-hospital mortality rate of patients with AKI ranged from 13% to 60%.5–7 Thus, the diagnosis of AKI and its classification is important. Up to now, three classification systems were developed for diagnosing AKI.

The first system developed by the Acute Dialysis Quality Initiative group was Risk, Injury, Failure, Loss to End stage kidney disease (RIFLE) classification.6^,8^,9 RIFLE criteria is graded by increasing the level of severity from risk, injury, failure, loss to end-stage kidney disease. The last two stages, loss and end-stage of kidney disease, were defined as clinical outcomes that correspond to a worse prognosis.9 As a predictor for mortality, there are several limitations for RIFLE classifications. First, it depended directly on obtaining baseline serum creatinine (SCr) for each patient, which did not necessarily reflect the state of renal involvement. Second, there was also no definition of which specific stage indicated patients requiring renal replacement therapy (RRT).9 Moreover, estimated glomerular filtration rate (eGFR) criteria were included in RIFLE classification. The use of eGFR criteria calculated as >25%, >50%, and >75% decrease for risk, injury, and failure stages has been questioned. GFR was estimated by SCr-based formulas, which is valid only when SCr is in equilibrium and steady-state.5

Compared to RIFLE criteria, Acute Kidney Injury Network (AKIN) classification was proposed to take into account minor SCr changes (>0.3 mg/dL) and the need for RRT.10 Besides, eGFR criteria were removed. AKIN criteria are the most restrictive of all AKI definitions, requiring a 48-h window for an acute SCr change to define AKI.11 Patients started RRT and those with SCr ≥ 4.0 mg/dL with an acute increase ≥ 0.5 mg/dL was classified as stage 3 in AKIN criteria. Kidney disease: improving global outcomes (KDIGO) proposed a new set of guidelines and a definition of AKI, based on two previous classifications.12 The notable difference from AKIN criteria are as followed: (1) time frame differences for absolute versus relative changes in SCr; (2) increase of ≥ 0.5 mg/dL in those with SCr ≥ 4.0 mg/dL was no longer required if minimum AKI threshold met; (3) inclusion of eGFR criteria for children.13 Several studies demonstrated comparisons among RIFLE, KDIGO, and AKIN criteria.5^,6^,9^,11^,14 In predicting mortality, some study showed the AKI classification systems were comparable,9^,13 whereas some experts preferred KDIGO criteria.5^,6^,11 In diagnosing AKI, it is also demonstrated the KDIGO criteria are more sensitive than AKIN and RIFLE criteria.5^,11 However, most of these studies compared AKI staging systems (RIFLE, AKIN, and KDIGO criteria) with a logistic regression model, irrespective of the relation between time and survival.5^,9^,11 Moreover, percentage rise in SCr is inadequate to diagnose AKI in those with chronic kidney disease (ACKD).14 The aim of the present study was to compare the ability to identify AKI (including ACKD) and to predict mortality between AKIN and KDIGO criteria in those after surgery, after adjustment for several confounding factors of postoperative AKI, such as advanced age, emergent operation, liver disease, a high-risk operation, and chronic airway obstructive disease.15 Besides, the discriminatory power for mortality of postoperative AKI in different CKD stages was also evaluated.

Materials and methods

Ethics

This study was conducted in accordance with the Declaration of Helsinki (2000) by the World Medical Association. The protocol was approved by the Institutional Review Board of Taoyuan General Hospital (TYGH104018) and waived the need for informed consent because there was no breach of privacy or interference with decision-making processes related to patient care. This study was investigated in a single-center, and all patients in the study were directly diagnosed and followed up at Taoyuan General Hospital.

Materials

This is a retrospective observational study of 1167 patients who underwent surgery and were admitted to the intensive care unit (ICU) Department of Taoyuan General Hospital, from January of 2006 to December of 2013. Those undergoing chronic dialysis or RRT before surgery (n = 99), those less than 20 years of age (n = 7), lack of baseline demographic data (n = 20), and those who were hospitalized for less than 48 h (n = 62) were excluded. Finally, those who underwent any radiology study using intravenous contrast within the first seven postoperative days (n = 91) and those in whom AKI developed later than 7 d after surgery were excluded (n = 62) (Figure 1). We investigated the relation between time and survival. Those undergoing dialysis were excluded because they might have prolonged short-term survival. RRT could confound AKI duration and survival. We also excluded those diagnosed as contrast-induced acute kidney injury (CIAKI), because the pathogenesis and prognosis are different from AKI due to other etiologies. Besides, CIAKI might be developed immediately before surgery, which is also different from postoperative AKI by definition. Consequently, a total of 826 eligible patients were included for analysis.

Figure 1. Data flow diagram.

Methods

Demographic data, underlying disease, the severity of sepsis, surgery categories, serial renal function, the length of stay in the ICU, the length of hospitalization, the use of diuretics and vasoactive drugs, urine output, biochemical data and clinical characteristics, and ICU organ scoring system, such as Acute Physiology and Chronic Health Evaluation II (APACHE II) score, were recorded. The use of vasoactive drugs was recorded as inotrope infusion with inotrope equivalent score.16 All of the included critically ill non-dialysis patients were followed until death or discharge. As previous study, the endpoint of the study is in-hospital mortality censored at 180 d.17 The definitions of AKI are based on the KDIGO and AKIN criteria.13 The baseline SCr, were defined as the last SCr obtained before admission, and the SCr on ICU admission and within 48 h, the initiation date for AKI and peak AKI were also recorded for KDIGO and AKIN classification. SCr was measured by Beckman Coulter AU640 analyzer. Patients with non-dialysis AKI were stratified according to the CKD stages based on proteinuria, urinalysis, SCr, and eGFR (Modification of Diet in Renal disease (MDRD4 formula)) within 3 months before admission.18 The definitions of severe sepsis and septic shock are based on the 2012 International Guidelines for the Management of Severe Sepsis and Septic Shock.19 The need for a ventilator was defined as those with hypoxic and/or hypercapnic respiratory failure who needed ventilator after surgery. A surgery was defined as emergent based on time criterion (surgery less than 12 h), on a place criterion (patients that come from emergent department) and on code criterion (anesthesiologist indicates an emergent code as part of the American Society of Anesthesiology physical status code).15^,20

Statistical analysis

Continuous variables with a normal distribution are summarized as mean ± SD unless otherwise stated. Variables with a non-normal distribution are expressed as a median, the interquartile range. Based on KDIGO and AKIN criteria, patients without AKI were used as a reference group. Student’s t test, χ2, Mann–Whitney U, or Kruskal–Wallis tests are used to determine the differences in the demographic data, underlying disease, surgery category, scoring system, the laboratory variables, and the use of vasoactive and diuretics between patients with postoperative non-dialysis AKI and the non-AKI group. The in-hospital cumulative survival function is estimated using the Kaplan–Meier method. Cumulative survival curves for postoperative patients with different KDIGO and AKIN criteria use the log-rank test. The hazard ratios (HRs) and 95% confidence intervals (CIs) for in-hospital mortality are calculated using the Cox proportional hazard model, with adjustment for age, gender, the need for a ventilator, the organ scoring systems, the surgery category, underlying diseases, emergent operation, the use of vasoactive drugs and diuretics, the severity of sepsis, and the baseline renal function. The discriminative ability of the criteria to predict mortality was determined using the area under the curve (AUC) for the receiver operating characteristic (ROC) curve. All statistical analyses were performed using SPSS software (version 17.0, SPSS Inc., Chicago, IL) and a p value of less than 0.05 was considered statistically significant.

Results

In total, 826 patients who underwent surgery were enrolled. None of these patients received organ transplantation. The mean age of the patients was 66.4 years. There were 608 (73.6%) males and 218 (26.4%) females. The overall incidence of in-hospital mortality was 11.6% (96 cases) and that for AKI according to KDIGO and AKIN criteria were 30.0% (248 cases) and 31.0% (256 cases), respectively. AKI cases were diagnosed according to SCr criteria (65.8% vs. 70.4%), urine output criteria (8.8% vs. 8.1%), or both (25.4% vs. 21.5%). Baseline demographical, clinical, laboratory characteristics, scoring systems, and surgical categories for non-AKI and non-dialysis AKI cases sorted by KDIGO and AKIN classification are shown in Table 1. Laboratory parameters, such as the serial renal function (blood urea nitrogen and SCr), and demographic parameters, such as the need for a ventilator, the use of diuretics, the severity of sepsis, underlying disease, and the proportions of emergent operation, are significantly differed between non-dialysis AKI and non-AKI cases. The incidence of AKI cases using KDIGO and AKIN criteria are as followed: stage 1: 9.3% versus 9.7%; stage 2: 9.7% versus 12.6%; stage 3: 11% versus 8.7% (Table 2). According to AKIN criteria, the in-hospital mortality rate was 7.0% in non-AKI and 21.8%, 20.2%, 27.8% in stage 1, stage 2, and stage 3, respectively, whereas the mortality were 6.9%, 16.9%, 17.5%, and 34.1% according to KDIGO criteria. The in-hospital mortality stratified by CKD stages and AKI classifications were showed in Table 3. Most of the patients underwent abdominal (52.4%) and cardiovascular surgery (26.1%).

Table 1. Baseline characteristics and demographic data.

	Non-AKI (by KDIGO) N = 578	AKI (by KDIGO) N = 248	p values	Non-AKI (by AKIN) N = 570	AKI (by AKIN) N = 256	p values
Parameters
Age (years)	65.9 ± 13.5	66.0 ± 14.0	0.913	66.6 ± 15.4	65.8 ± 14.4	0.650
Male (n)	420	188	0.197	419	189	0.498
BMI (kg/m^2)	23.3 ± 4.1	23.3 ± 4.0	0.386	23.1 ± 3.8	23.8 ± 4.1	0.194
The need for a ventilator (n)	390	197	<0.001**	385	202	<0.001**
Emergent surgery (n)	81	51	0.013**	78	54	0.321
ICU stay (days)*	9.5 (3–14)	8.5 (3–16)	0.275	9.7 (3–14)	8.5 (3–16)	0.421
Length of hospitalization (days)^+	30 (13–39)	27 (14–43)	0.132	28 (13–39)	28 (14–43)	0.875
Mean arterial pressure (mm Hg)^+	91.5 (70.0–110)	90.5 (77–105)	0.644	91.5 (70.0–110)	90.7 (77–105)	0.712
The least urine output (L/day)	2.2 ± 1.1	1.4 ± 1.1	<0.001**	2.1 ± 1.1	1.5 ± 1.1	<0.001**
Underlying disease
Diabetes mellitus (n)	152	64	0.478	147	69	0.393
Hypertension (n)	311	122	0.127	308	125	0.095
Liver cirrhosis (n)	35	15	0.569	36	14	0.383
Coronary artery disease (n)	129	70	0.043*	127	72	0.043*
CHF, NYHA (n)	495	224	0.040*	487	232	0.024*
History of malignancy (n)	59	42	0.006*	56	45	0.102
Surgery category
Cardiovascular surgery (n)	220	108	0.112	228	106	0.412
Abdominal surgery (n)	112	56	0.316	109	59	0.377
Neurosurgery (n)	39	6	0.004*	38	7	0.004*
Chest surgery (n)	35	10	0.112	33	12	0.317
Trauma (n)	40	11	0.041	40	11	0.106
Orthopedic surgery (n)	25	7	0.146	25	7	0.047*
Gynecologic surgery (n)	20	9	0.351	20	9	0.216
Multiple surgeries (n)	12	6	0.414	12	6	0.317
Scoring system
APACHE II^+	8 (4–13)	8 (3–14.5)	0.675	8 (4–13)	8 (3–14.5)	0.145
Severity of sepsis (non-sepsis/sepsis/severe   sepsis/septic shock	18/91/217/252	3/32/110/103	0.004*	18/88/215/249	3/35/112/106	0.176*
AKI classification (non-AKI/stage1/2/3)	578/0/0/0	0/77/80/91	<0.001**	506/0/0/0	0/80/104/72	<0.001**
CKD stages (non-CKD/stage 1 and 2/≥3)	104/263/211	115/96/37	<0.001**	102/263/205	117/96/43	<0.001**
Serial renal function
Baseline creatinine (0.4–1.2 mg/dL)^+	1.6 (1.0–1.8)	1.5 (0.9–1.7)	0.040*	2.0 (1.2–2.2)	1.6 (0.9–2.3)	0.019*
Creatinine at AKI onset (0.4–1.2 mg/dL)^+	–	2.2 (1.6–2.7)		–	2.2 (1.6–2.9)
Peak creatinine during admission   (0.4–1.2 mg/dL)^+	1.9 (1.3–2.0)	2.7 (1.9–3.9)	<0.001**	2.1 (1.4–2.5)	2.5 (1.9–3.7)	0.005**
Baseline BUN (8–12 mg/dL)	34.4 ± 20.2	29.2 ± 14.7	0.001**	39.4 ± 22.4	29.1 ± 14.8	<0.001**
Peak BUN during admission (8–12 mg/dL)	40.1 ± 20.7	41.1 ± 20.4	0.125	41.8 ± 21.5	39.7 ± 20.3	<0.001
Laboratory values at ICU admission, first day
Albumin (g/dL)	3.14 ± 0.81	3.15 ± 0.74	0.868	3.23 ± 0.78	3.13 ± 0.76	0.412
Hemoglobin (g/dL)	11.3 ± 1.9	11.3 ± 2.6	0.999	11.4 ± 2.0	11.3 ± 2.4	0.742
WBC count (×10^3/μL)	11.1 ± 5.7	11.8 ± 5.9	0.182	10.4 ± 5.2	118 ± 5.9	0.061
Platelets (×10^9/L)	175 ± 112	177 ± 95	0.482	161 ± 77	178 ± 104	0.180
Prothrombin time INR^+	1.2 (1.0–1.5)	1.3 (1.0–1.3)	0.578	1.3 (1.0–1.5)	1.3 (1.0–1.4)	0.560
Lactate (mg/dL)^+	4.1 (2.6–4.9)	4.0 (1.8–6.4)	0.850	4.4 (2.6–4.9)	4.0 (1.8–4.5)	0.385
Sodium (mEq/L)	140.5 ± 6.1	139.6 ± 5.4	0.055	140.5 ± 6.3	139.8 ± 6.4	0.296
Potassium (mEq/L)	4.0 ± 0.6	4.1 ± 0.7	0.680	4.0 ± 0.6	4.0 ± 0.7	0.812
Medicine prescribed at peak AKI
Use of diuretics (n)	258	142	<0.001**	252	148	<0.001**
Furosemide (n)	210	120	<0.001**	203	127	<0.001**
Bumetanide (n)	42	26	0.082	40	28	0.041*
Mannitol (n)	6	1	0.328	6	1	0.308
Use of vasoactive (n)	324	142	0.405	319	147	0.377
^^IE	6.9 ± 17.9	10.3 ± 26.5	0.407	5.6 ± 8.5	10.5 ± 5.3	0.400

Notes: +interquartile range

* p < 0.05

** p < 0.001.

The statistical comparison uses an analysis of variance (ANOVA), χ2, Mann–Whitney U, or Kruskal–Wallis tests.

BMI, body mass index; ICU, intensive care unit; AKI, acute kidney injury; CHF, congestive heart failure; NYHA, New York Heart Association classification of heart failure; BUN, blood urea nitrogen; APACHE II, acute physiology and chronic health evaluation II score; INR, international normalized ratio; WBC, white blood cells.

^{^}IE (inotrope equivalent) = (dopamine + dobutamine) + milrinone × 15 + (epinephrine + norepinephrine + isoproterenol) × 100 in μg/kg/min.

Table 2. Cross tabulation of patients classified by AKIN versus KDIGO classification.

	AKIN
	KDIGO	Without AKI	Stage 1	Stage 2	Stage 3	Total
Without AKI	563 (68.1%)	15 (1.8%)	0	0	578 (69.9%)
Stage 1	5 (0.6%)	60 (7.3%)	12 (1.4%)	0	77 (9.3%)
Stage 2	0	5 (0.6%)	75 (9.1%)	0	80 (9.7%)
Stage 3	2 (0.2%)	0	17 (2.0%)	72 (8.7%)	91 (11.0%)
Total	570 (69.0%)	80 (9.7%)	104 (12.6%)	72 (8.7%)	826 (100%)

Notes: AKI, acute kidney injury.

The number of patients classified by the AKIN or KDIGO criteria is cross-tabulated. The results are expressed as numbers and percentages.

Table 3. The in-hospital mortality stratified by CKD stages and AKI classifications.

	AKIN criteria	KDIGO criteria
Non-CKD
Non-AKI	2 (1.9%)	1 (1.0%)
Stage 1	1 (4.5%)	2 (7.4%)
Stage 2	6 (12%)	5 (12.2%)
Stage 3	10 (23.3%)	11 (23.4%)
CKD stages 1 and 2
Non-AKI	19 (7.2%)	20 (7.6%)
Stage 1	10 (25.6%)	8 (18.2%)
Stage 2	11 (26.8%)	7 (24.1%)
Stage 3	5 (31.2%)	10 (43.5%)
CKD stage ≥3
Non-AKI	19 (9.3%)	19 (9.0%)
Stage 1	5 (29.4%)	1 (16.7%)
Stage 2	4 (30.8%)	4 (40.0%)
Stage 3	6 (46.2%)	10 (47.6%)

Notes: AKI, acute kidney injury; CKD, chronic kidney disease.

The results are expressed as numbers and percentages.

Figures 2 and 3 show the in-hospital survival curve based on AKI stages and CKD stages according to KDIGO and AKIN criteria, using a log-rank test (all ps < 0.05). Compared to non-AKI group, after adjustment for confounding factors, such as age, gender, the need for a ventilator, APACHE II, surgical category, underlying diseases, emergent operation, the use of vasoactive drugs and diuretics, the severity of sepsis, the baseline renal function, the adjusted hazard ratios (AHRs) for mortality in those with non-dialysis AKI were 3.675 (2.344–5.763) (KDIGO criteria) versus 3.351 (2.130–5.271) (AKIN criteria) (both ps < 0.001).

Figure 2. The cumulative survival rate for 826 postoperative patients stratified by CKD and AKI stages using KDIGO criteria. Cumulative survival curve according to KDIGO stages for AKI in different CKD stages. (a) All subjects. (b) Non-CKD. (c) Early-stage CKD. (d) Late-stage CKD. All p values for (a–d) are <0.001.

Figure 3. The cumulative survival rate for 826 postoperative patients stratified by CKD and AKI stages using AKIN criteria. The cumulative survival curves according to AKIN stages for AKI in different CKD stages. (a) All subjects. (b) Non-CKD. (c) Early-stage CKD. (d) Late-stage CKD. p Values for (a–d) are <0.001, <0.001, 0.002, and 0.001, respectively.

The multivariate-adjusted HRs (AKIN vs. KDIGO criteria) for in-hospital mortality stratified by CKD stages, compared to non-AKI group, is as follows: non-CKD group: 3.057 versus 6.671; early-stage CKD (stage 1 and 2): 5.110 versus 5.150; and late-stage CKD (stage ≥ 3): 1.844 versus 2.330 (all ps < 0.05, except both staging systems in late stage CKD). AHRs for in hospital mortality sorted by CKD and AKI stages were shown in Table 4. Since there is no significant difference for mortality between AKIN and KDIGO criteria in different CKD stages (by log-rank test), AUC for the ROC curve for mortality in different CKD stages were investigated. As shown in Figure 4, the discriminative power for in-hospital mortality in KDIGO criteria is greater than that in AKIN criteria in late-stage CKD. The AUC of the ROC for different CKD stages are as followed: all subjects: 0.678 (0.615–0.740) versus 0.670 (0.609–0.730); non-CKD: 0.800 (0.703–0.897) versus 0.809 (0.728–0.889); early-stage CKD: 0.676 (0.584–0.768) versus 0.676 (0.586–0.765); late-stage CKD: 0.674 (0.563–0.785) versus 0.660 (0.549–0.770) (all ps < 0.001, except ps in those with AKI superimposed on late stage CKD, they were 0.001 vs. 0.003).

Figure 4. ROC curves and discriminatory power for death in critical ill non-dialysis patients after surgery for AKIN and KDIGO criteria, stratified by CKD stages. The AUC of ROC curves in different CKD stages. (a) All subjects. (b) Non-CKD. (c) Early-stage CKD. (d) Late-stage CKD. All p values for (a–c) are <0.001. p Value for (d) according to AKIN and KDIGO criteria were 0.003 versus 0.001.

Table 4. Hazard ratios for in-hospital mortality, for different stages of AKI.

	All subjects		Non-CKD		CKD stages 1 and 2		CKD stage ≥ 3
Parameter	AHR (95% CI)	p values	AHR (95% CI)	p values	AHR (95% CI)	p values	AHR (95% CI)	p values
AKI by AKIN stages
Non-AKI	–	–	–	–	–	–	–	–
Stage 1	4.233 (2.268–9.912)	0.001*	2.251 (0.611–8.288)	0.222	5.703 (2.282–14.248)	<0.001**	2.864 (0.794–8.002)	0.108
Stage 2	2.427(1.329–4.430)	0.004*	4.087 (1.807–9.247)	0.001*	4.730 (1.573–14.238)	0.006*	1.643 (0.900–3.370)	0.601
Stage 3	3.943 (2.177–7.145)	<0.001**	4.856 (1.715–13.748)	0.003*	6.367 (3.276–16.687)	<0.001**	2.053 (1.380–11.094)	0.037*
AKI by KDIGO stages
Non-AKI	–	–	–	–	–	–	–	–
Stage 1	4.496 (2.247–8.970)	0.004*	6.898 (2.614–18.211)	0.011*	6.329 (3.453–14.467)	<0.001**	3.709 (0.834–9.554)	0.287
Stage 2	2.136 (1.071–4.257)	0.011*	5.522 (0.565–7.591)	0.199	1.714 (0.934–7.450)	0.074	1.265 (0.825–5.613)	0.076
Stage 3	5.293 (3.121–8.978)	<0.001**	7.709 (4.831–12.557)	0.001*	9.320 (6.216–13.171)	<0.001**	3.831 (1.869–16.880)	0.019*

Notes: *p < 0.05

** p < 0.001.

AKI, acute kidney injury; CKD, chronic kidney disease; CI, confidence interval; AHR, adjusted hazard ratio.

Adjusted for age, gender, need for a ventilator, APACHE II, surgical category, underlying diseases, emergent operation, the use of vasoactive drugs and diuretics, severity of sepsis, and baseline renal function.

Discussion

This retrospective study demonstrates that KDIGO and AKIN criteria for AKI predict in-hospital mortality in surgical patients, after adjustment for the organ scoring systems, underlying disease, surgical category, and severity of sepsis, the baseline, and clinical characteristics. The KDIGO criteria are superior to AKIN criteria in predicting in-hospital mortality after surgery, especially in those with advanced CKD.

The validity of the AKI stage, defined by either the AKIN, RIFLE criteria, or the KDIGO classification, is a well-known predictor for mortality.5^,6^,9^,11^,14 Some study showed the overall AKIN, RIFLE, and KDIGO criteria were comparable in predicting mortality,9^,13 whereas the others showed KDIGO criteria are superior to RIFLE and AKIN criteria.5^,6^,11 Some study demonstrated that KDIGO criteria detect more incident AKI than AKIN and RIFLE criteria do,5^,11 whereas some experts do not agree with it.21^,22 Most of these studies compared AKI staging systems irrespective of relation among time, survival, and CKD stages.5^,9^,11 It is also unclear whether current staging systems to detect ACKD is appropriate or not. CKD is a well-known risk factor for AKI. Those with ACKD lead to end stage renal disease (ESRD) at a higher frequency than does AKI alone.23^,24 Both AKI and CKD are risk factors for in-hospital mortality. The predictability of AKI stages might decrease after adjusting CKD stages. Current AKI staging systems might have lower prognostic values in CKD when based on absolute increases of SCr at the lower end of the spectrum.14 Percentage rise in SCr is curtailed in CKD and is inadequate detect ACKD.14 Owing to postoperative fluid accumulation with dilutional effect on SCr levels and loss of lean body mass after surgery, postoperative SCr values were lower than preoperative levels.25^,26 For those with postoperative AKI, dilutional effect on SCr levels and loss of lean body mass might lead to minimal or no increase of SCr after surgery; thus, AKI diagnosis will be missed or delayed in a 48 h diagnostic window according to AKIN criteria, which required absolute increase of SCr ≥ 0.5 mg/dL to diagnose ACKD.25^,26 The removal of absolute increase of ≥ 0.5 mg/dL in those with SCr ≥ 4.0 mg/dL and wide time frame difference for changes in SCr in KDIGO criteria might contribute to early diagnosis of ACKD. That is why KDIGO criteria are important and are superior to AKIN criteria in this population. We compared AKI staging systems for mortality sorted by CKD stages with Cox hazard proportional model. It demonstrated that, in predicting in-hospital mortality, KDIGO criteria are better than AKIN criteria for those with ACKD.

According to KDIGO and AKIN criteria, 8.8% (22 cases) and 8.1% (26 cases) of subjects with AKI were defined by urine output criteria. AKI might be defined more comprehensive when using data for urine output. However, they are less validated and might be influenced by several drugs, such as diuretics.5^,27 The changed urine output caused by diuretics might result in an underestimate of the severity of the AKI when using the urine output criteria alone.27 Besides, RIFLE criteria were not included in our comparisons, since it included eGFR criteria. Estimated GFR is valid only when SCr is in equilibrium and steady-state, but not for AKI.5^,28 Oliguria and eGFR did not always show parallel changes in AKI. Moreover, patients who required RRT during admission were excluded, because RRT can confound AKI duration and survival.29 The proportions of non-dialysis AKI diagnosed by KDIGO criteria is more than that of AKIN criteria in late-stage CKD (11.1% vs. 8.7%), whereas the proportions is less than that of AKIN criteria in early stage CKD group. This result might be explained by the removal of absolute increase of ≥ 0.5 mg/dL in those with SCr ≥ 4.0 mg/dL in KDIGO criteria. The minimal increase of SCr less than 0.5 mL/dL might be regarded as AKI in KDIGO criteria. AKI event is not an accurate predictor of in-hospital mortality for those with AKI less than 3 d.17^,30 In this study, most of the AKI duration were less than 14 d (92.1%) and only 8.9% of the AKI duration were less than 48 h.27^,29^,30 As the previous study, the endpoint of the present study was in-hospital mortality censored at 180 d.17 However, patients might join the hospital for other reasons, especially those with prolonged hospitalization. The average length of hospitalization in similar studies was less than 30 d.17^,25^,26^,30 In the study, most of the lengths of hospitalization are less than 30 d (78.9%). Moreover, the proportions of disagreement between staging systems are higher between KDIGO and RIFLE classifications than that of KDIGO and AKIN classifications (from 3.7% to 18%).6^,9^,11^,14 The proportion of disagreement between different AKI staging systems is 6.8% in this study.

This study has several limitations. First, the retrospective study population is confined to one medical center and those who receiving non-cardiac surgery, so the result cannot be extrapolated to other patient populations. Second, the exclusion of patients less than 20 years of age, history of dialysis, and those suffered from AKI 7 d after surgery may lead to a selection bias. Additionally, we do not perform renal ultrasonography and renal biopsy during ICU admission routinely; therefore, obstructive AKI could not be excluded. The prognosis of obstructive AKI is different from acute tubular injury and acute glomerulonephritis. Third, the definition of baseline renal function is still controversy. Using a short time frame (SCr < 7 d before admission) or a longer time frame (SCr between 7 and 365 d before admission) for detection of AKI is different. The latter definition might detect more but less severe AKI events. RIFLE and KDIGO recommended either back calculation, longer time frame method, or the lowest SCr during hospitalization. The lack of uniform approach to estimate baseline renal function is one of the limitations. Finally, there may be residual confounding factors that could attenuate the relationship between AKI and in-hospital mortality, such as the duration of AKI and treatment strategies.30

In summary, this retrospective study demonstrates that both KDIGO and AKIN criteria for AKI predict in-hospital mortality in surgical patients, after adjustment for demographic and laboratory data. The KDIGO criteria are superior to AKIN criteria in predicting in-hospital mortality after surgery, especially in those with advanced CKD. The overall incidence of AKI is similar between KDIGO and AKIN criteria. This study is limited by its retrospective nature and sample size. Multi-center, larger observational cohort studies are warranted to verify the conclusion.

Acknowledgments

Special thanks to Dr. Lin-Chien Lee in statistical analysis.

Declaration of interest

The authors report no conflicts of interest.