Original Article. Correspondence and offprint requests to: Matthew T. James;

Nephrol Dial Transplant (2016) 31: 2049 2056 doi: 10.1093/ndt/gfw374 Original Article Prognostic implications of adding urine output to serum creatinine measurements for staging of acute kidney injury after major surgery: a cohort study Samuel Quan 1, Neesh Pannu 2, Todd Wilson 1, Chad Ball 3, Zhi Tan 1, Marcello Tonelli 1,4, Brenda R. Hemmelgarn 1,4, Elijah Dixon 3,4 and Matthew T. James 1,4 1 Department of Medicine, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada, 2 Department of Medicine, University of Alberta, Edmonton, AB, Canada, 3 Department of Surgery, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada and 4 Department of Community Health Sciences, Cumming School of Medicine, University of Calgary, Calgary, AB, Canada Correspondence and offprint requests to: Matthew T. James; E-mail: mjames@ucalgary.ca ABSTRACT Background. Current guidelines recommend staging acute kidney injury () according to the serum creatinine (SCr) or urine output (UO) criteria that achieve the highest stage. There is little information about the implications of adding UO to SCr measurements for staging outside intensive care units and after cardiac surgery. Methods. We performed a cohort study of all adults without endstage renal disease who underwent major noncardiac surgery between January 2005 and March 2011 in Calgary, AB, Canada. Participants required at least two SCr and UO measurements to be included. We examined the implications of adding UO to SCr to stage based on Kidney Disease: Improving Global Outcomes criteria. Logistic and linear regression models were used to examine the associations between stage and 30-day mortality or hospital length of stay (LOS), respectively. Results. A total of 4229 (17%) surgical patients had sufficient SCr and UO measurements for inclusion in the cohort. The apparent incidence of postoperative substantially increased with the addition of UO to SCr criteria (8.1% with SCr alone versus 64.0% with SCr and UO). Mortality for a given stage of was lower when UO was added to SCr criteria (0.3, 3.2, 1.9 and 3.0% for no and Stages 1, 2 and 3, respectively) versus with SCr alone (1.2, 4.2, 15.4 and 12.8%). However, among participants without based on the SCr criterion, the odds of mortality and mean LOS both significantly increased with lower UO. Models that reclassified stage based on UO in addition SCr criteria had the best discrimination for mortality and LOS. Conclusions. Adding UO to SCr criteria substantially increases the apparent incidence of on hospital wards and significantly changes the prognostic implications of identification and staging. These measures should not be considered equivalent criteria in staging. Keywords: acute kidney injury, prognosis, risk stratification, surgery, urine output INTRODUCTION Acute kidney injury () is a serious complication of surgery and is associated with significant morbidity, mortality, lengthy hospitalization and large financial costs [1 4]. Major surgery can lead to through a variety of mechanisms, including intravascular volume depletion, hemodynamic instability and nephrotoxic drug exposure [5 7]. Identification of is important because it is potentially reversible [8] and recognition of its severity helps identify patients at high risk of adverse events, prompts enhanced clinical monitoring, and guides administration of fluids and avoidance of nephrotoxic exposures [5, 7]. The most recent recommendations for identification and classification of suggest that its severity should be defined according to three stages based on the highest stage achieved from one of two criteria: serum creatinine (SCr) change from baseline and decreased urine output (UO) [3, 9, 10]. Although The Author 2016. Published by Oxford University Press on behalf of ERA-EDTA. All rights reserved. 2049

ORIGINAL ARTICLE several studies have validated published classification criteria, most of these studies have focused on predicting outcomes based on SCr changes alone [1, 4, 11 25]. Relatively little research has been done to assess how the addition of UO to SCr measurements affects classification and prognosis. While UO measurement may increase detection of, more sensitive criteria for diagnosis may alter the prognostic implications of case identification [26 28]. Recent guidelines from Kidney Disease: Improving Global Outcomes (KDIGO) reported that the effect of incorporation of UO data on classification and prognosis is still uncertain and recommended further research to address this knowledge gap [3]. The purpose of this study was to investigate how the addition of postoperative UO measurements, collected during routine care on surgical hospital wards, influenced the classification of in patients who had undergone major noncardiac surgery. We also sought to compare how the addition of UO measurements to SCr criteria for influenced associations with important clinical outcomes of mortality and length of hospital stay (LOS). MATERIALS AND METHODS Study design and cohort formation We conducted a retrospective cohort study including all residents of Alberta who underwent major noncardiac surgery between 28 January 2005 and 1 March 2011 at any of three hospitals in Calgary, AB, Canada. We used each patient s unique Personal Health Number to link data sources for the study. We identified patients 18 years of age who underwent major surgery from Hospital Discharge Abstracts using Canadian Classification of Intervention (CCI) codes for vascular, retroperitoneal, abdominal, and thoracic surgeries, as previously described [29]. Patients undergoing cardiac surgery in Calgary are admitted to the Cardiac Intensive Care Unit for postoperative management and therefore were excluded. Patients were also excluded if they had end-stage renal disease prior to surgery, defined by one or more of the following: a record for chronic dialysis or a kidney transplant from the Southern or Northern Alberta Renal Programs as previously described [30], procedure codes for dialysis [International Classification of Diseases (ICD): Z49.1, Z49.2, Z99.2 or CCI: 1.PZ.21.^^] or a catheter inserted for dialysis [1.JQ.53.^^ or 1.JT.53.^^] on or prior to the day of surgery. To be eligible for inclusion, patients also required at least one SCr measured within 30 days prior to surgery, at least one follow-up SCr measurement within 7 days after surgery and at least two UO measurements within 24 h during the first 7 days after surgery. For patients who underwent multiple surgeries, only the first surgery for each patient was identified for inclusion in the study. SCr test results, for all hospitalized patients that received these measurements in Calgary. We used the KDIGO criteria to stage within 7 days of surgery [3]. Patients were categorized using four different approaches: classification 1, a 4-category system (i.e. no, stage 1, 2, or 3 based on SCr alone); classification 2, a 4-category system based on UO alone; classification 3, a 4-category system based on the highest stage of achieved from either SCr or UO criteria; classification 4, a 10-category system where stage was reclassified to a higher stage based on the addition of UO criteria to SCr criteria. For ascertainment of according to changes in SCr, we defined the baseline SCr as the closest measurement preceding surgery. Times and volumes of urine measurements charted in the electronic health record were converted to weight-based UO rates (Supplementary data 1) according to the KDIGO criteria. We standardized UO to the charted patient weight using the closest measurement preceding the time of surgery. If the patient s weight was missing we used an inputted value of the sex-specific mean weight (84.2 kg for males or 70.4 kg for females) of the cohort. We also repeated our analyses after excluding patients missing weight measurements. Outcomes We followed all patients for 30 days after surgery for the primary outcome, all-cause 30-day mortality, determined from vital statistics records from Alberta Health. The secondary outcome of the study was postoperative hospital LOS, calculated from the day of surgery until the day of discharge from the hospital. Covariates Covariates were identified from Alberta Health hospital discharge abstracts. These data capture all hospital admissions and discharges in Alberta and include up to 25 diagnostic codes per record, which were used to ascertain the presence of Charlson comorbidities [31] and hypertension [32] up to 3 years prior to hospital admission for surgery, according to validated algorithms. Additional covariates included baseline estimated glomerular filtration rate (egfr) and proteinuria. These measurements were obtained from the Alberta Kidney Disease Network (AKDN) repository of provincial laboratory data, linked using Personal Health Numbers, and categorized as previously described [30]. egfr was calculated using the Chronic Kidney Disease Epidemiology Collaboration (CKD-EPI) 2009 equation [33]. The closest measurement prior to surgery was identified as the baseline egfr. For proteinuria, the median of all urine dipstick and albumin:creatinine ratio measurements 1 year prior to surgery was taken and were categorized as normal, mild, heavy or unmeasured/missing, as previously described [30]. Identification and staging of We obtained all SCr and UO measurements from the Calgary Zone hospital electronic health record (Sunrise Clinical Manager). This electronic health record collects clinical data charted directly at the point of care by nurses. The record contains UO measurements and all laboratory results, including Statistical analysis We compared patient characteristics according to status by analysis of variance, χ 2 or Kruskal Wallis tests as appropriate. We constructed a reclassification table to tabulate the incidence of each stage of based on SCr criteria alone versus that based on the addition of UO to SCr measurements [34]. 2050 S. Quan et al.

To further evaluate how differences in the use of UO and SCr measurements influenced prognostic estimates for, we developed separate logistic regression models, each incorporating one of the four different staging approaches based on SCr and UO criteria, for predicting 30-day mortality. We fit unadjusted logistic regression models to evaluate the crude associations between stage of and 30-day mortality. We also fit multivariable logistic regression models for 30-day mortality adjusted for age, sex, baseline proteinuria, baseline egfr, each of the Charlson comorbidities and hypertension. We calculated the C-statistic to assess model discrimination and utilized the Hosmer Lemeshow test to assess the calibration of these models. We also fit four separate linear regression models incorporating the different approaches to staging to estimate LOS. We applied a logarithmic transformation to normalize the distribution of LOS and fit both unadjusted and adjusted linear regression models. Coefficients from these models were exponentiated to report mean differences (Δ) inthelosin days with 95% confidence intervals (CIs), with patients without forming the reference group. Model fit was assessed using R 2 values. In subsequent subgroup analyses we stratified the results by the type of surgery (abdominal versus other types). Statistical analyses were carried out using STATA/MP v.11 (StataCorp, College Station, TX, USA) and SAS v.9.3 (SAS, Cary, NC, USA) programs. The University of Calgary Conjoint Health Research Ethics Board approved the study. RESULTS Cohort formation and baseline characteristics A total of 4229 eligible patients underwent noncardiac surgery between 28 January 2005 and 1 March 2011 and were included in the final study cohort (Figure 1). Characteristics of the cohort, stratified by stage according to SCr criteria alone or the highest stage from the combination of SCr and UO criteria are shown in Table 1. Mortality associated with stages Based on SCr measurements alone (classification 1), 341 patients (8.1%) were identified with. However, when was identified by either SCr or UO criteria (classification 3), 2707 patients (64.0%) were identified. The risk of 30-day mortality increased with higher stages of based on SCr criteria alone (classification 1), from 1.2% in those without to 4.2, 15.8 and 12.8% for those with stages 1, 2 and 3, respectively. In comparison, the risk of 30-day mortality across the three stages based on the combination of SCr and UO (classification 3) ranged from 0.3% in those without to 3.2, 1.9 and 3.0% for those with stages 1, 2 and 3, respectively. Compared with those without, patients with stage 1, 2 or 3 had higher unadjusted and adjusted 30-day mortality either when was based on SCr criteria alone (classification 1), UO criteria alone (classification 2) or when based on the combination of SCr and UO criteria (classification 3) (Table 2). The discrimination of models was comparable when was based on SCr criteria alone (classification 1), UO criteria alone (classification 2) or the combination of SCr and UO criteria (classification 3) (C-statistic: 0.88, 0.86 and 0.88, respectively, for adjusted models). Models for 30-day mortality based on SCr criteria alone, UO criteria alone or the combination of SCr and UO criteria were well calibrated (Hosmer Lemeshow P-value >0.10). Cross-tabulation of stages according to SCr measurements alone versus the combination of SCr and UO measurements showed a large portion of patients reclassified into a higher stage of when UO measurements were considered in addition to SCr changes (Table 3). The majority of the difference in staging resulted from 2366 (55.9%) of the patients ORIGINAL ARTICLE FIGURE 1: Cohort formation. Adding urine output for acute kidney injury 2051

ORIGINAL ARTICLE Table 1. Baseline characteristics of patients that underwent noncardiac surgery according to stage, determined using either SCr criteria only or based on the highest stage ascertained from SCr and UO criteria in combination SCr criteria UO and SCr criteria No (n = 3888) Stage 1 (n = 263) Stage 2 (n = 39) without based on SCr changes being reclassified as having when UO measurements were added. Of 302 patients with stage 1 or 2 based on SCr alone, 233 (77.2%) were reclassified into a higher stage (stage 2 or 3 ) when UO measurements were also considered. The odds of 30-day mortality changed significantly across the categories when adding the UO stages to the SCr stages to form a 10-category classification system (classification 4) (Table 3). Among those without based on SCr criteria, there was a graded increase in both unadjusted and adjusted 30-day mortality with increasing stage with the addition of UO criteria. Compared with no by SCr or UO criteria, the relative odds of unadjusted and adjusted mortality associated with Stage 3 UO criteria were 10- and 5-fold greater, respectively. Mortality was highest among those with based on the combination of the highest stages of both SCr and UO criteria, with up to 50 70-fold increases in unadjusted mortality and approximately a 20-fold increase in Stage 3 (n = 39) P-value No (n = 1522) Stage 1 (n = 187) Stage 2 (n = 1319) Stage 3 (n = 1201) Female, n (%) 1782 (45.8) 102 (38.8) 17 (43.59) 13 (33.3) 0.064 785 (51.6) 73 (39.0) 572 (43.4) 484 (40.3) <0.001 Age, years, mean (SD) 59.3 (17.5) 67.0 (15.3) 70.3 (12.2) 70.5 (12.2) <0.001 54.8 (18.0) 58.2 (18.0) 60.9 (16.9) 66.1 (15.0) <0.001 Weight, kg, mean (SD) 77 (19.5) 81.3 (20.3) 87.3 (24.3) 79.9 (20.3) <0.001 73.0 (17.4) 79.4 (19.1) 76.5 (19.2) 83.4 (21.2) <0.001 Missing weight, n (%) 788 (20.3) 41 (15.6) 4 (10.3) 9 (23.1) 0.116 328 (21.6) 43 (23.0) 275 (20.9) 196 (16.3) 0.003 Baseline egfr, ml/min/ 86.0 (24.0) 73.1 (26.4) 73.0 (25.3) 58.6 (34.9) <0.001 91.0 (23.6) 83.8 (26.4) 84.2 (23.5) 77.8 (24.8) <0.001 1.73 m 2, mean (SD) Baseline proteinuria, n (%) <0.001 0.001 None 1746 (44.9) 115 (43.7) 21 (53.9) 17 (43.6) 664 (49.4) 85 (45.5) 592 (44.9) 558 (46.5) Mild 224 (5.8) 27 (10.3) 3 (7.7) 4 (10.3) 78 (5.1) 12 (6.4) 78 (5.9) 90 (7.5) Heavy 67 (1.7) 14 (5.3) 1 (2.6) 5 (12.8) 28 (1.8) 9 (4.8) 17 (1.3) 33 (2.8) Missing measurement 1851 (47.6) 107 (40.7) 14 (35.9) 13 (33.3) 752 (49.3) 81 (43.3) 632 (47.9) 520 (43.3) Comorbidities, n (%) Cancer, no metastasis 1637 (42.1) 133 (50.6) 19 (48.7) 22 (56.4) 0.013 566 (37.2) 52 (27.8) 570 (43.2) 623 (51.9) <0.001 Cancer, with metastasis 536 (13.8) 33 (12.6) 6 (15.4) 8 (20.5) 0.592 175 (11.5) 12 (6.4) 191 (14.5) 205 (17.1) <0.001 Cerebrovascular disease 310 (8.0) 26 (9.9) 5 (12.8) 4 (10.3) 0.460 116 (7.6) 14 (7.5) 111 (8.4) 104 (8.7) 0.749 Chronic heart failure 270 (6.9) 45 (17.1) 8 (20.5) 9 (23.1) <0.001 60 (3.9) 15 (8.0) 102 (7.7) 155 (12.9) <0.001 Chronic obstructive 805 (20.7) 74 (28.1) 10 (25.6) 13 (33.3) 0.008 273 (17.9) 39 (20.9) 308 (23.4) 282 (23.5) 0.001 pulmonary disease Dementia 127 (3.3) 12 (4.6) 3 (7.7) 1 (2.6) 0.311 35 (2.3) 6 (3.2) 43 (3.3) 59 (4.9) 0.003 Diabetes, complicated 218 (5.6) 42 (16.0) 7 (18.0) 10 (25.6) <0.001 61 (4.0) 14 (7.5) 81 (6.1) 121 (10.1) <0.001 Diabetes, uncomplicated 443 (11.4) 37 (14.1) 8 (20.5) 5 (12.8) 0.190 153 (10.0) 22 (11.8) 143 (10.8) 175 (14.6) 0.002 HIV/AIDS 5 (0.1) 0 (0.0) 0 (0.0) 0 (0.0) 0.936 0 (0.0) 0 (0.0) 3 (0.2) 2 (0.2) 0.309 Mild liver disease 131 (3.4) 11 (4.2) 2 (5.1) 3 (7.7) 0.406 48 (3.2) 7 (3.7) 50 (3.8) 42 (3.5) 0.825 Moderate or severe liver 32 (0.8) 5 (1.9) 2 (5.1) 0 (0.0) 0.011 7 (0.5) 3 (1.6) 12 (0.9) 17 (1.4) <0.001 disease Myocardial infarction 237 (6.1) 36 (13.7) 5 (12.8) 6 (15.4) <0.001 61 (4.0) 10 (5.4) 91 (6.9) 122 (10.2) <0.001 Paraplegia or hemiplegia 48 (1.2) 5 (1.9) 3 (7.7) 2 (5.1) 0.001 16 (1.1) 3 (1.6) 22 (1.7) 17 (1.4) 0.551 Peptic ulcer disease 264 (6.8) 27 (10.3) 4 (10.3) 5 (12.8) 0.069 75 (4.9) 8 (4.3) 97 (7.4) 120 (10.0) <0.001 Peripheral vascular disease 665 (17.1) 88 (33.5) 12 (30.8) 18 (46.2) <0.001 230 (15.1) 49 (26.2) 257 (19.5) 247 (20.6) <0.001 Rheumatic disease 112 (2.9) 9 (3.4) 2 (5.1) 0 (0.0) 0.551 46 (3.0) 4 (2.1) 38 (2.9) 35 (2.9) 0.926 Hypertension 390 (10.0) 34 (12.9) 5 (12.8) 7 (18.0) 0.172 147 (9.7) 19 (10.2) 137 (10.4) 133 (11.1) 0.689 Type of surgery, n (%) <0.001 <0.001 Vascular 540 (13.9) 81 (30.8) 11 (28.2) 15 (38.5) 213 (14.0) 46 (24.6) 206 (15.6) 182 (15.2) Retroperitoneal 163 (4.2) 13 (4.9) 1 (2.6) 3 (7.7) 92 (6.0) 17 (9.1) 42 (3.2) 29 (2.4) Abdominal 2836 (72.9) 156 (59.3) 25 (64.0) 19 (48.7) 1092 (71.8) 118 (63.0) 939 (71.2) 887 (73.9) Thoracic 349 (9.0) 13 (4.9) 2 (5.1) 2 (5.1) 125 (8.2) 6 (3.2) 132 (10.0) 103 (8.6) Percentages reflect the proportion of patients within the specified stage with the characteristic being described. P-value adjusted mortality compared with those with no by either criterion. Compared with the models based on SCr criteria alone (classification 1), UO criteria alone (classification 2) or based on the highest stage achieved by the combination of SCr and UO criteria (classification 3), the models using categorized with both UO and SCr criteria separately (classification 4) showed superior discrimination (C-statistic 0.90 for the adjusted model). The model for 30-day mortality using stages categorized based on the addition of UO criteria to SCr criteria was well calibrated (Hosmer Lemeshow P-value >0.10). LOS associated with stages The LOS increased in a graded manner with higher stages defined using SCr criteria only (classification 1), UO criteria only (classification 2) and when based on the highest stage from the combination of SCr and UO criteria (classification 3) (Table 4). When examining the 10-category system for defining 2052 S. Quan et al.

Table 2. Unadjusted and adjusted odds ratios for 30-day mortality associated with stages determined based on serum creatinine (SCr) criteria only, urine output (UO) only or the highest stage ascertained from SCr and UO criteria in combination. The results displayed represent odds ratios with 95% confidence intervals Unadjusted estimates Adjusted estimates a SCr criteria UO criteria UO and SCr criteria SCr criteria UO criteria UO and SCr criteria No [Reference] [Reference] [Reference] [Reference] [Reference] [Reference] Stage 1 8.40 (5.09 13.87) 2.07 (0.45 9.44) 12.58 (3.52 45.00) 5.27 (3.05 9.10) 1.55 (0.33 7.39) 9.90 (2.66 36.80) Stage 2 14.54 (5.82 36.33) 2.64 (1.29 5.40) 7.33 (2.54 21.12) 8.62 (3.14 23.66) 1.75 (0.83 3.68) 4.89 (1.67 14.33) Stage 3 11.76 (4.41 31.38) 5.92 (3.06 11.47) 16.14 (5.81 44.86) 5.00 (1.65 15.17) b 2.84 (1.41 5.70) 7.85 (2.76 22.33) C-statistic 0.68 0.68 0.71 0.88 0.86 0.88 Values are given as odds ratio (95% CI). a Models were adjusted for sex, age, type of surgery, baseline proteinuria, baseline egfr, each of the Charlson comorbidities and hypertension. b There was no statistically significant difference in adjusted estimates of 30-day mortality between Stage 2 and Stage 3 incorporating SCr criteria (P = 0.745). Table 3. Using a combination of SCr and UO criteria to predict 30-day mortality 30-day mortality UO and SCr measures SCr measures No Stage 1 Stage 2 Stage 3 No Number of participants 1522 131 1233 1002 Number (%) of events 4 (0.26) 1 (0.76) 18 (1.46) 25 (2.50) Unadjusted odds ratio (95% CI) [Reference] 2.92 (0.32 26.31) 5.62 (1.90 16.66) 9.71 (3.37 27.99) Adjusted odds ratio (95% CI) [Reference] 2.38 (0.26 22.07) 3.80 (1.26 11.50) 5.10 (1.72 15.12) Stage 1 Number of participants 56 73 134 Number (%) of events 5 (8.93) 5 (6.85) 15 (11.19) Unadjusted odds ratio (95% CI) 37.21 (9.70 142.67) 27.90 (7.33 106.25) 47.84 (15.63 146.41) Adjusted odds ratio (95% CI) 29.95 (7.22 124.17) 13.97 (3.43 56.89) 17.72 (5.50 57.10) Stage 2 Number of participants 13 26 Number (%) of events 2 (15.38) 4 (15.38) Unadjusted odds ratio (95% CI) 69.00 (11.43 416.59) 69.00 (16.21 293.68) Adjusted odds ratio (95% CI) 25.36 (10.10 422.86) 22.10 (4.66 104.95) Stage 3 Number of participants 39 Number (%) of events 5 (12.82) Unadjusted odds ratio (95% CI) 55.81 (16.21 217.01) Adjusted odds ratio (95% CI) 17.87 (4.13 77.41) Adjusted models included covariates for age, sex, type of surgery, baseline proteinuria, baseline egfr, each of the Charlson comorbidities and hypertension. Table 4. Unadjusted and adjusted differences in mean LOS associated with stages determined using SCr only, UO only or the highest stage ascertained from SCr and UO in combination. The results displayed below represent the difference in mean hospital LOS in days with 95% confidence interval Unadjusted estimates Adjusted estimates a SCr criteria UO criteria UO and SCr criteria SCr criteria UO criteria UO and SCr criteria No [Reference] [Reference] [Reference] [Reference] [Reference] [Reference] Stage 1 1.65 (1.49 1.84) 0.98 (0.85 1.13) 1.07 (0.95 1.21) 1.48 (1.34 1.64) 0.98 (0.86 1.12) 1.09 (0.97 1.22) Stage 2 2.28 (1.75 2.99) 1.61 (1.52 1.71) 1.62 (1.53 1.72) 1.82 (1.42 2.33) 1.49 (1.41 1.58) 1.48 (1.40 1.57) Stage 3 2.87 (2.19 3.75) 2.17 (2.04 2.30) 2.22 (2.09 2.36) 2.37 (1.85 3.04) 1.86 (1.75 1.97) 1.87 (1.76 1.98) R 2 0.046 0.140 0.142 0.145 0.238 0.254 Values are given as the difference in the mean hospital LOS (95% CI), in days. a Models were adjusted for sex, age, type of surgery, baseline proteinuria, baseline egfr, each of the Charlson comorbidities and hypertension. ORIGINAL ARTICLE, the unadjusted and adjusted LOS increased significantly across the categories based on the addition of UO staging criteria to SCr criteria (classification 4) (Table 5). Compared to those without by SCr or UO criteria, the unadjusted and adjusted differences in mean LOS associated with Stage 3 UO criteria were 2.07 and 1.77 days greater, respectively. The mean difference in LOS was greatest among those with based on Stage 2 or 3 SCr criteria and Stage 3 UO criteria, with up to a 3.25-day increase in adjusted mean LOS. The best model fit was observed for models that used categories Adding urine output for acute kidney injury 2053

ORIGINAL ARTICLE Table 5. Using a combination of SCr and UO criteria to predict LOS LOS after surgery based on classification 4, the addition of UO staging criteria to SCr criteria (R 2 = 0.271 for the fully adjusted model). Findings were similar when the analyses were stratified by surgery type (abdominal versus other surgeries; vascular, retroperitoneal and thoracic) (Supplementary data 2). Similar results were seen in sensitivity analyses when patients with missing weights were removed from the cohort (Supplementary data 3). DISCUSSION UO and SCr measures SCr measures No Stage 1 Stage 2 Stage 3 No LOS, mean (SD) 7.75 (18.3) 6.02 (8.22) 10.06 (12.33) 12.81 (14.78) Unadjusted difference in mean LOS (95% CI) [Reference] 0.96 (0.78 1.18) 1.59 (1.44 1.75) 2.07 (1.86 2.29) Adjusted difference in mean LOS (95% CI) [Reference] 1.02 (0.79 1.21) 1.46 (1.33 1.60) 1.77 (1.60 1.96) Stage 1 LOS, mean (SD) 9.45 (12.57) 13.93 (16.22) 18.77 (18.83) Unadjusted difference in mean LOS (95% CI) 1.39 (1.04 1.86) 2.13 (1.64 2.77) 2.97 (2.42 3.63) Adjusted difference in mean LOS (95% CI) 1.44 (1.04 1.99) 1.91 (1.44 2.54) 2.35 (1.89 2.93) Stage 2 LOS, mean (SD) 18.92 (16.86) 24.12 (30.13) Unadjusted difference in mean LOS (95% CI) 2.60 (2.17 4.52) 3.53 (2.35 5.31) Adjusted difference in mean LOS (95% CI) 2.38 (1.24 4.60) 2.55 (1.59 4.08) Stage 3 LOS, mean (SD) 33.31 (46.09) Unadjusted difference in mean LOS (95% CI) 3.99 (2.86 5.58) Adjusted difference in mean LOS (95% CI) 3.25 (2.21 4.80) Adjusted models included covariates for age, sex, type of surgery, baseline proteinuria, baseline egfr, each Charlson comorbidity and hypertension. We examined the implications of using routinely collected UO measurements in combination with and in addition to SCr measurements to stage and characterize prognosis in patients who had undergone major noncardiac surgery. We found that the addition of UO measurements to SCr changes resulted in a large increase in the apparent incidence of (8.1% using the SCr criterion only versus 64.0% based on the highest stage by SCr or UO). Compared with patients without, those with defined using the SCr-only criterion, UO-only criterion or the combination of SCr and UO criteria had higher risks of 30-day mortality and postoperative LOS. When UO staging was used in addition to SCr staging to form a 10-stage system to reclassify, significant increases in 30-day mortality and LOS were observed among patients with a similar SCr stage. Our findings confirm the clinical intuition that adding routine UO measurement to SCr changes for staging can help to identify patients with poor prognosis after major noncardiac surgery. Several studies have examined the predictive validity of staging systems; however, most have focused on staging based on SCr alone and the few that have assessed the implications of including UO measurements have been restricted to intensive care unit (ICU) settings [1, 11 20]. To our knowledge, no study has examined the addition of UO measurements for staging on hospital wards, where postoperative UO is also commonly monitored [1, 3]. Five studies have compared staging methods based on SCr with and without UO measurements in the ICU [26 28, 35, 36]. These studies have shown that the addition of UO data may alter the diagnosis and prognosis of. Oliguria without SCr elevation was found to be a common occurrence [26 28, 36], and the addition of UO to SCr was also found to change the apparent risk of mortality associated with a given stage of [27, 28, 36]. Our study extends these findings by including novel information relevant to the interpretation of UO measurements captured routinely on hospital wards after surgery for identifying and considering patient prognosis. One of the rationales for incorporating UO measurements into staging is based on the observation that UO may decrease before or in the absence of a detectable increase in SCr, suggesting that UO can serve as a more sensitive and early indicator of than SCr alone [3]. Our findings support this claim, as the addition of UO classified many more patients with, including patients at higher risk of 30-day mortality independent of the SCr change. Importantly, our findings also illustrate that SCr and UO criteria for staging carry different prognostic implications and should not be considered equivalent or interchangeable for staging. Our findings suggest that prognosis is best determined by categorizing based on UO and SCr changes as separate variables rather than considering the stages achieved by SCr and UO measurements as equivalent. There are limitations to our study. First, because we restricted the cohort to patients with two or more SCr and UO measurements after surgery, our study likely selected for patients with a higher acuity of illness and greater risk of. This is especially significant as only 17% of postoperative patients otherwise eligible for our study had sufficient UO and SCr measurements to define, which may lead to a higher apparent incidence of in our study. Future prospective studies incorporating standardized UO and SCr measurements 2054 S. Quan et al.

on hospital wards are required to confirm this finding. Second, we were unable to account for diuretic use, which may influence UO. Third, UO may be prone to misclassification, which is an inherent limitation of such measurements obtained in routine clinical care. For participants in the study who did not have urinary catheters there may be missing UO collection, leading to misclassification of patients with. Although urinary catheters may improve measurement accuracy, they may lead to unintended consequences, including infection, and thus are often avoided or removed when possible on hospital wards. Dementia was more common in patients with based on UO, and these patients may have been misclassified because they were incontinent and unable to participate in UO collection. Sicker or higher-risk patients may have more frequent UO measurements ordered, which may have increased the apparent incidence of based on UO in this study. Furthermore, some of the urine collections occurred beyond a 6-h interval, potentially affecting the diagnostic accuracy of UO measurements. For example, measurement of UO was frequently completed once every nursing shift, which often does not align with the KDIGO criteria of 6-, 12- or 24-h measurement intervals. Fourth, our study focused on care on hospital wards and thus we did not include UO data from the operating or recovery room, nor fluid balance, blood loss or acuity during surgery. Fifth, our study only demonstrates a correlation between classification and LOS. Physicians may be inclined to delay discharge should UO decline or SCr rise, which could confound these relationships. Sixth, there were a relatively small number of deaths in our cohort, which decreased the precision of estimates and may have reduced the ability to detect a dose response relationship between stage and mortality. Finally, our study was limited to data from three hospitals and may not be generalizable to practices at other centers. However, our findings are consistent with those from studies conducted in ICUs in other centers. In conclusion, our findings illustrate that routine postoperative UO measurements can be added to SCr criteria to increase the sensitivity of identification and staging and to refine estimates of prognosis among patients who have received noncardiac surgery. This information may be useful for designing strategies for identification of high-risk patients to enroll in clinical studies and for evaluating the effectiveness of interventions based on early identification and treatment. SUPPLEMENTARY DATA Supplementary data are available online at http://ndt.oxfordjournals.org. ACKNOWLEDGEMENTS This study is based in part on data provided by Alberta Health and Alberta Health Services. The interpretation and conclusions contained herein are those of the researchers and do not necessarily represent the views of the government of Alberta or Alberta Health Services. Neither the government of Alberta nor Alberta Health or Alberta Health Services express any opinion in relation to this study. The study was funded by the Departments of Medicine and Surgery Research Development Fund, Cumming School of Medicine, University of Calgary. S.Q. was supported by an Alberta-Innovates Summer Studentship Award. M.T.J. was supported by a KRESCENT New Investigator Award from the Kidney Foundation of Canada, Canadian Society of Nephrology and Canadian Institutes for Health Research. CONFLICT OF INTEREST STATEMENT The results presented in this paper have not been published previously in whole or part. REFERENCES 1. Ricci Z, Cruz D, Ronco C. The RIFLE criteria and mortality in acute kidney injury: a systematic review. Kidney Int 2008; 73: 538 546 2. Kerr M, Bedford M, Matthews B et al. The economic impact of acute kidney injury in England. Nephrol Dial Transplant 2014; 29: 1362 1368 3. Kidney Disease: Improving Global Outcomes (KDIGO) Acute Kidney Injury Work Group. KDIGO clinical practice guideline for acute kidney injury. Kidney Int 2012; 2: 1 138 4. Hoste EA, Bagshaw SM, Bellomo R et al. Epidemiology of acute kidney injury in critically ill patients: the multinational -EPI study. Intensive Care Med 2015; 41: 1411 1423 5. Borthwick E, Ferguson A. Perioperative acute kidney injury: risk factors, recognition, management, and outcomes. BMJ 2010; 341: c3365 6. Josephs SA, Thakar CV. Perioperative risk assessment, prevention, and treatment of acute kidney injury. Int Anesthesiol Clin 2009; 47: 89 105 7. Chenitz KB, Lane-Fall MB. Decreased urine output and acute kidney injury in the postanesthesia care unit. Anesthesiol Clin 2012; 30: 513 526 8. Tang IY, Murray PT. Prevention of perioperative acute renal failure: what works? Best Pract Res Clin Anaesthesiol 2004; 18: 91 111 9. Mehta RL, Kellum JA, Shah SV et al. Acute Kidney Injury Network: report of an initiative to improve outcomes in acute kidney injury. Crit Care 2007; 11: R31 10. Bellomo R, Ronco C, Kellum JA et al. Acute renal failure definition, outcome measures, animal models, fluid therapy and information technology needs: the Second International Consensus Conference of the Acute Dialysis Quality Initiative (ADQI) Group. Crit Care 2004; 8: R204 R212 11. Bagshaw SM, George C, Bellomo R et al. A comparison of the RIFLE and N criteria for acute kidney injury in critically ill patients. Nephrol Dial Transplant 2008; 23: 1569 1574 12. Joannidis M, Metnitz B, Bauer P et al. Acute kidney injury in critically ill patients classified by N versus RIFLE using the SAPS 3 database. Intensive Care Med 2009; 35: 1692 1702 13. Levi TM, de Souza SP, de Magalhaes JG et al. Comparison of the RIFLE, N and KDIGO criteria to predict mortality in critically ill patients. Rev Bras Ter Intensiva 2013; 25: 290 296 14. Reddy NP, Ravi KP, Dhanalakshmi P et al. Epidemiology, outcomes and validation of RIFLE and N criteria in acute kidney injury () in critically ill patients: Indian perspective. Ren Fail 2014; 36: 831 837 15. Ostermann M, Chang RW. Acute kidney injury in the intensive care unit according to RIFLE. Crit Care Med 2007; 35: 1837 1843; quiz 1852 16. Basu G, Chrispal A, Boorugu H et al. Acute kidney injury in tropical acute febrile illness in a tertiary care centre RIFLE criteria validation. Nephrol Dial Transplant 2011; 26: 524 531 17. Wang HE, Muntner P, Chertow GM et al. Acute kidney injury and mortality in hospitalized patients. Am J Nephrol 2012; 35: 349 355 18. Kuitunen A, Vento A, Suojaranta-Ylinen R et al. Acute renal failure after cardiac surgery: evaluation of the RIFLE classification. Ann Thorac Surg 2006; 81: 542 546 ORIGINAL ARTICLE Adding urine output for acute kidney injury 2055

ORIGINAL ARTICLE 19. Lin CY, Chen YC, Tsai FC et al. RIFLE classification is predictive of shortterm prognosis in critically ill patients with acute renal failure supported by extracorporeal membrane oxygenation. Nephrol Dial Transplant 2006; 21: 2867 2873 20. Ahlstrom A, Kuitunen A, Peltonen S et al. Comparison of 2 acute renal failure severity scores to general scoring systems in the critically ill. Am J Kidney Dis 2006; 48: 262 268 21. Roy AK, Mc Gorrian C, Treacy C et al. A comparison of traditional and novel definitions (RIFLE, N, and KDIGO) of acute kidney injury for the prediction of outcomes in acute decompensated heart failure. Cardiorenal Med 2013; 3: 26 37 22. Chung KK, Stewart IJ, Gisler C et al. The Acute Kidney Injury Network (N) criteria applied in burns. J Burn Care Res 2012; 33: 483 490 23. Lopez-Delgado JC, Esteve F, Torrado H et al.influence of acute kidney injury on short- and long-term outcomes in patients undergoing cardiac surgery: risk factors and prognostic value of a modified RIFLE classification. Crit Care 2013; 17: R293 24. Fujii T, Uchino S, Takinami M et al. Validation of the kidney disease improving global outcomes criteria for and comparison of three criteria in hospitalized patients. Clin J Am Soc Nephrol 2014; 9: 848 854 25. Abosaif NY, Tolba YA, Heap M et al. The outcome of acute renal failure in the intensive care unit according to RIFLE: model application, sensitivity, and predictability. Am J Kidney Dis 2005; 46: 1038 1048 26. Wlodzimirow KA, Abu-Hanna A, Slabbekoorn M et al. A comparison of RIFLE with and without urine output criteria for acute kidney injury in critically ill patients. Crit Care 2012; 16: R200 27. Macedo E, Malhotra R, Bouchard J et al. Oliguria is an early predictor of higher mortality in critically ill patients. Kidney Int 2011; 80: 760 767 28. Ralib AM, Pickering JW, Shaw GM et al. The urine output definition of acute kidney injury is too liberal. Crit Care 2013; 17: R112 29. Siddiqui NF, Coca SG, Devereaux PJ et al. Secular trends in acute dialysis after elective major surgery 1995 to 2009. CMAJ 2012; 184: 1237 1245 30. Hemmelgarn BR, Clement F, Manns BJ et al. Overview of the Alberta Kidney Disease Network. BMC Nephrol 2009; 10: 30 31. Quan H, Sundararajan V, Halfon P et al. Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data. Med Care 2005; 43: 1130 1139 32. Quan H, Khan N, Hemmelgarn BR et al. Validation of a case definition to define hypertension using administrative data. Hypertension 2009; 54: 1423 1428 33. Levey AS, Stevens LA, Schmid CH et al. A new equation to estimate glomerular filtration rate. Ann Intern Med 2009; 150: 604 612 34. Janes H, Pepe MS, Gu W. Assessing the value of risk predictions by using risk stratification tables. Ann Intern Med 2008; 149: 751 760 35. Han SS, Kang KJ, Kwon SJ et al. Additional role of urine output criterion in defining acute kidney injury. Nephrol Dial Transplant 2012; 27: 161 165 36. Kellum JA, Sileanu FE, Murugan R et al. Classifying by urine output versus serum creatinine level. J Am Soc Nephrol 2015; 26: 2231 2238 Received for publication: 21.4.2016; Accepted in revised form: 17.9.2016 2056 S. Quan et al.