Kaplan-Meier curves for time to incident chronic kidney disease (CKD) (A), progression of CKD stage (B), and long-term mortality after cardiac surgery (C) by increase in creatinine (Cr) class. Log-rank P <.001 for all 3 parts.
Adjusted hazard ratio (HR) over time for incident chronic kidney disease (CKD) (A), progression of CKD stage (B), and long-term mortality after cardiac surgery (C) by increase in creatinine (Cr) class. Ref indicates reference.
Ishani A, Nelson D, Clothier B, Schult T, Nugent S, Greer N, Slinin Y, Ensrud KE. The Magnitude of Acute Serum Creatinine Increase After Cardiac Surgery and the Risk of Chronic Kidney Disease, Progression of Kidney Disease, and Death. Arch Intern Med. 2011;171(3):226-233. doi:10.1001/archinternmed.2010.514
Long-term outcomes after acute kidney injury remain poorly defined. We determined the association between the magnitude of creatinine increase after cardiac surgery and the risk of incident chronic kidney disease (CKD), CKD progression, and death.
We identified 29 388 individuals who underwent cardiac surgery at Veterans Affairs hospitals between November 1999 and September 2005. The magnitude of creatinine increase was defined by the percent change from baseline to peak creatinine levels after cardiac surgery and categorized as none (≤0%) or as class I, (1%-24%), II (25%-49%), III (50%-99%), or IV (≥100%). Cox proportional hazard models were used to examine the association between the magnitude of creatinine increase and outcomes.
The relative hazards for outcomes increased monotonically with greater increases in creatinine levels compared with no change in creatinine levels. The relative hazards for adverse outcomes were significantly higher immediately after the creatinine increase and attenuated over time. Three months after surgery, creatinine increase classes I, II, III, and IV were associated with a greater risk of incident CKD (hazard ratios [HRs] 2.1, 4.0, 5.8, and 6.6, respectively; all P < .01), progression of CKD stage (HRs 2.5, 3.8, 4.4, and 8.0; all P < .01), and long-term mortality (HRs 1.4, 1.9, 2.8, and 5.0; all P < .01). At 5 years, the associations were lower in magnitude: incident CKD (HRs 1.4, 1.9, 2.3, and 2.3; all P < .01), CKD progression (HRs 1.5, 1.7, 1.7, and 2.4; all P < .01), and mortality (HRs 1.0, 1.2, 1.4, and 1.8; all P < .01, except class I).
The magnitude of creatinine increase after cardiac surgery is associated in a graded manner with an increased risk of incident CKD, CKD progression, and mortality.
Acute kidney injury (AKI) is common among hospitalized individuals,1 particularly among those undergoing cardiac surgery.2 Numerous studies have demonstrated that AKI is associated in the short term with increased hospital duration,3 an increased risk for infection,4 increased hospital expenditures,3 and increased mortality.5- 9 Historically, it has been assumed that if an individual survived an episode of AKI there was no long-term sequelae.10- 13 This assumption has been recently challenged by studies suggesting an association between AKI and an increased risk of mortality and end-stage kidney disease (ESKD) after the AKI event.14- 20 However, many of these studies were unable to adjust for baseline kidney function and have assumed a constant risk over time, whereas the risk of adverse outcomes may be highest early in the period after the AKI episode. Furthermore, to our knowledge, the association between an AKI and both the incidence and the progression of kidney disease has not been comprehensively evaluated.
Using a national cohort of veterans who underwent cardiac surgery, we investigated the association between the magnitude of postoperative increases in creatinine levels and the risk of adverse outcomes, including incidence and progression of chronic kidney disease (CKD) and death. We also aimed to determine whether the association between increases in creatinine levels and these adverse outcomes varied over time.
The Veteran Affairs (VA) Surgical Quality Improvement Program (VASQIP) is a national program that was designed to improve the quality of surgical care at VA hospitals. As part of this program, nurses at each participating VA facility routinely extract comprehensive data on all veterans undergoing cardiac surgery. Using the VASQIP data set, we identified all individuals who underwent cardiac bypass, with or without placement of a concomitant valve, between November 1999 and September 2005.
Between November 1, 1999, and September 30, 2005, a total of 30 662 individuals underwent cardiac surgery at 1 of 44 VA hospitals performing cardiac surgery nationwide for which the procedure was risk abstracted by VASQIP. Among these, a number of individuals were excluded from the final cohort for preexisting ESKD (n = 165), need for acute dialysis in the 30 days before surgery (n = 123), paralysis (n = 964), and missing age (n = 22), leaving a final cohort of 29 388 individuals.
Serum creatinine measurements were obtained for all members of the cohort from the national VA Decision Support System. The last creatinine measurement obtained before, but within 30 days of, cardiac surgery was used to calculate the baseline estimated glomerular filtration rate (eGFR). The baseline eGFR, calculated using the 4-variable Modification of Diet in Renal Disease formula,21 was used to categorize each individual's stage of CKD with a modified version of the National Kidney Foundation staging system.22 Chronic kidney disease was considered present if the eGFR was less than 60 mL/min/1.73m2. Among individuals with CKD, we further categorized the stage of kidney function as follows: eGFR value, 45 to 59 mL/min/1.732 (stage 3a); 30 to 44 mL/min/1.732 (stage 3b); and less than or equal to 29 mL/min/1.732 (stages 4 and 5). Individuals with stage 4 or 5 CKD at baseline were combined for the current analyses, as there were few individuals with stage 5 CKD who were not receiving dialysis and who underwent cardiac surgery.
We extracted all serum creatinine values obtained over the first 7 days after cardiac surgery. The magnitude of creatinine change was determined based on the percent change in serum creatinine values comparing the peak creatinine value with the baseline creatinine value. We defined the following classes of postoperative percent increases in creatinine values ([peak/baseline – 1] × 100): none (≤0%), class I (1%-24%), class II (25%-49%), class III (50%-99%), and class IV (≥100%).
For all individuals, all outpatient serum creatinine values, measured from 30 days after their cardiac surgery until the end of the study follow-up period [September 30, 2008]), were extracted from the VA Decision Support System database. Each individual had to have a minimum of 2 outpatient creatinine values in the follow-up period starting 30 days after surgery.
A moving average eGFR was determined for each individual. For this moving average eGFR, all eGFR values starting 30 days after surgery were used to identify the first eGFR that classified an individual into a lower CKD stage. Starting on the date of this first lower eGFR value, we then went forward and determined the average eGFR over at least 3 months (at least 2 creatinine measurements were required). If the average eGFR over this period continued to categorize an individual into a lower CKD stage, the individual was considered to have progressed into a new stage of kidney disease, and the date of the first eGFR in the new stage was considered as the date of progression. Only individuals with baseline CKD were included in the progression of CKD stage analysis (n = 6725).
Individuals were considered to have developed incident CKD if their baseline eGFR was greater than or equal to 60 mL/min/1.73m2 and if, during follow-up, their moving average eGFR went below the threshold of 60 mL/min/1.73m2 consistently for at least 3 months.
Dates of death were obtained from the VA death file, which has a sensitivity of 98.3% and a specificity of 99.3% when compared against the national death index. To decrease the influence of short-term mortality associated with AKI,23 we required that individuals survive at least 30 days after surgery before cohort entry
Comorbid conditions for all individuals were identified using administrative data and classified using the Agency for Healthcare Research and Quality–Elixhauser algorithm, which clusters and categorizes administrative data into clinically homogeneous categories.24 The Agency for Healthcare Research and Quality–Elixhauser classification system does not include either myocardial infarction or cerebrovascular disease within the classification system. These specific comorbid conditions were separately identified based on the definition from the Charlson-enhanced International Classification of Diseases, Ninth Revision (myocardial infarction, 410.x, 412.x; cerebrovascular disease, 362.34, 430.x-438.x).25 Also, we extracted the prescribed medication list at the time of surgery for each individual in the final cohort from the VA Pharmacy Benefits Package. Medications were collapsed into drugs classes (eg, angiotensin-converting enzyme inhibitors) for all analyses.
Cox proportional hazard models were used to assess the association between the class of creatinine change and the development or progression of kidney disease as well as mortality, adjusted for baseline eGFR and the covariates discussed previously. To develop a proportional hazard models for each outcome, we implemented a bootstrap model selection procedure, as discussed by Austin and Tu.26 One hundred bootstrapped versions of the sample were created, and for each sample, a stepwise selection process was performed to develop a regression model. The model selection procedure used P value–based selection and retention criteria of .05 and .10, respectively, for the significance tests of individual predictors. Variables that were selected in at least 80 of the 100 models were then used for the final Cox model in addition to class of creatinine change and baseline eGFR, along with facility as a blocking variable. Potential covariates included, race, sex, age, Agency for Healthcare Research and Quality–Elixhauser comorbidities, myocardial infarction, cerebrovascular disease, baseline prescribed medications, and surgical emergency status.
The proportionality of the hazards was tested by assessing for nonzero slope in generalized linear regressions of the scaled Schoenfeld residuals on functions of time. For each outcome, the tests indicated nonproportionality. We modeled this nonproportionality using piecewise Cox proportional hazards models for the outcome, with knots at 1, 2, 3, and 6 years. Point estimates of the hazard ratios (HRs) and corresponding 95% confidence intervals (CIs) at 3, 6, 12, 24, 36, and 60 months were estimated for each class of creatinine change. We also considered an interaction between class of creatinine change and baseline eGFR. This interaction was not significant for analyses evaluating either incident CKD or CKD-stage progression. While the interaction was significant (P < .05) for death, the patterns of association were similar to those in the additive model, and for simplicity, we present the results of this simpler model.
Individuals in the final cohort had relatively well-preserved kidney function at baseline (mean [SD] eGFR, 73.7 [21.6] mL/min/1.732). In the first 7 days after cardiac surgery, the mean (SD) number of creatinine values performed per individual was 9.6 (4.5). During an average follow-up of 5.1 years beginning 30 days after cardiac surgery, the median number of outpatient creatinine measurements annually per individual in the cohort was 2.7 (25th-75th percentile, 1.7-4.1).
In the first 7 days after cardiac surgery, there was a modest increase in the mean creatinine value of the entire cohort (mean [SD] peak creatinine level, 1.46 [0.77] mg/dL [to convert to micromoles per liter, multiply by 88.4] vs mean preoperative creatinine level, 1.18 [0.42] mg/dL). Overall, in this cohort, the postoperative changes in creatinine values were modest, with 35.3% (n = 10 369), 18.2% (n = 5357), 9.5% (n = 2719), and 4.5% (n = 1334) for postoperative increases in creatinine of 1% to 24%, 25% to 49%, 50% to 99%, and greater than or equal to 100%, respectively. Table 1 lists the baseline characteristics of patients stratified by the magnitude of their postoperative creatinine increase. In general, individuals with larger increases in creatinine values (≥25%) after cardiac surgery were older; more likely to have diabetes, hypertension, anemia, tumors, congestive heart failure, chronic obstructive pulmonary disease, or peripheral vascular disease; more likely to use angiotensin-converting enzyme inhibitors or angiotensin receptor blockers; and less likely to use β-blockers.
Among the subgroup of individuals without CKD at baseline (eGFR, ≥60 mL/min/1.73m2 [n = 20 263]), an acute increase in creatinine levels after cardiac surgery was associated with incident CKD (P < .01). The incidence of CKD during the entire follow-up period was 25.1% among patients with no CKD at baseline and no increase in creatinine levels after surgery. This percentage compared with an incidence of CKD of 33.7%, 44.1%, 51.1%, and 53.4% for patients with no CKD at baseline and creatinine increase classes I, II, III, and IV after surgery, respectively (P for linear trend, <.001). The relative risk of incident CKD was highest in the 3 months after the increase in creatinine levels and attenuated over time but persisted even at 5 years (Figure 1). Figure 2 displays the relative hazards for the development of incident CKD over time, which was not constant during follow-up. After adjustment for baseline eGFR and other potential confounders, the class of creatinine increase after cardiac surgery was associated in a graded manner with the development of incident CKD at 3 months (HRs 2.1, 4.0, 5.8, and 6.6 for creatinine increase classes I, II, III, and IV, respectively, compared with no increase in creatinine values; P for linear trend and P for each class of creatinine, <.01; compared with reference group with no increase in creatinine values) (Table 2). Although attenuated, the increase in risk compared with the group without an increase in creatinine values was still present at 5 years (HRs 1.4, 1.9, 2.3, and 2.3 for creatinine increase classes I, II, III, and IV, respectively; P for linear trend and P for each creatinine class, <.01; compared with reference group with no increase in creatinine values) (Table 2).
Of those individuals with at least CKD stage 3 included in this analysis (n = 6725), each individual had, on average, 19.9 (14.8) follow-up creatinine measurements. Among the individuals with CKD at baseline without an increase in creatinine levels after surgery, the incidence of CKD stage progression was 25%. Among individuals with CKD at baseline and an increase in creatinine levels after surgery, there was a graded increase in the proportion of individuals with the development of worsening CKD stage by magnitude of creatinine increase: class I, 33.7%; class II, 44.1%; class III, 51.1%; and class IV, 53.4% (P for linear trend, <.01) (Figure 1). The relative risk of CKD stage progression was also not constant over time (Figure 2). The relative risk of CKD stage progression was greatest immediately after the episode of creatinine increase (HRs at 3 months for progression of CKD by creatinine increase classes I, II, III, and IV: 2.5, 3.8, 4.4, and 8.0, respectively; P for linear trend and P for each class, <.01; compared with reference group with no increase in creatinine values). While the relative risk of progression by AKI severity decreased over time, at no time did the risk return to baseline. For example, after 5 years, the risk of CKD stage progression was markedly attenuated but remained elevated (HRs at 5 years for CKD stage progression by creatinine increase classes I, II, III, and IV: 1.5, 1.7, 1.7, and 2.4, respectively; P for linear trend, .21; and P for each creatinine increase class, <.05; compared with reference group with no increase in creatinine values) (Table 2).
During the follow-up period of the study, 6775 individuals (23.1%) died. Increasing severity of creatinine increase was associated in a graded manner with increasing risk of death. Mortality among those without an increase in creatinine levels (censoring deaths within 30 days of surgery) was 19.5%. Among those with an increase in creatinine classes I, II, III, and IV, mortality (censoring deaths within 30 days of surgery) was 21.0%, 26.4%, 31.7%, and 33.6%, respectively; P for linear trend, <.01) (Figure 1).
The association between severity of increase in creatinine values and mortality varied significantly over time. Piecewise proportional hazard models with knots described previously were fit to model the changing relative hazards for mortality over time. In these multivariable adjusted models that included baseline eGFR, patients with creatinine increase classes I, II, III, and IV had HRs of 1.4, 1.9, 2.8, and 5.0 for mortality at 3 months (reference no creatinine increase, P for linear trend <.01; and P for each of these 3 creatinine increase classes, <.05; compared with reference group with no AKI) (Table 2). The mortality risk associated with an increase in creatinine values attenuated over time (Figure 2) but did not return to baseline for creatinine classes II, III, and IV. At 5 years, the relative hazard for mortality continued to be elevated (HRs 1.2, 1.4, and 1.8 for creatinine increase classes II, III, and IV; P for linear trend, <.01; and P for each of these 3 classes of creatinine increase, <.01; compared with reference group with no increase in creatinine values).
The magnitude of creatinine increase after cardiac surgery was associated with an increased risk of incident CKD, progression of CKD, and death. The increase in risk of these adverse outcomes was most pronounced in the first 3 to 24 months of follow-up but persisted, albeit attenuated, at 5 years after surgery. Our results also suggest that even a minimal increase in serum creatinine values (eg, 1%-24%) after cardiac surgery is associated with the development of long-term adverse outcomes, including mortality, progression of CKD, and incident CKD.
Our findings are in agreement with those reported previously in published studies. We, and others, have previously reported that hospitalized AKI was independently associated with both ESKD and mortality and that this association was modified by the presence of baseline CKD.27,28 However, both of these studies were limited by the use of administrative data that lacked sensitivity for the identification of both CKD at baseline and AKI. The group from the University of Florida16,29 has determined the risk of long-term mortality associated with AKI among individuals undergoing both cardiac and noncardiac surgery. In both studies, a graded association between severity of postoperative AKI and risk of long-term mortality was present, but of a smaller magnitude than reported herein (cardiac surgery cohort, HR for death for 100% increase in creatinine, 1.23 [95% CI, 1.06-1.42] and for 200% increase in creatinine, 2.14 [95% CI, 1.73-2.66]). However, in both of these analyses, individuals with CKD at baseline were excluded, limiting the generalizability of study results, as CKD is the one of the strongest risk factors for the development of AKI. Also, many of the individuals in these analyses did not undergo assessment of a baseline level of kidney function before surgery. For these individuals, the authors imputed a baseline eGFR of 75 mL/min/1.73m2, which may have magnified the incidence and severity of AKI episodes identified in these data sets. Hsu et al17 have also demonstrated an increased risk of both long-term ESKD and mortality among individuals with relatively preserved kidney function (eGFR, >45 mL/min/1.732) who experienced an episode of AKI requiring dialysis. Finally, another group using VA data of hospitalized patients found an increased risk of mortality by increasing the severity of AKI (HR for mortality associated with an increase in creatinine of 50%, 1.36 [95% CI, 1.34-1.38]; 100%, 1.46 [1.42-1.50]; and 200%, 1.59 [1.54-1.65]), even after adjustment for baseline kidney function.30 While previous studies are limited by their design or lack of baseline information or focused on the minority of individuals with AKI requiring dialysis, these results, together with our findings, suggest that AKI episodes after surgery or in hospitalized patients have long-term sequelae.
A novel finding of our study is that very small changes in serum creatinine levels were associated with adverse outcomes. While the association exists for all 3 outcomes that we examined, it is unclear whether these small changes in creatinine levels represent true episodes of AKI. Current classification systems for AKI ignore small changes in creatinine levels as they may represent variations in creatinine within the individual or within the laboratory measurement. However, random variations such as these should not increase the risk for adverse outcomes. Future studies should determine the source of these small variations in creatinine levels and the clinical significance of these small changes in serum creatinine levels in other populations.
While our results suggest that postcardiac surgery increases in creatinine levels are associated with higher risks of incident CKD, progression of CKD, and long-term mortality, the mechanism underlying these associations is uncertain. An AKI episode may initiate a maladaptive renal process that leads to the development and/or progression of kidney disease. Once established, CKD itself may mediate the association between the AKI episode and long-term mortality. Alternatively, the AKI episode may simply be a marker for an individual with poor renal reserve and the stress of cardiac surgery simply unmasks the lack of underlying renal reserve. Irrespective of the mechanism, an increase in serum creatinine levels after cardiac surgery appears to accurately identify a group of individuals who are at high risk for the development of incident CKD and the progression of kidney disease as well as long-term mortality.
Our results have significant clinical implications that suggest that even small increases in serum creatinine levels after cardiac surgery are associated with long-term renal complications, particularly in the first 1 to 2 years after the episode of creatinine increase. A difficulty currently in caring for individuals who are at high risk for kidney disease is identifying those likely to develop incident kidney disease or progressive kidney disease. An increase in creatinine levels after cardiac surgery appears to identify individuals who are at high renal risk. However, currently few individuals with AKI see a nephrologist in the first year after their AKI hospitalization, and many do not have follow-up serum creatinine measurements after an episode of AKI.31 Given our results, future studies of AKI should focus on long-term outcomes associated with AKI and identify factors that can ameliorate these long-term complications, particularly in the first 2 years after an AKI episode.
Our study has several strengths. First, we used a relatively homogeneous population of individuals who underwent cardiac surgery. Our results are based on a nationwide data set with exceptionally long renal follow-up; over an average follow-up period of 5 years, each individual had a median 2.7 annual outpatient creatinine measurements after cardiac surgery. Also, we were able to accurately determine the baseline level of kidney function for all members of our cohort. Our results suggest that the magnitude of creatinine increase after cardiac surgery appears to be associated in a graded manner with incident CKD, progression of kidney disease, and long-term mortality.
Finally, while other investigators have examined the risk of adverse outcomes associated with AKI, they have assumed that increases in risk were constant after the AKI episode. Our results suggest that the increased risk of adverse outcomes is greatest immediately after the episode of creatinine increase and slowly declines over time but may never disappear.
However, our study has a number of limitations. Our study results may not generalizable to the general population, as our study was derived from a VA cohort, which is older and contains fewer females compared with the general population. Also, our cohort comprised individuals who underwent cardiac surgery. Given the selection process for cardiac surgery, our cohort likely had better baseline kidney function than a comparable cohort not selected for cardiac surgery, as evidenced by our relatively high mean baseline eGFR. Furthermore, as we used an administrative data set, other clinically important measures were not available, leading to the possibility of residual confounding. Finally, our definition of incident and progressive kidney disease required an individual to have at least 3 months of a new kidney disease stage (with at least 2 creatinine measurements) before reclassification. We thus may have misclassified individuals who then either transiently or permanently regained kidney function longitudinally; however, our definition was based a priori on the National Kidney Foundation Kidney Disease Outcomes Quality Initiative requiring at least 3 months of a new CKD stage for reclassification.
An increase in creatinine levels after cardiac surgery, even of mild severity, is associated in a graded manner with a subsequent increase in the risk of incident CKD, kidney disease progression, and mortality. This increased risk is most pronounced during the 3 to 24 months after an episode of creatinine increase. Increases in creatinine levels after cardiac surgery may be a strong risk factor for both incident and progressive kidney disease. Future research should develop strategies to prevent AKI or long-term sequelae associated with AKI, particularly in the 3 to 24 months after AKI.
Correspondence: Areef Ishani, MD, MS, Section of Nephrology, Minneapolis Veterans Affairs Medical Center (111J), 1 Veterans Dr, Minneapolis, MN 55417 (firstname.lastname@example.org).
Accepted for Publication: July 1, 2010.
Author Contributions: Dr Ishani had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis. Study concept and design: Ishani, Nelson, Clothier, Nugent, and Ensrud. Acquisition of data: Ishani, Clothier, Schult, and Nugent. Analysis and interpretation of data: Ishani, Nelson, Clothier, Greer, Slinin, and Ensrud. Drafting of the manuscript: Ishani, Nelson, and Clothier. Critical revision of the manuscript for important intellectual content: Ishani, Nelson, Clothier, Schult, Nugent, Greer, Slinin, and Ensrud. Statistical analysis: Ishani, Nelson, and Clothier. Obtained funding: Ishani, Nelson, and Clothier. Administrative, technical, and material support: Clothier, Schult, Nugent, and Greer. Institutional review board approval: Greer.
Financial Disclosure: None reported.
Funding/Support: This study was funded in part by National Institute of Diabetes and Digestive and Kidney Diseases grant 1R21DK076780 and by VA Clinical Studies Research and Development Merit Review IIR 03-295.
Role of the Sponsors: The sponsors had no role in the design and conduct of the study; in the collection, analysis, and interpretation of the data; or in the preparation, review, or approval of the manuscript.
Disclaimer: The opinions expressed are those of the authors and not necessarily those of the Department of Veterans Affairs or the US government.
Additional Contributions: We acknowledge the VA Surgical Quality Data Use Group for its role as scientific advisors and for the critical review of data use and analysis presented in this article.