Visual Abstract
Keywords: acute kidney injury and ICU nephrology, acute kidney injury, morbidity, postoperative period, renal dialysis, renal replacement therapy, risk assessment, web-based
Abstract
Background
AKI after surgery is associated with high mortality and morbidity. The purpose of this study is to develop and validate a risk prediction tool for the occurrence of postoperative AKI requiring RRT (AKI-dialysis).
Methods
This retrospective cohort study had 2,299,502 surgical patients over 2015–2017 from the American College of Surgeons National Surgical Quality Improvement Program Database (ACS NSQIP). Eleven predictors were selected for the predictive model: age, history of congestive heart failure, diabetes, ascites, emergency surgery, hypertension requiring medication, preoperative serum creatinine, hematocrit, sodium, preoperative sepsis, and surgery type. The predictive model was trained using 2015–2016 data (n=1,487,724) and further tested using 2017 data (n=811,778). A risk model was developed using multivariable logistic regression.
Results
AKI-dialysis occurred in 0.3% (n=6853) of patients. The unadjusted 30-day postoperative mortality rate associated with AKI-dialysis was 37.5%. The AKI risk prediction model had high area under the receiver operating characteristic curve (AUC; training cohort: 0.89, test cohort: 0.90) for postoperative AKI-dialysis.
Conclusions
This model provides a clinically useful bedside predictive tool for postoperative AKI requiring dialysis.
Introduction
Postoperative dialysis-requiring AKI is a serious medical complication of surgery, with a rising incidence and significant morbidity and mortality. Postoperative AKI increases 5-year mortality regardless of level of injury and extent of kidney recovery (1). For patients who develop AKI requiring RRT (AKI-dialysis) after elective surgery, 90-day mortality rate was reported to be 42%; among patients undergoing cardiac surgery, dialysis conferred a 30-day mortality rate of 63.7% as compared with 4.3% without AKI (2,3). Of patients who survived the episode of AKI-dialysis, 27.2% required chronic dialysis, with its attendant morbidity and cost (3,4). It is important to identify patients at high risk of postoperative AKI as preventative measures may lower the risk. Implementation of the Kidney Disease Improving Global Outcomes (KDIGO) care bundles consisting of optimization of volume status and hemodynamics, avoidance of nephrotoxic drugs, and preventing hyperglycemia showed reduction of postoperative AKI (5).
Preoperative assessment of surgical risk is valuable for determining appropriateness of surgery, facilitating multidisciplinary surgical planning, and engaging patients in shared decision making (6). Tools for cardiac and pulmonary risk stratification have been widely adopted. Even though AKI prediction tools have been reported in the literature, these tools have not been incorporated into daily clinical practice (7–10). Existing postoperative AKI models are on the basis of small datasets or are specific to patients undergoing cardiac surgery (11–15). The universal American College of Surgeons (ACS) National Surgical Quality Improvement Program (NSQIP) Surgical Risk Calculator includes postoperative renal failure as one of its eight outcomes (16). The NSQIP surgical risk calculator does not use preoperative serum creatinine as a predictor, which makes it difficult to differentiate postoperative AKI risk between patients with normal kidney function and patients with advanced CKD.
Establishing an AKI risk prediction model for surgical patients, including the risk for RRT, is an unmet clinical need. The objective of this study is to develop and validate an AKI risk calculator using data from the NSQIP database.
Materials and Methods
Study Participants
Data were obtained from the ACS NSQIP database, which collects clinical data from >500 United States hospitals, including patient demographics, medical conditions, laboratory data, and 30-day postoperative outcomes, collected by trained surgical clinical reviewers (17,18). Patients undergoing surgery at participating hospitals from 2015 to 2017 were included (19). The dataset from 2015 to 2016 was used to develop the model, whereas the dataset from 2017 was held for validation. Patients who already had an NSQIP-assessed procedure within the previous 30 days were excluded (20). The study was approved by the Thomas Jefferson University Institutional Review Board, and informed consent was waived.
Outcome
The primary outcome of interest was AKI-dialysis within 30 days of surgery, a time frame chosen on the basis of the ACS NSQIP definition. ACS NSQIP defines AKI-dialysis as “worsening of renal dysfunction postoperatively requiring hemodialysis, peritoneal dialysis, hemofiltration, hemodiafiltration, or ultrafiltration in a patient who did not require dialysis preoperatively” (20). Instructions from the organization state that if a patient refuses a recommendation for dialysis, they should still be coded as developing AKI-dialysis as dialysis was deemed necessary (20).
The secondary outcome was postoperative major adverse kidney event (Postop-MAKE), a composite of AKI-dialysis or 30-day postoperative AKI (a rise in serum creatinine of >2 mg/dl from preoperative value but with no requirement for postoperative dialysis).
Data Analyses
We analyzed previously described risk factors for AKI, including demographic factors, medical history, current medical condition (e.g., presence of sepsis and American Society of Anesthesiologists class), and preoperative functional status. Multiple preoperative laboratory tests were analyzed, including serum sodium, creatinine, platelets, and hematocrit. Participant risk factors were compared using the chi-squared test for categorical variables and the Wilcoxon test for continuous variables. Multivariable logistic regression was performed with predictor candidates to determine the significance of their associations with the outcome.
Separate categories were made for missing values of laboratory concentrations of serum hematocrit. Backward elimination with P-value cutoff of 0.01 was used to select predictor variables with lowest P values. The model was trained on 2015–2016 cohort data and tested on 2017 cohort.
GridSearchCV was used to perform five-fold crossvalidation to select the best model parameters. The dataset from year 2017 (n=811,778) was used to further test and validate the predictive model. Area under the receiver operating characteristic curve (AUC) was used for model evaluation. Sensitivity analysis was performed after removing emergency cases from the cohort. TRIPOD statement was followed for this study.
Statistical analysis was performed with Python (version 3.65), Statsmodels (version 0.8.0), and R programming (RStudio version 1.1.463). Scikit-learn was used for the web-based model.
Results
Study Population
The study included 2,299,502 patients from 2015 to 2017; 56.4% were women. Patients from 2015 to 2016 (n=1,487,724) were included in the cohort to develop the model; 811,778 patients from 2017 were included in the validation cohort. Table 1 shows the characteristics of study patients for the years from 2015 to 2017. In total, 0.3% (n=6853) of patients developed AKI-dialysis, and 0.59% (n=13,482) of patients developed Postop-MAKE.
Table 1.
Patient characteristics (2015–2017)
| Characteristics | No AKI with Dialysis, n=2,292,649 | AKI with Dialysis, n=6853 |
| Age, mean (±SD) | 58.5 (±16.4) | 66.2 (±13.6) |
| Men, % | 43.52 | 61.67 |
| American Society of Anesthesiologists class, % | ||
| 1 | 5.80 | 0.28 |
| 2 | 42.41 | 6.89 |
| 3 | 45.35 | 45.56 |
| 4 | 6.27 | 40.62 |
| 5 | 0.17 | 6.65 |
| Diabetes, % | 17.22 | 34.26 |
| Disseminated cancer, % | 2.72 | 5.92 |
| Smoker within 1 yr, % | 17.77 | 24.19 |
| Chronic obstructive pulmonary disease, % | 4.94 | 14.59 |
| Ascites, % | 0.34 | 3.71 |
| Congestive heart failure, % | 0.90 | 8.84 |
| Hypertension requiring medication, % | 49.32 | 74.20 |
| Steroid, % | 4.03 | 9.54 |
| Weight loss, % | 1.38 | 4.71 |
| Bleeding disorder, % | 4.53 | 17.82 |
| Emergency surgery, % | 10.08 | 37.84 |
| Serum sodium, mEq/L | 139.2 | 138.1 |
| Serum creatinine, mg/dl | 0.92 | 1.97 |
| White blood cell, ×109/L | 8.2 | 11.4 |
| Hematocrit, % | 39.8 | 35.3 |
| Platelet, ×109/L | 249.5 | 233.0 |
Patients who required dialysis were older (66.2 versus 58.5 years, P<0.001); had higher preoperative serum creatinine level (1.97 versus 0.92 mg/dl, P<0.001); and were more likely to have congestive heart failure (8.84% versus 0.90%, P<0.001), diabetes (34.26% versus 17.22%, P<0.001), or other comorbidities. They were more likely to have undergone emergency surgery (37.84% versus 10.08%, P<0.001). Ascites was defined by the presence of fluid accumulation in the peritoneal cavity noted on physical examination, abdominal ultrasound, or abdominal computed tomography/magnetic resonance imaging within 30 days prior to primary procedure (20).
Supplemental Figure 1 shows a flowchart of study participants. Supplemental Tables 1 and 2 show training and validation cohort patient characteristics.
Mortality, Readmission, and Length of Stay Associated with AKI
Figure 1 shows a high unadjusted 30-day postoperative mortality rate (37.5%) from 2015 to 2017 for patients who developed AKI-dialysis compared with those who did not. The rate of 30-day hospital readmission was also higher in these patients (19.7% versus 5.0%). The patients who developed AKI-dialysis also had increased length of hospital stay (18.7 versus 2.9 days).
Figure 1.
Significant differences in thirty-day unadjusted mortality, hospital readmission rate and length of stay in patients who developed AKI requiring dialysis.
AKI-dialysis, AKI requiring RRT.
Supplemental Figure 2 illustrates unadjusted 30-day mortality rates after operation among patients who developed AKI requiring dialysis according to surgery types from 2015 to 2017. These varied widely, ranging from 16.2% to 53.3%. Although patients undergoing ENT surgery had a high mortality rate, the incidence of AKI-dialysis was low at 0.03%; 1.86% of patients who underwent cardiac surgery developed AKI-dialysis, with a mortality rate of 37.9%. Another study showed that the unadjusted all-cause mortality among patients requiring RRT after cardiac surgery was 64% at 1 year (21). Figure 2 shows unadjusted rates of postoperative AKI-dialysis according to surgery types. Aorta surgery has the highest risk of AKI-dialysis at 6.05%, with a mortality rate of 41.4%. Endovascular aneurysm repair has a lower AKI-dialysis rate at 0.98% compared with aorta surgery. Other surgery types that have higher AKI risk include liver, pancreas, spleen surgery (0.82%), intestinal surgery (0.8%), vascular surgery (0.59%), and thoracic surgery (0.48%). Surgery types in lower-risk categories include orthopedic upper extremity surgery (0.05%) and spine surgery (0.09%).
Figure 2.
Differences in unadjusted rates of postoperative AKI requiring dialysis according to surgery type. AKI-dialysis, AKI requiring RRT; ENT, Ear, Nose, Throat: OB&GYN, Obstetrics and Gynecology.
Effect of Preoperative Renal Function, Serum Sodium, and Hematocrit
As shown in Figure 3, preoperative renal dysfunction was associated with increased unadjusted risk of AKI-dialysis. The risk of AKI was higher with higher preoperative serum creatinine levels. Patients with hyponatremia and hypernatremia had higher incidence of AKI-dialysis (≤130 mEq/L: 1.5%; 130.1–135 mEq/L: 0.6%; 135.1–145 mEq/L: 0.2%; >145 mEq/L: 1.0%; P <0.001). Anemia also increased the risk of AKI-dialysis, with higher unadjusted rates among patients with hematocrit <30% (hematocrit ≤24%: 2.3%; 24.1%–30%: 1.4%; >30%: 0.2%; P<0.001).
Figure 3.
Significant differences in unadjusted risk of postoperative AKI requiring dialysis according to preoperative serum creatinine, sodium, and hematocrit. AKI-dialysis, AKI requiring RRT.
Prediction of AKI-Dialysis
Table 2 lists clinical factors that were strongly associated with the occurrence of postoperative 30-day AKI-dialysis, and these variables were incorporated into the prediction model. History of congestive heart failure was a significant predictor, with increased adjusted odds ratio (OR) of the outcome of 2.3 (95% confidence interval [95% CI], 2.05 to 2.58; P<0.001). Diabetes was also associated with an increased risk (OR, 1.95; 95% CI, 1.79 to 2.12; P<0.001 if insulin dependent and OR, 1.25; 95% CI, 1.14 to 1.36; P<0.001 if noninsulin dependent), a finding consistent with a previous study (22). Preoperative conditions, such as systemic inflammatory response syndrome (OR, 2.61; 95% CI, 2.34 to 2.91; P<0.001), sepsis (OR, 3.5; 95% CI, 3.09 to 3.85; P<0.001), and septic shock (OR, 8.3; 95% CI, 7.35 to 9.38; P<0.001), were strongly associated with postoperative AKI-dialysis. Laboratory findings associated with increased risk included preoperative serum sodium and hematocrit. Serum sodium abnormalities in either direction had an increased adjusted risk of the outcome: for serum sodium ≤130 mEq/L, OR of AKI-dialysis was 1.39 (95% CI, 1.21 to 1.6; P<0.001); for serum sodium of 130.1–135 mEq/L, OR was 1.21 (95% CI, 1.12 to 1.31; P<0.001); and for serum sodium >145 mEq/L, OR was 1.48 (95% CI, 1.24 to 1.76; P<0.001). Surgery type was also a predictive factor, with the highest rates of AKI-dialysis in patients undergoing aortic surgery, followed by cardiac surgery and surgery of the liver, pancreas, or spleen. Intestinal and other vascular procedures were also strongly associated.
Table 2.
Predictors of postoperative AKI requiring dialysis
| Characteristics | Coefficient | SEM | P Value | Adjusted Odds Ratio (95% Confidence Interval) |
| Intercept | −9.228 | 0.122 | <0.001 | |
| Age, per yr | 0.012 | 0.001 | <0.001 | 1.01 (1.01 to 1.02) |
| Ascites | 0.971 | 0.089 | <0.001 | 2.64 (2.22 to 3.14) |
| Congestive heart failure | 0.832 | 0.059 | <0.001 | 2.30 (2.05 to 2.58) |
| Emergency surgery | 0.773 | 0.042 | <0.001 | 2.17 (2.00 to 2.35) |
| Hypertension requiring medication | 0.440 | 0.038 | <0.001 | 1.55 (1.44 to 1.67) |
| Diabetes, reference: no diabetes | ||||
| Insulin dependent | 0.667 | 0.043 | <0.001 | 1.95 (1.79 to 2.12) |
| Noninsulin dependent | 0.219 | 0.046 | <0.001 | 1.25 (1.14 to 1.36) |
| Serum sodium, reference: 135.1–145, mEq/L | ||||
| 0–130 | 0.328 | 0.071 | <0.001 | 1.39 (1.21 to 1.60) |
| 130.1–135 | 0.193 | 0.041 | <0.001 | 1.21 (1.12 to 1.31) |
| >145 | 0.390 | 0.090 | <0.001 | 1.48 (1.24 to 1.76) |
| Missing | −0.165 | 0.181 | 0.36 | 0.85 (0.59 to 1.21) |
| Serum hematocrit, reference: >30, % | ||||
| 0–24 | 0.802 | 0.073 | <0.001 | 2.23 (1.93 to 2.57) |
| 24.1–30 | 0.688 | 0.041 | <0.001 | 1.99 (1.83 to 2.16) |
| Missing | −0.582 | 0.167 | <0.001 | 0.56 (0.40 to 0.78) |
| Preoperative sepsis, reference: no | ||||
| SIRS | 0.959 | 0.056 | <0.001 | 2.61 (2.34 to 2.91) |
| Sepsis | 1.238 | 0.056 | <0.001 | 3.5 (3.09 to 3.85) |
| Septic shock | 2.116 | 0.062 | <0.001 | 8.30 (7.35 to 9.38) |
| Surgery type, reference: anorectal, appendix | ||||
| ENT | 0.255 | 0.372 | 0.49 | 1.29 (0.62 to 2.68) |
| Bariatric, stomach, esophagus | 1.425 | 0.127 | <0.001 | 4.16 (3.24 to 5.33) |
| Brain | 0.872 | 0.210 | <0.001 | 2.39 (1.58 to 3.61) |
| Cardiac | 2.864 | 0.141 | <0.001 | 17.53 (13.30 to 23.09) |
| Endocrine, thyroid, parathyroid, adrenal, breast | −0.427 | 0.203 | 0.04 | 0.65 (0.44 to 0.97) |
| Gallbladder, biliary tract | 0.555 | 0.140 | <0.001 | 1.74 (1.33 to 2.29) |
| Hernia, peritoneum, omentum, diverticulum | 0.920 | 0.121 | <0.001 | 2.51 (1.98 to 3.18) |
| Intestine (not rectum) | 1.690 | 0.111 | <0.001 | 5.42 (4.36 to 6.73) |
| Liver, pancreas, spleen | 2.565 | 0.126 | <0.001 | 13.00 (10.16 to 16.64) |
| OBGYN | 0.203 | 0.173 | 0.24 | 1.23 (0.87 to 1.72) |
| Orthopedic lower extremity, pelvis | 0.732 | 0.118 | <0.001 | 2.08 (1.7 to 2.62) |
| Orthopedic upper extremity, shoulder | 0.020 | 0.260 | 0.94 | 1.02 (0.61 to 1.70) |
| Skin, subcutaneous, other musculoskeletal | 0.909 | 0.129 | <0.001 | 2.48 (1.93 to 3.20) |
| Spine | 0.593 | 0.148 | <0.001 | 1.81 (1.35 to 2.42) |
| Thoracic (nonesophageal) | 1.913 | 0.149 | <0.001 | 6.77 (5.06 to 9.07) |
| Urology | 1.590 | 0.124 | <0.001 | 4.90 (3.85 to 6.25) |
| Vascular (endovascular aneurysm repair) | 2.310 | 0.151 | <0.001 | 10.07 (7.50 to 13.53) |
| Vascular (open aorta surgery) | 3.993 | 0.128 | <0.001 | 54.20 (42.19 to 69.64) |
| Vascular (other) | 1.709 | 0.119 | <0.001 | 5.53 (4.37 to 6.98) |
| Preoperative serum creatinine per 1 mg/dl, mg/dl | 0.449 | 0.008 | <0.001 | 1.57 (1.54 to 1.59) |
SIRS, systemic inflammatory response syndrome; ENT, Ear, Nose, Throat: OBGYN, Obstetrics and Gynecology.
These predictors were used to develop a model for AKI-dialysis as described in Materials and Methods. Brier score for the training cohort was 0.003. The model has a training AUC of 0.89. Sensitivity analysis after removing emergency cases from the cohort showed training AUC of 0.87 and validation AUC of 0.87. The model has sensitivity of 0.89 and specificity of 0.717 with cutoff of 0.002 and sensitivity of 0.667 and specificity of 0.908 with cutoff of 0.005.
Supplemental Figure 3A shows the calibration plot of the prediction model, which is well matched with a 45° line.
Prediction of Postop-MAKE
Supplemental Table 3 shows predictors of Postop-MAKE. Anemia had an increased risk of Postop-MAKE: for serum hematocrit ≤24%, adjusted OR was 2.04 (95% CI, 1.82 to 2.29), and for hematocrit 24.1%–30%, OR was 1.90 (95% CI, 1.79 to 2.02). Systemic inflammatory response syndrome (OR, 2.38; 95% CI, 2.2 to 2.58), sepsis (OR, 3.07; 95% CI, 2.83 to 3.33), and septic shock (OR, 5.85; 95% CI, 5.29 to 6.47) also had an increased risk. The model had a training AUC of 0.85. Brier score for the training cohort was 0.006. Supplemental Figure 3B shows the excellent calibration plot of the prediction model.
Validation of the AKI Risk Model
The trained model was tested by applying the model to a validation cohort of 811,778 from 2017. The prediction model has excellent predictive power, with an AUC of 0.90 for the AKI-dialysis model and an AUC of 0.85 for the Postop-MAKE model. Five-fold crossvalidation from the training cohort also showed high average AUC (0.89) for AKI-dialysis model and AUC (0.85) for the Postop-MAKE model. Figure 4 shows the AUC of model validation for AKI-dialysis.
Figure 4.
Showing excellent predictive model performance with validation cohort for postoperative AKI requiring dialysis. AKI-dialysis, AKI requiring RRT; AUC, area under the receiver operating characteristic curve; ROC, receiver operating characteristic.
Examples of AKI Risk Calculator
A calculator for AKI-dialysis risk was programmed from coefficients and intercepts from multivariable logistic regression and the Scikit-learn machine learning library (23). The examples in Supplemental Figure 4 demonstrate the bedside applicability of this model as well as the effect of various clinical factors on the risk of the outcome. The mobile web model is available at https://www.predictionmodel.org/. The mobile web model and hypothetical cases are shown in Supplemental Figure 4.
Discussion
Preoperative evaluation has often focused on cardiac risk assessment, but AKI risk assessment has not been routinely incorporated in clinical practice, even though patients with CKD have high morbidity and mortality after surgery (24–26).
Accurate risk assessment of postoperative AKI will help clinicians to identify patients at high risk and implement individualized preventative measures. For high-risk patients, KDIGO bundle and biomarker-guided intervention may lower AKI risk after major surgery (5,27–30).
In addition, accurate risk assessment can allow appropriate anticipatory guidance to be given to patients in advance of emergent surgeries, allowing patients at high risk of postoperative AKI-dialysis to prepare themselves for that possibility (31).
In this study, a risk model was developed for postoperative AKI-dialysis in patients undergoing broad types of surgery. This model uses simple, easily available clinical data to deliver excellent predictive ability for the outcome, with a validation AUC of 0.90.
This study has strengths compared with prior models to predict postoperative AKI. A prior machine learning algorithm to predict postoperative AKI requires a complex set of electronic health record (EHR)–extracted data (32). There are also previously described machine learning prediction models for AKI in hospitalized patients, which are not specific to the postoperative population and often rely on complex data extracted from the EHR (33–36).
Other strengths of this study include the very large (2,299,502 patients) and recent (2015–2017) cohort used to build our model. The study population includes various surgery types, including cardiac surgery, and the risk model was validated across surgery types.
A limitation of our model is an inability to account for multiple factors as is possible in an EHR-integrated model: for example, zip code as a predictor of socioeconomic factors. The simplicity of our model was an intentional choice, making it useful in a variety of settings without requiring institutional investment in EHR integration. Additionally, the large dataset used to develop and validate the model portends a high degree of external validity. This could be further tested by using this model at institutions that do not participate in NSQIP and evaluating its performance in that population, which will assure the generalizability of the model.
Another limitation pertains to the clinical use of this model. Although certain factors investigated are modifiable (for example, preoperative hematocrit and sodium), this analysis cannot determine causality or the effect of modifying these factors on ultimate risk of AKI-dialysis. An individual with sodium <130 mEq/L can be counseled on an elevated postoperative risk of AKI-dialysis, but it is not known if correcting the sodium prior to surgery will be protective.
This analysis shows that preoperative renal function, older age, congestive heart failure, diabetes, and anemia are all significantly associated with the development of postoperative AKI-dialysis. A clinical model incorporating 11 readily available patient factors can be easily used by the clinician at bedside and has high predictive ability for this outcome, making it a useful and individualized tool for shared surgical decision making. Further study will be needed to see if optimizing modifiable predictors can improve renal outcomes.
Disclosures
All authors have nothing to disclose.
Funding
None.
Acknowledgments
This manuscript was previously posted on the medRxiv preprint server at https://doi.org/10.1101/2020.07.03.20145094.
The ACS NSQIP and the hospitals participating in the ACS NSQIP are the source of the data used herein; they have not verified and are not responsible for the statistical validity of the data analysis or the conclusions derived by the authors.
Author Contributions
S.H. Woo conceptualized the study; S.H. Woo was responsible for data curation, formal analysis, project administration, software, validation, and visualization; L. Ackermann, S.W. Cowan, O.H. Maarouf, S.H. Woo, J. Zavodnick, and J. Zhang were responsible for investigation; S.H. Woo and J. Zavodnick were responsible for methodology; S.H. Woo provided supervision; L. Ackermann, S.H. Woo, and J. Zavodnick wrote the original draft; and L. Ackermann, S.W. Cowan, O.H. Maarouf, S.H. Woo, J. Zavodnick, and J. Zhang reviewed and edited the manuscript.
Supplemental Material
This article contains supplemental material online at http://kidney360.asnjournals.org/lookup/suppl/doi:10.34067/KID.0004732020/-/DCSupplemental.
Study participant flow chart. Download Supplemental Figure 1, PDF file, 564 KB (559.7KB, pdf)
30-day postoperative unadjusted mortality rates of patients who developed AKI requiring dialysis according to surgery types. Download Supplemental Figure 2, PDF file, 564 KB (559.7KB, pdf)
(A) Calibration plot for postoperative acute kidney injury requiring dialysis. Observed vs predicted AKI-Dialysis risk. (B) Calibration plot for Postoperative major adverse kidney event (Postop-MAKE). Observed vs predicted AKI-MAKE risk. Download Supplemental Figure 3, PDF file, 564 KB (559.7KB, pdf)
Mobile web application of AKI risk model. Download Supplemental Figure 4, PDF file, 564 KB (559.7KB, pdf)
Training cohort patient characteristics (2015–2016). Download Supplemental Table 1, PDF file, 564 KB (559.7KB, pdf)
Validation cohort patient characteristics (2017). Download Supplemental Table 2, PDF file, 564 KB (559.7KB, pdf)
Predictors of postoperative major adverse kidney event (Postop-MAKE). Download Supplemental Table 3, PDF file, 564 KB (559.7KB, pdf)
References
- 1.Bihorac A, Yavas S, Subbiah S, Hobson CE, Schold JD, Gabrielli A, Layon AJ, Segal MS: Long-term risk of mortality and acute kidney injury during hospitalization after major surgery. Ann Surg 249: 851–858, 2009. 10.1097/SLA.0b013e3181a40a0b [DOI] [PubMed] [Google Scholar]
- 2.Chertow GM, Lazarus JM, Christiansen CL, Cook EF, Hammermeister KE, Grover F, Daley J: Preoperative renal risk stratification. Circulation 95: 878–884, 1997. 10.1161/01.CIR.95.4.878 [DOI] [PubMed] [Google Scholar]
- 3.Siddiqui NF, Coca SG, Devereaux PJ, Jain AK, Li L, Luo J, Parikh CR, Paterson M, Philbrook HT, Wald R, Walsh M, Whitlock R, Garg AX: Secular trends in acute dialysis after elective major surgery--1995 to 2009. CMAJ 184: 1237–1245, 2012. 10.1503/cmaj.110895 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 4.Hobson C, Ozrazgat-Baslanti T, Kuxhausen A, Thottakkara P, Efron PA, Moore FA, Moldawer LL, Segal MS, Bihorac A: Cost and mortality associated with postoperative acute kidney injury. Ann Surg 261: 1207–1214, 2015. 10.1097/SLA.0000000000000732 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 5.Meersch M, Schmidt C, Hoffmeier A, Van Aken H, Wempe C, Gerss J, Zarbock A: Prevention of cardiac surgery-associated AKI by implementing the KDIGO guidelines in high risk patients identified by biomarkers: The PrevAKI randomized controlled trial [published correction appears in Intensive Care Med 43: 1749, 2017 10.1007/s00134-017-4735-y]. Intensive Care Med 43: 1551–1561, 2017. [DOI] [PMC free article] [PubMed] [Google Scholar]
- 6.Ajitsaria P, Eissa SZ, Kerridge RK: Risk assessment. Curr Anesthesiol Rep 8: 1–8, 2018. 10.1007/s40140-018-0246-9 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 7.Fleisher LA, Fleischmann KE, Auerbach AD, Barnason SA, Beckman JA, Bozkurt B, Davila-Roman VG, Gerhard-Herman MD, Holly TA, Kane GC, Marine JE, Nelson MT, Spencer CC, Thompson A, Ting HH, Uretsky BF, Wijeysundera DN; American College of Cardiology; American Heart Association: 2014 ACC/AHA guideline on perioperative cardiovascular evaluation and management of patients undergoing noncardiac surgery: A report of the American College of Cardiology/American Heart Association Task Force on practice guidelines. J Am Coll Cardiol 64: e77–e137, 2014. 10.1016/j.jacc.2014.07.944 [DOI] [PubMed] [Google Scholar]
- 8.Kristensen SD, Knuuti J, Saraste A, Anker S, Bøtker HE, Hert SD, Ford I, Gonzalez-Juanatey JR, Gorenek B, Heyndrickx GR, Hoeft A, Huber K, Iung B, Kjeldsen KP, Longrois D, Lüscher TF, Pierard L, Pocock S, Price S, Roffi M, Sirnes PA, Sousa-Uva M, Voudris V, Funck-Brentano C; Authors/Task Force Members: 2014 ESC/ESA Guidelines on non-cardiac surgery: Cardiovascular assessment and management: The Joint Task Force on non-cardiac surgery: Cardiovascular assessment and management of the European Society of Cardiology (ESC) and the European Society of Anaesthesiology (ESA). Eur Heart J 35: 2383–2431, 2014. 10.1093/eurheartj/ehu282 [DOI] [PubMed] [Google Scholar]
- 9.Gupta PK, Gupta H, Sundaram A, Kaushik M, Fang X, Miller WJ, Esterbrooks DJ, Hunter CB, Pipinos II, Johanning JM, Lynch TG, Forse RA, Mohiuddin SM, Mooss AN: Development and validation of a risk calculator for prediction of cardiac risk after surgery. Circulation 124: 381–387, 2011. 10.1161/CIRCULATIONAHA.110.015701 [DOI] [PubMed] [Google Scholar]
- 10.Smetana GW, Lawrence VA, Cornell JE; American College of Physicians: Preoperative pulmonary risk stratification for noncardiothoracic surgery: Systematic review for the American College of Physicians. Ann Intern Med 144: 581–595, 2006. 10.7326/0003-4819-144-8-200604180-00009 [DOI] [PubMed] [Google Scholar]
- 11.Pannu N, Graham M, Klarenbach S, Meyer S, Kieser T, Hemmelgarn B, Ye F, James M; APPROACH Investigators and the Alberta Kidney Disease Network: A new model to predict acute kidney injury requiring renal replacement therapy after cardiac surgery. CMAJ 188: 1076–1083, 2016. 10.1503/cmaj.151447 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 12.Jiang W, Teng J, Xu J, Shen B, Wang Y, Fang Y, Zou Z, Jin J, Zhuang Y, Liu L, Luo Z, Wang C, Ding X: Dynamic predictive scores for cardiac surgery-associated acute kidney injury. J Am Heart Assoc 5: e003754, 2016. 10.1161/JAHA.116.003754 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 13.Berg KS, Stenseth R, Wahba A, Pleym H, Videm V: How can we best predict acute kidney injury following cardiac surgery?: A prospective observational study. Eur J Anaesthesiol 30: 704–712, 2013. 10.1097/EJA.0b013e328365ae64 [DOI] [PubMed] [Google Scholar]
- 14.Thakar CV, Arrigain S, Worley S, Yared J-P, Paganini EP: A clinical score to predict acute renal failure after cardiac surgery. J Am Soc Nephrol 16: 162–168, 2005. 10.1681/ASN.2004040331 [DOI] [PubMed] [Google Scholar]
- 15.Mehta RH, Grab JD, O’Brien SM, Bridges CR, Gammie JS, Haan CK, Ferguson TB, Peterson ED; Society of Thoracic Surgeons National Cardiac Surgery Database Investigators: Bedside tool for predicting the risk of postoperative dialysis in patients undergoing cardiac surgery. Circulation 114: 2208–2216, 2006. 10.1161/CIRCULATIONAHA.106.635573 [DOI] [PubMed] [Google Scholar]
- 16.Bilimoria KY, Liu Y, Paruch JL, Zhou L, Kmiecik TE, Ko CY, Cohen ME: Development and evaluation of the universal ACS NSQIP surgical risk calculator: A decision aid and informed consent tool for patients and surgeons. J Am Coll Surg 217: 833–842.e1–e3, 2013. 10.1016/j.jamcollsurg.2013.07.385 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 17.Cohen ME, Ko CY, Bilimoria KY, Zhou L, Huffman K, Wang X, Liu Y, Kraemer K, Meng X, Merkow R, Chow W, Matel B, Richards K, Hart AJ, Dimick JB, Hall BL: Optimizing ACS NSQIP modeling for evaluation of surgical quality and risk: Patient risk adjustment, procedure mix adjustment, shrinkage adjustment, and surgical focus. J Am Coll Surg 217: 336–46.e1, 2013. 10.1016/j.jamcollsurg.2013.02.027 [DOI] [PubMed] [Google Scholar]
- 18.Hall BL, Hamilton BH, Richards K, Bilimoria KY, Cohen ME, Ko CY: Does surgical quality improve in the American College of Surgeons National Surgical Quality Improvement Program: An evaluation of all participating hospitals. Ann Surg 250: 363–376, 2009. 10.1097/SLA.0b013e3181b4148f [DOI] [PubMed] [Google Scholar]
- 19.American College of Surgeons: ACS National Surgical Quality Improvement Program. Available at: https://www.facs.org/quality-programs/acs-nsqip. Accessed November 30, 2020
- 20.American College of Surgeons National Surgical Quality Improvement Program : User Guide for the 2014 ACS NSQIP Participant Use Data File, 2015. Available at: https://www.facs.org/-/media/files/quality-programs/nsqip/nsqip_puf_userguide_2014.ashx. Accessed November 30, 2020
- 21.Thongprayoon C, Cheungpasitporn W, Shah IK, Kashyap R, Park SJ, Kashani K, Dillon JJ: Long-term outcomes and prognostic factors for patients requiring renal replacement therapy after cardiac surgery. Mayo Clin Proc 90: 857–864, 2015. 10.1016/j.mayocp.2015.03.026 [DOI] [PubMed] [Google Scholar]
- 22.Suen WS, Mok CK, Chiu SW, Cheung KL, Lee WT, Cheung D, Das SR, He GW: Risk factors for development of acute renal failure (ARF) requiring dialysis in patients undergoing cardiac surgery. Angiology 49: 789–800, 1998. 10.1177/000331979804900902 [DOI] [PubMed] [Google Scholar]
- 23.Pedregosa F, Varoquaux G, Gramfort A, Michel V, Thirion B, Grisel O, Blondel M, Prettenhofer P, Weiss R, Dubourg V, Vanderplas J, Passos A, Cournapeau D, Brucher M, Perrot M, Duchesnay É: Scikit-learn: Machine learning in Python. J Mach Learn Res 12: 2825–2830, 2011 [Google Scholar]
- 24.Meersch M, Schmidt C, Zarbock A: Patient with chronic renal failure undergoing surgery. Curr Opin Anaesthesiol 29: 413–420, 2016. 10.1097/ACO.0000000000000329 [DOI] [PubMed] [Google Scholar]
- 25.Ackland GL, Moran N, Cone S, Grocott MPW, Mythen MG: Chronic kidney disease and postoperative morbidity after elective orthopedic surgery. Anesth Analg 112: 1375–1381, 2011. 10.1213/ANE.0b013e3181ee8456 [DOI] [PubMed] [Google Scholar]
- 26.Mathew A, Devereaux PJ, O’Hare A, Tonelli M, Thiessen-Philbrook H, Nevis IFP, Iansavichus AV, Garg AX: Chronic kidney disease and postoperative mortality: A systematic review and meta-analysis. Kidney Int 73: 1069–1081, 2008. 10.1038/ki.2008.29 [DOI] [PubMed] [Google Scholar]
- 27.Park JT: Postoperative acute kidney injury. Korean J Anesthesiol 70: 258–266, 2017. 10.4097/kjae.2017.70.3.258 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 28.Mårtensson J, Bellomo R: Prevention of renal dysfunction in postoperative elderly patients. Curr Opin Crit Care 20: 451–459, 2014. 10.1097/MCC.0000000000000107 [DOI] [PubMed] [Google Scholar]
- 29.Göcze I, Jauch D, Götz M, Kennedy P, Jung B, Zeman F, Gnewuch C, Graf BM, Gnann W, Banas B, Bein T, Schlitt HJ, Bergler T: Biomarker-guided intervention to prevent acute kidney injury after major surgery: The prospective randomized BigpAK study. Ann Surg 267: 1013–1020, 2018. 10.1097/SLA.0000000000002485 [DOI] [PubMed] [Google Scholar]
- 30.Engelman DT, Crisafi C, Germain M, Greco B, Nathanson BH, Engelman RM, Schwann TA: Using urinary biomarkers to reduce acute kidney injury following cardiac surgery. J Thorac Cardiovasc Surg 160: 1235–1246.e2, 2020. 10.1016/j.jtcvs.2019.10.034 [DOI] [PubMed] [Google Scholar]
- 31.Politi MC, Street RL Jr: The importance of communication in collaborative decision making: Facilitating shared mind and the management of uncertainty. J Eval Clin Pract 17: 579–584, 2011. 10.1111/j.1365-2753.2010.01549.x [DOI] [PubMed] [Google Scholar]
- 32.Thottakkara P, Ozrazgat-Baslanti T, Hupf BB, Rashidi P, Pardalos P, Momcilovic P, Bihorac A: Application of machine learning techniques to high-dimensional clinical data to forecast postoperative complications. PLoS One 11: e0155705, 2016. 10.1371/journal.pone.0155705 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 33.Kate RJ, Perez RM, Mazumdar D, Pasupathy KS, Nilakantan V: Prediction and detection models for acute kidney injury in hospitalized older adults. BMC Med Inform Decis Mak 16: 39, 2016. 10.1186/s12911-016-0277-4 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 34.Davis SE, Lasko TA, Chen G, Siew ED, Matheny ME: Calibration drift in regression and machine learning models for acute kidney injury. J Am Med Inform Assoc 24: 1052–1061, 2017. 10.1093/jamia/ocx030 [DOI] [PMC free article] [PubMed] [Google Scholar]
- 35.Forni LG, Dawes T, Sinclair H, Cheek E, Bewick V, Dennis M, Venn R: Identifying the patient at risk of acute kidney injury: A predictive scoring system for the development of acute kidney injury in acute medical patients. Nephron Clin Pract 123: 143–150, 2013. 10.1159/000351509 [DOI] [PubMed] [Google Scholar]
- 36.Koyner JL, Carey KA, Edelson DP, Churpek MM: The development of a machine learning inpatient acute kidney injury prediction model. Crit Care Med 46: 1070–1077, 2018. 10.1097/CCM.0000000000003123 [DOI] [PubMed] [Google Scholar]
Associated Data
This section collects any data citations, data availability statements, or supplementary materials included in this article.
Supplementary Materials
Study participant flow chart. Download Supplemental Figure 1, PDF file, 564 KB (559.7KB, pdf)
30-day postoperative unadjusted mortality rates of patients who developed AKI requiring dialysis according to surgery types. Download Supplemental Figure 2, PDF file, 564 KB (559.7KB, pdf)
(A) Calibration plot for postoperative acute kidney injury requiring dialysis. Observed vs predicted AKI-Dialysis risk. (B) Calibration plot for Postoperative major adverse kidney event (Postop-MAKE). Observed vs predicted AKI-MAKE risk. Download Supplemental Figure 3, PDF file, 564 KB (559.7KB, pdf)
Mobile web application of AKI risk model. Download Supplemental Figure 4, PDF file, 564 KB (559.7KB, pdf)
Training cohort patient characteristics (2015–2016). Download Supplemental Table 1, PDF file, 564 KB (559.7KB, pdf)
Validation cohort patient characteristics (2017). Download Supplemental Table 2, PDF file, 564 KB (559.7KB, pdf)
Predictors of postoperative major adverse kidney event (Postop-MAKE). Download Supplemental Table 3, PDF file, 564 KB (559.7KB, pdf)