A model for predicting AKI after cardiopulmonary bypass surgery in Chinese patients with normal preoperative renal function

Abstract

Objective

To develop and validate a predictive model for acute kidney injury (AKI) after cardiopulmonary bypass (CPB) surgery in Chinese patients with normal preoperative renal function.

Method

From January 1, 2015, to September 1, 2022, a total of 1003 patients were included in the analysis as a development cohort. We used the ratio of 7:3 to divide the patients into a training group (n = 703) and a testing group (n = 300). In addition, a total of 178 patients were collected as an external validation cohort from January 1, 2023, to May 1, 2023. In the training group, independent risk factors for postoperative AKI were identified through the least absolute shrinkage and selection operator (LASSO) regression and multifactor logistic regression analysis. A nomogram predictive model was then established. The area under the curve (AUC) of receiver operating characteristic (ROC) curve, as well as calibration curve and decision curve, were used for validation of the model.

Results

Age, body mass index (BMI), emergent surgery, CPB time, intraoperative use of adrenaline, and postoperative procalcitonin (PCT) were identified as important risk factors for AKI after CPB surgery (P < 0.05). The nomogram predictive model demonstrated good discrimination (AUC: 0.772 (95%CI: 0.735 − 0.809), 0.780 (95% CI: 0.724 − 0.835), and 0.798 (95% CI: 0.731 − 0.865)), calibration (Hosmer and Lemeshow goodness of fit test: P-value 0.6941, 0.9539, and 0.2358), and clinical utility (the threshold probability values in the decision curves are respectively > 12%, > 10%, and 16% ~ 75%) in the training, testing, and external validation groups.

Conclusion

The predictive model, which was established in Chinese patients with normal preoperative renal function, has high accuracy, calibration, and clinical utility. Clinicians can utilize this model to predict and potentially reduce the incidence of AKI after CPB surgery in the Chinese population.

Introduction

Cardiopulmonary bypass (CPB) is a method of extracorporeal circulation that can temporarily substitute the heart and lungs’ functions to maintain the circulation of blood and oxygen in the patient, and it is used in more than 80% of cardiac surgeries (according to the American Heart Association) [1, 2]. Since it was produced almost 60 years ago, CPB has benefited lots of people. However, CPB is associated with several issues that persist, such as hemolysis, capillary leak syndrome, and acute kidney injury (AKI) [3]. AKI is a common and important complication, with reported occurrence rates ranging from 5 to 42% in patients undergoing CPB surgery. It is linked to increased rates of morbidity and mortality, prolonged stays in the intensive care unit (ICU), and longer hospital stays [4,5,6]. Furthermore, research has shown that AKI is related with unfavourable long-term outcomes, even after apparent recovery, including a higher risk of mortality, cardiovascular events, and the development of chronic kidney disease (CKD) [7]. Currently, the clinical diagnosis of AKI primarily depends on the assessment of the serum creatinine (Scr) and blood urea nitrogen (BUN) levels. However, these conventional biomarkers are not effective in early AKI prediction. Thus, early identification of high-risk AKI patients and the timely implementation of renal protection strategies are crucial.

So far, numerous studies have aimed to prevent AKI and identified new biomarkers for its prediction. Various laboratory parameters, such as neutrophil gelatinase-associated lipocalin (NGAL), kidney injury molecule-1 (KIM-1), N-acetyl-β-glucosaminidase (NAG), interleukin-18 (IL-18), and cystatin C, have been proposed as potential AKI biomarkers in recent years [8]. Additionally, research by Stefan J Schunk et. found that preoperative urinary dickkopf-3 is also an independent predictor of postoperative AKI [9]. These biomarkers offer several advantages over Scr, such as their ability to rapidly increase in plasma and urine within hours of renal injury, making them more specific and sensitive. However, due to limited validation, availability, and reimbursement, these biomarkers are currently used only for research purposes and not in clinical practice.

Furthermore, due to the multifactorial nature of AKI development, postoperative AKI cannot be predicted solely based on a single risk factor. Thus, it is essential to establish a efficient bedside tool for patients undergoing CPB surgery. Clinical prediction models estimate the probability of risk by statistically combining a set of patient and disease characteristics. These models serve as valuable clinical support tools, assisting clinicians in making informed decisions regarding medical treatments. Conventional prediction models such as the Cleveland Clinic score [10], Mehta score [11], and Simplified Renal Index (SRI) score [12] are commonly used in Europe and the United States to predict cardiac surgery-associated AKI. However, their predictive capabilities have not been adequately validated in China due to racial differences. Moreover, these models were developed to AKI in patients requiring renal replacement therapy. Compared to mild AKI, which does not necessitate dialysis, the incidence of renal replacement therapy is low and occurs late in clinical practice, limiting the practical application of these models. Therefore, the objectives of our study was to develop and validate a predictive model for AKI after CPB surgery in Chinese patients.

Patients and methods

In our study, we adhered to the guidelines outlined in the TRIPOD [13] (Transparent Reporting of a Multivariable Prediction Model for Individual Prognosis or Diagnosis) statement for reporting the development and validation of multivariable prediction model.

Patient selection

As the development cohort, we collected data from a consecutive series of 1003 patients who underwent CPB surgery at the First Affiliated Hospital of Sun Yat-sen University from January 1, 2015, to September 1, 2022.

To validate the prediction model externally, we retrospectively extracted relevant data from 178 patients who underwent CPB surgery at the same hospital between January 1, 2023, and May 1, 2023.

Inclusion criteria: Patients confirmed by echocardiography, coronary angiography, and total abdominal CT, with clear indications for cardiopulmonary surgery. Exclusion criteria included: (1) preoperative kidney diseases or abnormal creatinine elevation; (2) patients under the age of 18; (3) lack of clinical data. According to the C-statistic of 0.8, 20 possible risk factors were expected to be counted, and based on a previous AKI incidence of 0.34 from the literature [14], the estimated sample size was 647. Our study complies with the relevant provisions of ethics and adheres to the Declaration of Helsinki. Approval for the study protocol was obtained from the Ethical Committee for Clinical Research and Animal Trials at the First Affiliated Hospital of Sun Yat-sen University (Approval Number: 2023–445). The data had been completely anonymized to remove any identifying information, thus exempting the need for informed consent.

Variable collection

Many published articles have evaluated the clinical risk factors associated with AKI after CPB surgery. These factors can be generally divided into preoperative, intraoperative, and postoperative categories based on the timing of the surgery. Common preoperative factors include advanced age, female sex, smoking history, and certain preoperative complications such as diabetes, heart failure, and chronic obstructive pulmonary disease (COPD). Intraoperative factors comprise the type of surgery, CPB time, aortic occlusion time, hemodilution, and exposure to positive inotropic drugs. Postoperative factors encompass hemodynamic changes such as anemia, inadequate blood volume, and cardiogenic shock [3, 6]. Moreover, AKI following cardiac surgery is essentially another specific manifestation of the “cardiorenal syndrome”. It is therefore important to closely monitor functional indicators of the heart prior to CPB surgery. In our study, relevant cardiac functional indicators such as left ventricular ejection fraction (LVEF), New York Heart Association (NYHA) classification, and N-terminal pro-B-type natriuretic peptide (NT-proBNP) levels were also considered. In addition, some studies have also indicated that the underlying physical condition of patients may be associated with the incidence of AKI after surgery. For instance, anaemic patients are more likely to develop AKI than patients with normal hemoglobin levels [15], and malnourished patients also appear to have a higher incidence of AKI [16]. Recent studies have shown that elevated postoperative procalcitonin (PCT) levels are more common in patients at a high risk of various complications after cardiac surgery [17]. Therefore, measuring PCT levels may help in identifying patients at a high risk of developing AKI after CPB. Considering the aforementioned background information and the observations in the clinical work, the clinical indicators included in this study were carefully chosen.

We selected the predictor variables that are assessable as follows: (I) basic demographic data such as age, gender, body mass index (BMI), American Society of Anesthesiologists (ASA) class, and primary comorbidities; (II) preoperative laboratory tests including white blood cell (WBC), red blood cell (RBC), hemoglobin (Hb), hematokrit (HCT), platelet (PLT), albumin (ALB), BUN, Scr, uric acid (UA), NT-proBNP, etc.; (III) intraoperative data encompassing surgical types, surgical time, CPB time, aortic occlusion time, minimum temperature, intraoperative medication, volume of various fluids administered, total blood loss, and total urine output; (IV) postoperative data covering postoperative PCT on day 1 (routinely tested on the first day after cardiac surgery at the First Affiliated Hospital of Sun Yat-sen University).

Outcome and definition

The primary outcome of the study was the development of AKI within 7 days after CPB surgery. According to the Kidney Disease: Improving Global Outcomes (KDIGO) clinical practice guidelines, AKI is defined as follow: an increase in Scr of 0.3 mg/dl (26.5µmol/L) within 48 h, or Scr increased to 1.5 times the baseline level within 7 days, or urine volume < 0.5 ml/kg/h for 6 h [18]. Due to the limitations of retrospective studies that prevent accurate collection of urine volume for each patient, blood creatinine levels were used to establish the presence of AKI. The baseline serum creatinine concentration was determined as the most recent value obtained within 7 days prior to the surgery.

Model development and validation

Randomization was generated using R software to divide the 1003 patients into training and testing cohorts with a 7:3 allocation ratio based on a random sequence. The accuracy of our model was also externally validated using 178 additional patients.

The least absolute shrinkage and selection operator (LASSO) method was utilized to select the most predictive characteristics of risk factors for postoperative AKI in patients who underwent CPB surgery. Identify the variables in the LASSO regression model that have non-zero coefficients. Multicollinearity of the variables was tested and excluded by calculating the variance inflation factor (VIF). Subsequently, a prediction model was created by using variables with P < 0.05 after multivariate logistic regression analysis. A nomogram was established by the final model through the “rms” package of R software. The area under the curve (AUC) of receiver operating characteristic (ROC) curve, as well as calibration curve and decision curve were used to validate the nomogram prediction model to assess its discriminative ability, calibration, and clinical usefulness.

Statistical analysis

In this study, all statistical analyses were performed using R software, version 4.2.2 (R Foundation for Statistical Computing, Vienna, Austria), along with the use of “rms” package and MSTATA software.

The data were complete for demographics, comorbidities, surgical time, intraoperative medication, volume of various fluids administered, and postoperative PCT on day 1. Variables with a deletion rate exceeding 5% were excluded, and missing data were imputed using the technique of multiple imputation by chained equations with 20 iterations. The results were pooled according to Rubin’s rule [19].

Continuous variables were presented as the mean ± standard deviation (SD) and compared using Student’s independent t-test. Non-normally distributed data were reported as the median [interquartile range (IQR, P25, P75)] and tested by nonparametric test (Mann‒Whitney U test). Categorical variables were shown as counts (%) and compared by χ2 test or Fisher’s exact test. Statistical significance was considered present when P < 0.05.

Results

Patient demographic and clinical characteristics

From January 1, 2015, to September 1, 2022, a total of 1003 patients were included in the analysis, divided into an AKI group (n = 328) and a Non-AKI group (n = 675) according to KDIGO standards. They were randomly divided into a training group (n = 703) and a testing group (n = 300) to assess the model’s accuracy. Moreover, a total of 178 patients were included from January 1, 2023, to May 1, 2023, to serve as an external validation group for verifying the model’s performance. This group comprised 59 patients with AKI and 119 patients without AKI. The detailed flowchart is shown in Fig. 1.

Fig. 1

Flow diagram of study design

Table 1 shows the overall characteristics of the AKI group and the Non-AKI group of patients. The gender ratio was 1.37:1 (male/female = 580/423), with an average age of 52.57 ± 13.35 years. Patients in the AKI group exhibited significantly higher age and BMI, a higher prevalence of previous cardiac surgery, hypertension, diabetes, and heart failure, as well as elevated baseline levels of WBC, BUN, SCr, and NT-proBNP. They also had lower baseline levels of LVEF, RBC, Hb, HCT value, PLT, and ALB compared to the Non-AKI group (P < 0.05).

Table 1 Clinical characteristics of the study population in the Non-AKI and AKI groups

In the development cohort, the incidence of postoperative AKI was 32.7%. Out of the total, 611 patients (60.9%) underwent valve surgery, 127 patients (12.7%) underwent coronary artery bypass grafting (CABG), 39 patients (3.9%) underwent valve surgery combined with CABG, 74 patients (7.4%) underwent vascular surgery, 87 patients (8.7%) had congenital heart surgery, 2 patients (0.2%) had heart transplant surgery, and 63 patients (6.3%) underwent other cardiac surgeries. In the external validation cohort, the incidence rate of AKI was 33.1%. Additional details can be found in Table S1.

Table 2 displays the general characteristics of the training group and the testing group of patients. The incidence of postoperative AKI was 32.6% (n = 229) and 33% (n = 99) in two groups respectively. Except for the P-values of UA, congenital heart surgery, and postoperative PCT on day 1 which were less than 0.05, there was no significant difference in other any characteristics between the two groups. With so many similar variables present, the imbalance of a few variables did not affect the model construction.

Table 2 Clinical characteristics of the study population in the training and testing groups

Comparison of surgical outcomes between Non-AKI and AKI groups

Table S2 indicates that patients in the AKI group had longer hospital stays (30.72 ± 13.84 days vs. 25.82 ± 9.41 days, P < 0.001), longer postoperative hospital stays (19.86 ± 11.74 days vs. 16.54 ± 7.07 days, P < 0.001), and longer ICU stays (4.32 ± 6.31 days vs. 1.81 ± 1.69 days, P < 0.001). Additionally, they had a higher likelihood of requiring CRRT (P < 0.001) and experiencing in-hospital mortality (P < 0.001).

Selection of variables for AKI after CPB surgery

Using the LASSO algorithm, 11 potential risk factors were selected with the chosen Lambda.1se being 0.040 (Fig. 2A and B). As shown in Table S3, the maximum VIF of the factors was 2.70, indicating no multicollinearity among the variables. After multivariate logistic regression analysis, 6 factors remained independently associated with the risk of AKI (Table 3). Age, BMI, emergent surgery, CPB time, intraoperative use of adrenaline, and postoperative PCT were identified to be independent risk factors for AKI after CPB surgery.

Fig. 2

Selection of predictive features for patients who underwent CPB using the LASSO logistic regression model

Table 3 The multivariate logistic regression model of predicting AKI after CPB in the training cohort

Construction of a nomogram model for predicting AKI after CPB surgery

Taking the development of AKI as the dependent variable and using the variables determined in the multivariate logistic regression analysis as predictive variables to construct a nomogram. The scores of each factor (age, BMI, emergent surgery, CPB time, intraoperative use of adrenaline, and postoperative PCT) in the nomogram were calculated. The total score represents the risk of AKI and can be easily calculated by adding individual scores, with a scoring range from 0 to 160 (Fig. 3).

Fig. 3

Nomogram prediction model for AKI after CPB surgery

Validation of the AKI nomogram model

The AUC of the nomogram model for the training group and testing group is 0.772 (95% CI: 0.735 − 0.809) and 0.780 (95% CI: 0.724 − 0.835) respectively (Fig. 4A and B). Correspondingly, the model achieved an AUC of 0.798 (95% CI: 0.731 − 0.865) in the external validation group (Fig. 7A), indicating that the model is highly accurate in estimating the probability of AKI following CPB surgery. The model’s calibration curves of the training, testing, and external validation cohorts, are depicted in Figs. 5A, B, and 7B, demonstrating good consistency between prediction and observation (Hosmer and Lemeshow goodness of fit test: P-value 0.6941, 0.9539, and 0.2358).

Fig. 4

ROC curves of the predictive nomogram for AKI risk

Fig. 5

Calibration curves of the predictive nomogram for AKI risk

Clinical application of the AKI nomogram model

When the threshold probability values in the decision curves of the training, testing, and external validation groups are respectively > 12%, > 10%, and16% ~ 75%, the model used to predict the risk of AKI occurrence has a higher net benefit, suggesting that the predictive model has good clinical usability (Figs. 6A, B, and 7C).

Fig. 6

Decision curve analysis for the AKI risk nomogram

Fig. 7

ROC curve, calibration curve, and decision curve for the model in the external validation cohort

Discussion

Most cardiac surgeries require the establishment of CPB, with the aim of replacing the functions of the heart and lungs to ensure adequate blood perfusion throughout the body and provide surgeons with a clear view of the surgical field. However, CPB does not replicate physiological circulation. For instance, CPB creates non-pulsatile perfusion, leading to a decrease in glomerular perfusion pressure, as well as inducing hemolysis and hemodilution, resulting in reduced oxygen delivery. Additionally, CPB triggers inflammation in vivo, leading to a systemic inflammatory response, exacerbating renal ischemia and hypoxia during surgery, increasing the risk of renal injury, and negatively impacting patient prognosis [20]. Our study found that approximately 32.7% of patients undergoing CPB surgery experienced AKI, aligning closely with research findings globally(5% ~ 42%) [6]. Furthermore, in terms of prognosis, we observed that patients in the AKI group had longer overall hospital stays, longer postoperative hospital stays, and extended stays in the ICU, as well as a higher likelihood of requiring CRRT and experiencing in-hospital mortality, which is consistent with the findings of previous studies [4,5,6].

Numerous articles have assessed clinical and surgical risk factors associated with AKI following CPB surgery. Prediction models such as the Cleveland Clinic score, Mehta score, and SRI score have demonstrated good discriminatory value. However, applying these models to the Chinese patient population poses challenges due to differences in demographics, clinical characteristics, and healthcare systems. Additionally, studies have shown that these three traditional models perform well in predicting RRT-AKI, but demonstrate poorer performance in predicting non-RRT-AKI [21]. Therefore, we developed a model to predict postoperative AKI in Chinese patients undergoing CPB surgery. Notably, we integrated clinical variables with a biomarker (postoperative PCT on day 1) in our model, demonstrating good discrimination, calibration capabilities, and clinical utility.

Age emerged as a significant risk factor for postoperative AKI following CPB surgery, especially in individuals aged over 70 [3, 6, 22, 23], as supported by various studies. Comorbidities that accumulate with age, such as renovascular disease and congestive heart failure, may contribute to the higher incidence of AKI in elderly patients. Additionally, age-related structural and functional changes in the kidneys impair their ability to withstand and recover from injury, rendering the elderly more susceptible to AKI [24].

As previously reported, BMI is a factor that contributes to AKI after cardiac surgery [6, 23]. Obesity can lead to increased oxidative stress, endothelial dysfunction, and inflammation, but it is unclear how it affects postoperative AKI. Billings et al. suggested that the association between obesity and AKI is partly influenced by oxidative stress [25]. Moreover, a retrospective cross-sectional study examined kidney biopsy findings in patients with morbid obesity and found that morbidly obese patients have a high incidence of acute tubular necrosis (ATN), which may demonstrate a vulnerability to insult that reduce renal perfusion [26]. Additionally, BMI and AKI incidence are directly proportional, with a higher BMI resulting in a higher incidence [27]. However, in terms of prognosis, several epidemiological studies indicated that the prognosis for patients with AKI is better among moderately obese individuals than those of normal or below-normal weight, a phenomenon known as the obesity paradox [28].

Researches have demonstrated that emergent surgery can serve as a predictor for AKI [29], possibly because patients requiring emergent surgery are frail [29, 30]. Our findings revealed that patients undergoing emergent cardiac surgeries with CPB are at a higher risk of postoperative AKI.

In our research, we identified two intraoperative variables as risk factors: CPB time and the use of adrenaline during surgery. Most studies have come to the same conclusion that CPB time is a one of the independent risk factors of AKI after CPB surgery [31, 32]. As mentioned above, AKI may be related to renal hypoperfusion, hemodilution, and systemic inflammatory response cause by CPB. Additionally, similar to the research of Peng et al. and Li et al. [33, 34], our findings indicated that the use of adrenaline during surgery is a significant factor in the development of early postoperative kidney impairment. Some researches have demonstrated that adrenaline α-adrenoceptor activity can result in a reduction of renal blood flow. Furthermore, the administration of intraoperative adrenaline is associated with hemodynamic impairment, most commonly due to cardiogenic shock, which is also a significant cause of the decrease in renal perfusion and consequently leads to AKI [35].

To date, serum PCT levels have been found to be higher in patients with AKI than in those without AKI in various clinical settings [36,37,38,39]. There is no clear pathophysiological mechanism by which high-serum PCT concentrations are connected to AKI. A possible underlying mechanism may be the inflammatory response induced by CPB that causes the development of AKI, with proinflammatory cytokines significantly stimulating PCT production [40, 41]. Our study’s results showed that PCT levels are independently linked to AKI in CPB surgery. However, studies examining the connection between PCT and AKI in CPB surgery are currently few.

Collectively, by combining these risk factors, we can create a simple, precise, and reliable bedside risk tool. This tool can facilitate discussions between clinicians and patients regarding the risks of AKI following CPB surgery. High-risk patients should be the focus of renal protection strategies, and clinicians should prioritize the prevention of AKI in these patients following CPB surgery.

There are limitations to our study. Firstly, this model lacked external validation from another institution, which hinders its broad applicability. A multicenter prospective study is needed to confirm our findings in the future. Secondly, data were missing for some variables. We deleted variables with a missing rate of more than 5% and used multiple imputation to minimize the impact of missing data on model performance. Despite its ability to provide reasonable estimates of missing data, it still relies on assumptions that may not be entirely accurate. Thirdly, the urine criterion was not applied to diagnose AKI in our study due to the unavailability of data on urine output per hour, potentially leading to an underestimation of the overall incidence of AKI. Thus, a prospective study is warranted to remedy this deficiency.

Conclusion

This study aimed to develop and validate a risk predictive model for AKI following CPB surgery in Chinese patients with normal preoperative renal function. LASSO regression was employed to select variables. The model exhibited strong predictive capabilities in the training, testing, and external validation groups. Clinicians can utilize this model to predict and potentially reduce the incidence of AKI after CPB surgery in the Chinese population.