Application of machine learning to develop a predictive model for hyperbilirubinemia after cardiac surgery

Application of machine learning to develop a predictive model for hyperbilirubinemia after cardiac surgery | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Application of machine learning to develop a predictive model for hyperbilirubinemia after cardiac surgery Weibin Lin, Haowei Zhu This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-6456253/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Objective This study aimed to develop a machine learning-based model to predict hyperbilirubinemia after cardiac surgery. Methods A retrospective analysis was performed on 554 patients who underwent cardiopulmonary bypass surgery at the First Hospital of Sun Yat-sen University in 2021. The predictors of HB were determined by Least Absolute Shrinkage and Selection Operator (LASSO)regression, and eight different machine learning algorithms were constructed, including naive Bayes (NB), support vector machine (SVM), and decision tree (DT), random forest (RF), light gradient boosting machine (LightGBM), gradient boosting decision tree (GBDT), extra trees, and adaptive boosting (AdaBoost). Shapley additive explanation (SHAP) is used for interpretability analysis of the model. Results A total of 401 patients were enrolled, and 20 features were identified via LASSO regression. Among the 8 algorithms, the Area Under Curve (AUC)value of extra trees was 0.846, which was superior to those of the other models. The top 4 features of SHAP analysis were the preoperative total bilirubin level and international normalized ratio (INR). Other important risk factors included intraoperative red blood cell infusion and dexmedetomidine (DEX) use. Conclusions The extra-trees model is a good model for predicting the occurrence of hyperbilirubinemia after cardiac surgery. The most important risk factors for postoperative HB were an increase in total bilirubin and the INR before the operation, an increase in red blood cell transfusion during the operation and no use of DEX. LASSO Extra Trees SHAP Transfusion Dexmedetomidine Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 INTRODUCTION Since 1967, Prof. Mundth has documented that elevated total bilirubin levels following heart surgery are a common complication 1 . About 10.1–35.1% of individuals experience this syndrome following heart surgery, and it has the potential to significantly raise in-hospital mortality 2 . Traditional techniques (logistic regression, nomograms) have confirmed a number of risk factors for the development of hyperbilirubinemia(HB) following cardiac surgery, including hepatic insufficiency, infective endocarditis, the number of valve replacements, elevated right atrial pressure, blood transfusions, aging, diabetes mellitus, duration of extracorporeal circulation, and duration of aortic occlusion 3 – 5 . Machine learning is used for medical data mining and shows better predictive performance than traditional methods for other diseases 6 – 9 . While machine learning models have been applied to assess the risk of elevated total bilirubin levels in vascular surgery 3 , there is a lack of corresponding research in the overall field of cardiac surgery. Therefore, a set of models capable of assessing the risk of elevated bilirubin levels is essential to help physicians identify high-risk patients and take preventive measures. Such models could help reduce perioperative complications and improve patient prognosis. By analyzing and predicting the risk of HB, we can take better care of our patients and reduce surgical complications. METHODS Study population and demographic characteristics Retrospective analysis was used in this study to gather data from patients who had heart surgery at Sun Yat-sen University's First Affiliated Hospital in 2021. Patients did not need to sign an informed consent form as the study used deidentified data. The following were the requirements for the study population's inclusion: 18 years of age or older and had extracorporeal circulation-assisted heart surgery. Patients with malignancies, those who had heart transplants, those without recorded postoperative bilirubin data, and those with preoperative bilirubin levels greater than 3 mg/dl were all excluded. This study has been approved by Sun Yat-sen University's First Affiliated Hospital Ethics Committee (approval number: [2023] 493). Considering the anonymization of the data, all identifying information was removed; therefore, the requirement for informed consent was waived. The study was conducted in a stepwise manner, strictly adhering to the relevant regulations and the principles of the Declaration of Helsinki. The detailed study flowchart is shown in Fig. 1 . Our hospital employs extracorporeal circulation with the median incision utilized in cardiac surgery. The anesthesiologist administers positive inotropic and vasoconstrictive agents to regulate blood pressure following aortic clamped. The extracorporeal circulation is discontinued after the requisite conditions are fulfilled. When the evacuation of extracorporeal circulation proves difficult, we initiate intra-aortic balloon pump or extracorporeal membrane oxygenation and generally sustain the overall reperfusion duration at approximately one-third of the aortic occlusion interval. The electronic medical records system grants access to all critical patient information, including demographics, comorbidities, and laboratory findings. To evaluate postoperative bilirubin (HB) levels, we utilized the criteria from other research and established the threshold as a total bilirubin above 3 mg/dL within 7 days post-surgery. 3 , 10 . Machine Learning Model Building and Model Evaluation We partitioned the training and test sets in an 8:2 ratios to identify the optimal model. We developed eight distinct machine learning algorithms: naïve Bayes (NB), support vector machine (SVM), decision tree (DT), random forest (RF), light gradient boosting machine (LightGBM), gradient boosting decision tree (GBDT), Extra Trees, and AdaBoost. The efficacy of these models was evaluated by 10-fold cross-validation on the training dataset. A grid search was employed to identify the ideal hyperparameter combinations for each model to enhance its performance. We assessed the models' performance by calculating the F1 score, accuracy, recall, precision, and area under the curve, and we plotted the receiver operating characteristic (ROC) curve for each model. Feature importance and model interpretability analysis Based on the model evaluation results, the most effective machine learning model is chosen. Feature importance analysis was conducted on the selected models to ascertain the significance of each feature. This study involved screening 20 predictors by LASSO regression, selecting the optimal extra tree model for prediction, and performing a detailed interpretive analysis to elucidate the unique influence of each feature on model decisions. A transparent and interpretable prediction framework was developed by quantitatively assessing the contribution of attributes to the prediction outcomes using the Shapley additive explanation (SHAP) approach. 11 . Furthermore, to investigate the global and local interpretations of the model, comprehensive studies were performed using the SHAP summary plot and SHAP force plot, respectively. The summary plot depicted the distribution of SHAP values for each feature, illustrating the range of values and indicating the positive and negative directions as well as the strength of the feature's prediction of the target variable. Additionally, a force plot is employed to locally elucidate the prediction process for a particular sample, illustrating how the model integrates the positive and negative contributions of feature values in that sample and quantifies the cumulative effect of the features on the final prediction via an intuitive arrow plot. Utilizing these methodologies, we developed a predictive model to assess the likelihood of hyperbilirubinemia following heart surgery. Statistical analyses Within our dataset, merely 13 data points were absent (< 5%) (left ventricular systole 9/401; right ventricular diastole 4/401), and these missing values were supplemented using repeated interpolation prior to subsequent statistical analysis. All statistical analyses were conducted using R software version 4.2.2. Continuous variables are presented as means ± standard deviations, whereas non-normally distributed data are presented as medians [interquartile ranges]. Categorical variables are expressed as numerical values and percentages [n (%)]. RESULTS Participant characteristics We encompassed a total of 401 participants. Among them, 115 individuals, accounting for 28.7% of the entire group, were found to be developed HB. Table 1 shows the characteristics of the participants in this study. Table 1 Clinical features of HB patients and non-HB patients Variables non-HB(n = 286) HB(n = 115) Age (years) 53.3 ± 14.1 54.1 ± 10.7 Male, n (%) 179 (62.6%) 73 (63.5%) BMI (kg/m2) 23.2 ± 3.6 22.9 ± 3.5 Smoker 71 (24.8%) 26 (22.6%) Diabetes 38 (13.3%) 11 (9.6%) Atrial fibrillation 40 (14.0%) 28 (24.3%) COPD 7 (2.4%) 1 (0.9%) AHF 8 (2.8%) 2 (1.7%) PCI 34 (11.9%) 10 (8.7%) PH 96 (33.6%) 43 (37.4%) PVD 29 (10.1%) 2 (1.7%) RFD 30 (10.5%) 14 (12.2%) Emergency 4 (1.4%) 6 (5.2%) Hb(g/L) 132.3 ± 21.1 131.5 ± 21.4 PLT(10 9 /L) 220.7 ± 67.0 214.9 ± 72.7 ALB(g/L) 38.9 ± 4.0 38.6 ± 4.0 AST(U/L) 23.0 (10.0-221.0) 28.0 (10.0-313.0) TBIL.pre(umol/L) 13.8 ± 5.9 21.6 ± 9.7 Cr(umol/L) 80.0 (36.0-845.0) 80.0 (50.0-581.0) pre.UA(umol/L) 412.8 ± 134.7 437.7 ± 144.1 INR 1.1 ± 0.3 1.2 ± 0.3 BNP(pg/mL) 277.4 (10.0-89957.0) 720.7 (10.0-32557.0) EF (%) 63.9 ± 10.6 63.1 ± 11.2 Preoperative.intubation 1 (0.3%) 0 (0.0%) Preoperative.vasoactive 34 (11.9%) 17 (14.8%) ACCT 87.3 ± 43.3 114.4 ± 58.5 CPBT 158.6 ± 77.4 202.3 ± 98.7 Minimum.temperature 34.4 ± 2.0 34.0 ± 1.9 Blood loss 522.7 ± 277.2 687.0 ± 418.5 RBC transfusion 230.0 (0.0-1500.0) 460.0 (0.0-2600.0) Albumin 10 (3.5%) 10 (8.7%) DEX 193 (67.5%) 68 (59.1%) Dopamine 36 (12.6%) 17 (14.8%) Dobutamine 163 (57.0%) 71 (61.7%) Norepinephrine 237 (82.9%) 104 (90.4%) Adrenaline 44 (15.4%) 21 (18.3%) 1 mg/dL = 17.1 µmol/L ACCT, aortic cross-clamp time; CPBT, cardiopulmonary bypass time Feature selection LASSO regression was used for feature selection, and the penalty was increased by introducing an L1 regularization term. Two key λ values were determined through cross-validation: λ.min = 0.0197 and λ.1 se = 0.0547 (Fig. 2 ). Owing to the subsequent need for machine learning for prediction, the λ.min model was chosen to have the smallest mean square error and the best fit. Under λ.min, the model screened out 20 nonzero variables, and the variables and their coefficient sizes are shown in Fig. 3 . Construction of predictive models The ratio of the training set to the test set of the model is 8:2, and the best model parameters are determined by 10 fold cross-validation. The ROC curve of the model (Fig. 4 ) and the results of evaluating the machine learning model are shown in Table 2 . The curves and model evaluation results show that the extra trees model has the highest Area Under Curve (AUC)value of 0.846 among the 8 models in the test set. AdaBoost also performs well (AUC = 0.835), and the decision tree has the lowest AUC value (0.642). Table 2 Model performance Model Accuracy Precision Recall F1 Score AUC Decision Tree 0.698 0.434 0.525 0.469 0.642 SVM 0.730 0.518 0.326 0.396 0.723 Naive Bayes-Bernoulli 0.743 0.625 0.079 0.137 0.672 Random Forest 0.773 0.669 0.461 0.544 0.802 AdaBoost 0.800 0.674 0.530 0.591 0.835 GBDT 0.795 0.668 0.576 0.612 0.817 Extra Trees 0.822 0.845 0.479 0.604 0.846 LightGBM 0.798 0.687 0.555 0.606 0.827 GBDT: Gradient boosting decision tree SHAP value calculation for features in the extra trees model Figure 5 displays the top 20 features and their significance in the feature importance analysis. According to the feature ranking and value, a high preoperative total bilirubin level was a significant risk factor for the development of HB. Other potential risk factors include the number of blood transfusions, the volume of blood loss, the international normalized ratio (INR), and the use of dexmedetomidine (DEX). The SHAP summary diagram (Fig. 6) and SHAP example diagram (Fig. 7 ) of the HB model were developed using SHAP interpretability analysis. Explanatory analysis SHAP helps reveal how feature impacts differ from person to person. Although all of the samples in Fig. 7abc have predictive values larger than the base value (positive sample), the positive drivers are not the same. For example, at 7a, the main drivers of positive contribution are PLTT = 3, RBCT = 800, DEX = 0, and INR = 1.44; at 7b, the positive drivers are Hb = 177, UA = 556, PH = 0, and TB = 28.2; and at 7c, the superimposed contribution of more features causes the predicted value to deviate from the baseline value toward the target category. Similarly, in the negative samples (which tended to be predicted as HB = 0), the negative drivers were distinct (7d, 7e, and 7f). DISCUSSION A machine learning methodology was employed to develop a predictive model for the risk of HB in patients following cardiac surgery. The extra trees model surpassed other strategies in predicting and classifying this dataset, and the model evaluation demonstrated superior performance. The four primary parameters in the model were preoperative total bilirubin level, preoperative INR, intraoperative erythrocyte transfusion, and DEX utilization. The primary risk factors for postoperative HB included elevated preoperative total bilirubin and INR levels, increased intraoperative erythrocyte infusion volume, and the absence of DEX administration. Classical prediction models can estimate the likelihood of postoperative hyperbilirubinemia; however, their risk factor ranking is less effective than that of machine learning, and they are affected by variable collinearity and sample size 12 . In contrast to the prior outcomes derived from our center's conventional predictive model, preoperative total bilirubin and intraoperative red blood cell infusion volume emerged as significant predictive indicators, whereas traditional factors, including operation time and aortic occlusion time, displayed minimal influence on the prediction. The intraoperative application of DEX may be acknowledged as a crucial component affecting HB. Utilizing the training set data, we developed eight prevalent supervised machine learning models, evaluated their performance with the test set data, and ultimately selected extra trees based on the ROC curve, calibration curve, decision curve, confusion matrix, and model performance evaluation metrics, including accuracy and sensitivity. The efficacy of the AdaBoost model is likewise substantial. Two risk factors, increased preoperative total bilirubin and intraoperative blood transfusion, were also discovered when used traditional modeling approaches 3 – 5 , 13 . This aligns with the findings of earlier research on conventional risk modeling and suggests that the results will remain consistent. Elevated preoperative total bilirubin levels suggested a potential loss in liver reserve function, resulting in the onset of postoperative HB 4 . While the volume of blood transfused is recognized as a risk factor for postoperative hyperbilirubinemia, the storage period of transfused red blood cells may also be an overlooked component. A NEJM study indicated that the long-term red cell storage group (exceeding 21 days) exhibited elevated bilirubin levels compared to the short-term storage group (under 10 days) 14 . This transpired due to the hemolysis of red blood cells during storage. Post-transfusion, preserved red blood cells have an elevated incidence of hemolysis in vivo owing to increased fragility. Nevertheless, we excluded this indicator due to the retrospective nature of the data collection.T Limited findings indicate that increased preoperative INR is recognized as a risk factor for liver injury. The standard INR range at our institution was 0.8–1.15, but the INR in the HB group was marginally increased at 1.2 ± 0.3 in comparison to the other groups. Our prior research demonstrated that even a minor alteration in the INR can result in inappropriate postoperative coagulation. These alterations may enhance drainage during the initial postoperative stage, resulting in increased blood transfusions and subsequent hyperbilirubinaemia, primarily observed in the early postoperative period. This observation coincides with our findings in another article, which indicate that postoperative hyperbilirubinaemia is predominantly evident within the first three days following surgery. It is noteworthy that the intraoperative administration of DEX can be identified as a significant factor affecting HB. Numerous studies have indicated that DEX may enhance hepatic recovery through the following mechanisms: 1. Inhibition of the NLRP3 inflammasome: Wu et al. in LO2 hepatocytes demonstrated that DEX can reduce hepatic ischemia/reperfusion (I/R)damage through the PI3K/AKT/Nrf2-NLRP3 pathway 15 . 2. Oxidative stress and endoplasmic reticulum stress: In a rat model, Zhang et al. reported that DEX could reduce hepatic I/R injury in vivo and in vitro 16 . Ferroptosis: Dexamethasone (DEX) elevates iron concentration in hepatic I/R injury through the upregulation of the Nrf2/GPx4 signaling pathway 17 . 4. NO pathway: Lee's investigations demonstrated that serum aminotransferase levels exhibited a reduction in a dose-dependent manner following administration of DEX 18 . The limitations of the investigation are outlined as follows: First, given that the patients included in the study predominantly originated from Asian populations, the applicability of the model to our demographic may not extend to other nations and regions. Consequently, it is imperative that future research encompasses a more extensive and diverse population. Second, this was a retrospective study conducted at a single location, which may have resulted in selection bias. Third, all participants in this study were sourced from a single institution; therefore, there is an immediate necessity for a more extensive external validation cohort. Fourth, given that the existing criteria for postoperative hyperbilirubinemia have not been standardized, the findings cannot be generalized to other studies employing varying criteria. 5 . CONCLUSIONS The Extra Trees model is a proficient machine learning framework for predicting the progression of hyperbilirubinemia subsequent to cardiac surgery. The most significant risk factors for postoperative HB include elevated preoperative total bilirubin and INR levels, increased intraoperative red blood cell transfusions, and nonuse of DEX. Abbreviations Area Under Curve (AUC) Cardiopulmonary bypass time (CPBT) Dexmedetomidine (DEX) Ejection fraction (EF) Hyperbilirubinemia (HB) International normalized ratio (INR) Shapley additive explanation （SHAP） Total bilirubin (TBIL) Declarations Acknowledgements We thank Zhongkai Wu and Liting Kuang for their strong support in the study. Author contributions W.L. projected management, validation and editing. H.Z. and W.L. wrote the main manuscript text. All authors had read and approved the manuscript. Funding There is no funding support for this study. Data availability The datasets used and analyzed during the current study are available from the corresponding author upon reasonable request Ethics approval and consent to participate This study was approved by the ethics committee of the First Affiliated Hospital of Sun Yat-sen University (No. [2023]493). Owing to the retrospective nature of the study, the need for informed consent was waived by the ethics committee of the First Affiliated Hospital of Sun Yat-sen University. The study complied with all regulations and complied with the Helsinki Declaration. Consent for publication Not applicable. Competing interests The authors declare no competing interests References Mundth ED, Keller AR, Austen WG. Progressive hepatic and renal failure associated with low cardiac output following open-heart surgery. J Thorac Cardiovasc Surg. 1967;53(2):275–84. 10.1016/S0022-5223(19)43231-5 . Wang MJ, Chao A, Huang CH, et al. Hyperbilirubinemia after cardiac operation. J Thorac Cardiovasc Surg. 1994;108(3):429–36. 10.1016/S0022-5223(94)70252-7 . Wang Z, Xu J, Cheng X, Zhang L, Wang D, Pan J. Hyperbilirubinemia after surgical repair for acute type a aortic dissection: A propensity score-matched analysis. Front Physiol. 2022;13:1009007. 10.3389/fphys.2022.1009007 . Gao Y, Li D, Dong H, et al. Risk factors analysis of hyperbilirubinemia after off-pump coronary artery bypass grafting: a retrospective observational study. J Cardiothorac Surg. 2021;16(1):294. 10.1186/s13019-021-01678-6 . Chen X, Bai M, Zhang W, Sun S, Chen X. The incidence, risk factors, and prognosis of postoperative hyperbilirubinemia after cardiac surgery: a systematic review and meta-analysis. Annals Palliat Med. 2021;10(7):7247257–7247257. 10.21037/apm-21-410 . Song X, Liu X, Liu F, Wang C. Comparison of machine learning and logistic regression models in predicting acute kidney injury: A systematic review and meta-analysis. Int J Med Informatics. 2021;151:104484. 10.1016/j.ijmedinf.2021.104484 . Choi H, Lee JY, Sul Y, et al. Comparing machine learning and logistic regression for acute kidney injury prediction in trauma patients: A retrospective observational study at a single tertiary medical center. Medicine. 2023;102(33):e34847. 10.1097/MD.0000000000034847 . Soni J, Ansari U, Sharma D, Soni S. Predictive data mining for medical diagnosis: An overview of heart disease prediction. IJCA. 2011;17(8):43–8. 10.5120/2237-2860 . Uddin S, Khan A, Hossain ME, Moni MA. Comparing different supervised machine learning algorithms for disease prediction. BMC Med Inf Decis Mak. 2019;19(1):281. 10.1186/s12911-019-1004-8 . An Y. Hyperbilirubinemia after extracorporeal circulation surgery: A recent and prospective study. World J Gastroenterol. 2006;12(41):6722. 10.3748/wjg.v12.i41.6722 . Chen JW, Chen HA, Liu TC, Wu TE, Lu CJ. The potential of SHAP and machine learning for personalized explanations of influencing factors in myopic treatment for children. Medicina. 2024;61(1):16. 10.3390/medicina61010016 . Norman G, Monteiro S, Salama S. Sample size calculations: should the emperor’s clothes be off the peg or made to measure? BMJ. 2012;345(aug23 1):e5278–5278. 10.1136/bmj.e5278 . Sharma P, Ananthanarayanan C, Vaidhya N, Malhotra A, Shah K, Sharma R. Hyperbilirubinemia after cardiac surgery: An observational study. Asian Cardiovasc Thorac Annals. 2015;23(9):1039–43. 10.1177/0218492315607149 . Steiner ME, Ness PM, Assmann SF, et al. Effects of red-cell storage duration on patients undergoing cardiac surgery. N Engl J Med. 2015;372(15):1419–29. 10.1056/NEJMoa1414219 . Wu Y, Qiu G, Zhang H, et al. Dexmedetomidine alleviates hepatic ischaemia-reperfusion injury via the PI3K/AKT/Nrf2‐NLRP3 pathway. J Cell Mol Medi. 2021;25(21):9983–94. 10.1111/jcmm.16871 . Zhang S, Tang J, Sun C, et al. Dexmedetomidine attenuates hepatic ischemia-reperfusion injury-induced apoptosis via reducing oxidative stress and endoplasmic reticulum stress. Int Immunopharmacol. 2023;117:109959. 10.1016/j.intimp.2023.109959 . Zhang Y, Wei H, Wang M, et al. Dexmedetomidine alleviates ferroptosis following hepatic ischemia-reperfusion injury by upregulating Nrf2/GPx4-dependent antioxidant responses. Biomed Pharmacother. 2023;169:115915. 10.1016/j.biopha.2023.115915 . Lee JE, Jung H, Cho JD et al. The Role of Dexmedetomidine in Hepatic Ischemia-Reperfusion Injury Via a Nitric Oxide-Dependent Mechanism in Rats. Transplantation Proceedings . 2021;53(6):2060–2069. 10.1016/j.transproceed.2021.05.008 Additional Declarations No competing interests reported. Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-6456253","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":449104769,"identity":"9cefcb90-ea9b-4a54-a57c-6dca2ad5d2d5","order_by":0,"name":"Weibin Lin","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA7ElEQVRIiWNgGAWjYBACPmYGBmaGChsgydwAEzTAq4UNrOVMGpBkJFYLA0hxy2EgRbQWdt5j0oUN56P52xkbGH+21SU2sDdvk2CouYPHYXxp0jN33M6dcZixgZm37XBiA8+xMgmGY8/waOExu8175nZuA0gLY9uBxAaJHDMJxobDBLS0ncudfxjmMPk3RGk5kLsBqIWBt40ZaAsPQS3mv2ecSc7dCNRymOfcYeM2nrRii4RjuLXw858xNi6osMudd/7wwYc/yupk+9kPb7zxoQa3FhRwgJENElMMCcRpAIE/xCsdBaNgFIyCkQMAUPdQrwNTM50AAAAASUVORK5CYII=","orcid":"","institution":"Sun Yat-sen University","correspondingAuthor":true,"prefix":"","firstName":"Weibin","middleName":"","lastName":"Lin","suffix":""},{"id":449104770,"identity":"f97b9f96-e270-4c9d-a42a-a74a47ce9c10","order_by":1,"name":"Haowei Zhu","email":"","orcid":"","institution":"Guangzhou University of Chinese Medicine","correspondingAuthor":false,"prefix":"","firstName":"Haowei","middleName":"","lastName":"Zhu","suffix":""}],"badges":[],"createdAt":"2025-04-15 15:23:17","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-6456253/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-6456253/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":82140235,"identity":"b5ad7a49-45ea-4832-ae87-17b15b291c3d","added_by":"auto","created_at":"2025-05-07 06:29:57","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":114043,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/3ca2ad3e6b6d5540aef45623.jpeg"},{"id":82138010,"identity":"fe615d61-c0a4-4883-b448-75c7c49c3dd9","added_by":"auto","created_at":"2025-05-07 06:21:57","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":205441,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/970afd6262c744ba7e252919.jpeg"},{"id":82138007,"identity":"96fada18-ada6-480b-be81-fc56bc729389","added_by":"auto","created_at":"2025-05-07 06:21:57","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":328642,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/d19999af2f63eefc20885316.jpeg"},{"id":82142359,"identity":"34298569-2cc8-44cd-9cc3-d70771f08b97","added_by":"auto","created_at":"2025-05-07 06:37:57","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":123987,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/2aad4c351e8c1b704d64d5f8.jpeg"},{"id":82140236,"identity":"f3a7c36e-6cfe-4ed0-be49-b77b5f6330a2","added_by":"auto","created_at":"2025-05-07 06:29:57","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":59090,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage511.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/8dd1a22dbc0936ee2e87302c.jpeg"},{"id":82140238,"identity":"abadc14f-639b-4cd2-9593-ed4af6c47086","added_by":"auto","created_at":"2025-05-07 06:29:57","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":211046,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/0d823c5485371f12434b8abb.png"},{"id":82138014,"identity":"3b9b8420-4a7e-4da4-a45f-2a60695f1712","added_by":"auto","created_at":"2025-05-07 06:21:57","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":641673,"visible":true,"origin":"","legend":"\u003cp\u003eLegend not included with this version\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/c1744807c22aea8e100a0e4a.jpeg"},{"id":98423273,"identity":"26f62d14-0a18-415a-b052-dd5a4ed1ea6d","added_by":"auto","created_at":"2025-12-17 16:32:02","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":2339382,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-6456253/v1/4121f18f-a09c-4660-a850-e14582dfffae.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Application of machine learning to develop a predictive model for hyperbilirubinemia after cardiac surgery","fulltext":[{"header":"INTRODUCTION","content":"\u003cp\u003eSince 1967, Prof. Mundth has documented that elevated total bilirubin levels following heart surgery are a common complication\u003csup\u003e\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u003c/sup\u003e. About 10.1\u0026ndash;35.1% of individuals experience this syndrome following heart surgery, and it has the potential to significantly raise in-hospital mortality\u003csup\u003e\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u003c/sup\u003e. Traditional techniques (logistic regression, nomograms) have confirmed a number of risk factors for the development of hyperbilirubinemia(HB) following cardiac surgery, including hepatic insufficiency, infective endocarditis, the number of valve replacements, elevated right atrial pressure, blood transfusions, aging, diabetes mellitus, duration of extracorporeal circulation, and duration of aortic occlusion\u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eMachine learning is used for medical data mining and shows better predictive performance than traditional methods for other diseases\u003csup\u003e\u003cspan additionalcitationids=\"CR7 CR8\" citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR9\" class=\"CitationRef\"\u003e9\u003c/span\u003e\u003c/sup\u003e. While machine learning models have been applied to assess the risk of elevated total bilirubin levels in vascular surgery\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u003c/sup\u003e, there is a lack of corresponding research in the overall field of cardiac surgery. Therefore, a set of models capable of assessing the risk of elevated bilirubin levels is essential to help physicians identify high-risk patients and take preventive measures. Such models could help reduce perioperative complications and improve patient prognosis. By analyzing and predicting the risk of HB, we can take better care of our patients and reduce surgical complications.\u003c/p\u003e"},{"header":"METHODS","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy population and demographic characteristics\u003c/h2\u003e \u003cp\u003eRetrospective analysis was used in this study to gather data from patients who had heart surgery at Sun Yat-sen University's First Affiliated Hospital in 2021. Patients did not need to sign an informed consent form as the study used deidentified data. The following were the requirements for the study population's inclusion: 18 years of age or older and had extracorporeal circulation-assisted heart surgery. Patients with malignancies, those who had heart transplants, those without recorded postoperative bilirubin data, and those with preoperative bilirubin levels greater than 3 mg/dl were all excluded. This study has been approved by Sun Yat-sen University's First Affiliated Hospital Ethics Committee (approval number: [2023] 493). Considering the anonymization of the data, all identifying information was removed; therefore, the requirement for informed consent was waived. The study was conducted in a stepwise manner, strictly adhering to the relevant regulations and the principles of the Declaration of Helsinki. The detailed study flowchart is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003eOur hospital employs extracorporeal circulation with the median incision utilized in cardiac surgery. The anesthesiologist administers positive inotropic and vasoconstrictive agents to regulate blood pressure following aortic clamped. The extracorporeal circulation is discontinued after the requisite conditions are fulfilled. When the evacuation of extracorporeal circulation proves difficult, we initiate intra-aortic balloon pump or extracorporeal membrane oxygenation and generally sustain the overall reperfusion duration at approximately one-third of the aortic occlusion interval. The electronic medical records system grants access to all critical patient information, including demographics, comorbidities, and laboratory findings.\u003c/p\u003e \u003cp\u003eTo evaluate postoperative bilirubin (HB) levels, we utilized the criteria from other research and established the threshold as a total bilirubin above 3 mg/dL within 7 days post-surgery.\u003csup\u003e\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e,\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eMachine Learning Model Building and Model Evaluation\u003c/h3\u003e\n\u003cp\u003eWe partitioned the training and test sets in an 8:2 ratios to identify the optimal model. We developed eight distinct machine learning algorithms: na\u0026iuml;ve Bayes (NB), support vector machine (SVM), decision tree (DT), random forest (RF), light gradient boosting machine (LightGBM), gradient boosting decision tree (GBDT), Extra Trees, and AdaBoost. The efficacy of these models was evaluated by 10-fold cross-validation on the training dataset. A grid search was employed to identify the ideal hyperparameter combinations for each model to enhance its performance. We assessed the models' performance by calculating the F1 score, accuracy, recall, precision, and area under the curve, and we plotted the receiver operating characteristic (ROC) curve for each model.\u003c/p\u003e\n\u003ch3\u003eFeature importance and model interpretability analysis\u003c/h3\u003e\n\u003cp\u003eBased on the model evaluation results, the most effective machine learning model is chosen. Feature importance analysis was conducted on the selected models to ascertain the significance of each feature. This study involved screening 20 predictors by LASSO regression, selecting the optimal extra tree model for prediction, and performing a detailed interpretive analysis to elucidate the unique influence of each feature on model decisions. A transparent and interpretable prediction framework was developed by quantitatively assessing the contribution of attributes to the prediction outcomes using the Shapley additive explanation (SHAP) approach.\u003csup\u003e\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e\u003c/sup\u003e. Furthermore, to investigate the global and local interpretations of the model, comprehensive studies were performed using the SHAP summary plot and SHAP force plot, respectively. The summary plot depicted the distribution of SHAP values for each feature, illustrating the range of values and indicating the positive and negative directions as well as the strength of the feature's prediction of the target variable. Additionally, a force plot is employed to locally elucidate the prediction process for a particular sample, illustrating how the model integrates the positive and negative contributions of feature values in that sample and quantifies the cumulative effect of the features on the final prediction via an intuitive arrow plot. Utilizing these methodologies, we developed a predictive model to assess the likelihood of hyperbilirubinemia following heart surgery.\u003c/p\u003e\n\u003ch3\u003eStatistical analyses\u003c/h3\u003e\n\u003cp\u003eWithin our dataset, merely 13 data points were absent (\u0026lt;\u0026thinsp;5%) (left ventricular systole 9/401; right ventricular diastole 4/401), and these missing values were supplemented using repeated interpolation prior to subsequent statistical analysis. All statistical analyses were conducted using R software version 4.2.2. Continuous variables are presented as means\u0026thinsp;\u0026plusmn;\u0026thinsp;standard deviations, whereas non-normally distributed data are presented as medians [interquartile ranges]. Categorical variables are expressed as numerical values and percentages [n (%)].\u003c/p\u003e"},{"header":"RESULTS","content":"\u003cdiv id=\"Sec8\" class=\"Section2\"\u003e \u003ch2\u003eParticipant characteristics\u003c/h2\u003e \u003cp\u003e We encompassed a total of 401 participants. Among them, 115 individuals, accounting for 28.7% of the entire group, were found to be developed HB. Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e1\u003c/span\u003e shows the characteristics of the participants in this study.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 1\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eClinical features of HB patients and non-HB patients\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"3\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eVariables\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003enon-HB(n\u0026thinsp;=\u0026thinsp;286)\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eHB(n\u0026thinsp;=\u0026thinsp;115)\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAge (years)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e53.3\u0026thinsp;\u0026plusmn;\u0026thinsp;14.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e54.1\u0026thinsp;\u0026plusmn;\u0026thinsp;10.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMale, n (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e179 (62.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e73 (63.5%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBMI (kg/m2)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23.2\u0026thinsp;\u0026plusmn;\u0026thinsp;3.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e22.9\u0026thinsp;\u0026plusmn;\u0026thinsp;3.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSmoker\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e71 (24.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e26 (22.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDiabetes\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38 (13.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e11 (9.6%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAtrial fibrillation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e40 (14.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28 (24.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCOPD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e7 (2.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1 (0.9%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAHF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e8 (2.8%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (1.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePCI\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e34 (11.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10 (8.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePH\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e96 (33.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e43 (37.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePVD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e29 (10.1%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e2 (1.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRFD\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e30 (10.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e14 (12.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEmergency\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e4 (1.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e6 (5.2%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eHb(g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e132.3\u0026thinsp;\u0026plusmn;\u0026thinsp;21.1\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e131.5\u0026thinsp;\u0026plusmn;\u0026thinsp;21.4\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePLT(10\u003csup\u003e9\u003c/sup\u003e/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e220.7\u0026thinsp;\u0026plusmn;\u0026thinsp;67.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e214.9\u0026thinsp;\u0026plusmn;\u0026thinsp;72.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eALB(g/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e38.9\u0026thinsp;\u0026plusmn;\u0026thinsp;4.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e38.6\u0026thinsp;\u0026plusmn;\u0026thinsp;4.0\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAST(U/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e23.0 (10.0-221.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e28.0 (10.0-313.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eTBIL.pre(umol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e13.8\u0026thinsp;\u0026plusmn;\u0026thinsp;5.9\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21.6\u0026thinsp;\u0026plusmn;\u0026thinsp;9.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCr(umol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e80.0 (36.0-845.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e80.0 (50.0-581.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003epre.UA(umol/L)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e412.8\u0026thinsp;\u0026plusmn;\u0026thinsp;134.7\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e437.7\u0026thinsp;\u0026plusmn;\u0026thinsp;144.1\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eINR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1.1\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBNP(pg/mL)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e277.4 (10.0-89957.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e720.7 (10.0-32557.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eEF (%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e63.9\u0026thinsp;\u0026plusmn;\u0026thinsp;10.6\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e63.1\u0026thinsp;\u0026plusmn;\u0026thinsp;11.2\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePreoperative.intubation\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e1 (0.3%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e0 (0.0%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003ePreoperative.vasoactive\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e34 (11.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eACCT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e87.3\u0026thinsp;\u0026plusmn;\u0026thinsp;43.3\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e114.4\u0026thinsp;\u0026plusmn;\u0026thinsp;58.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eCPBT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e158.6\u0026thinsp;\u0026plusmn;\u0026thinsp;77.4\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e202.3\u0026thinsp;\u0026plusmn;\u0026thinsp;98.7\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMinimum.temperature\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e34.4\u0026thinsp;\u0026plusmn;\u0026thinsp;2.0\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e34.0\u0026thinsp;\u0026plusmn;\u0026thinsp;1.9\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eBlood loss\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e522.7\u0026thinsp;\u0026plusmn;\u0026thinsp;277.2\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e687.0\u0026thinsp;\u0026plusmn;\u0026thinsp;418.5\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRBC transfusion\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e230.0 (0.0-1500.0)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e460.0 (0.0-2600.0)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAlbumin\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e10 (3.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e10 (8.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDEX\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e193 (67.5%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e68 (59.1%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDopamine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e36 (12.6%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e17 (14.8%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDobutamine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e163 (57.0%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e71 (61.7%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNorepinephrine\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e237 (82.9%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e104 (90.4%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdrenaline\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c2\"\u003e \u003cp\u003e44 (15.4%)\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"left\" colname=\"c3\"\u003e \u003cp\u003e21 (18.3%)\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003e1 mg/dL\u0026thinsp;=\u0026thinsp;17.1 \u0026micro;mol/L\u003c/td\u003e\u003c/tr\u003e \u003ctr\u003e\u003ctd colspan=\"3\"\u003eACCT, aortic cross-clamp time; CPBT, cardiopulmonary bypass time\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eFeature selection\u003c/h3\u003e\n\u003cp\u003eLASSO regression was used for feature selection, and the penalty was increased by introducing an L1 regularization term. Two key λ values were determined through cross-validation: λ.min\u0026thinsp;=\u0026thinsp;0.0197 and λ.1 se\u0026thinsp;=\u0026thinsp;0.0547 (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e). Owing to the subsequent need for machine learning for prediction, the λ.min model was chosen to have the smallest mean square error and the best fit. Under λ.min, the model screened out 20 nonzero variables, and the variables and their coefficient sizes are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003e.\u003c/p\u003e\n\u003ch3\u003eConstruction of predictive models\u003c/h3\u003e\n\u003cp\u003eThe ratio of the training set to the test set of the model is 8:2, and the best model parameters are determined by 10 fold cross-validation. The ROC curve of the model (Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e) and the results of evaluating the machine learning model are shown in Table\u0026nbsp;\u003cspan refid=\"Tab2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. The curves and model evaluation results show that the extra trees model has the highest Area Under Curve (AUC)value of 0.846 among the 8 models in the test set. AdaBoost also performs well (AUC\u0026thinsp;=\u0026thinsp;0.835), and the decision tree has the lowest AUC value (0.642).\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab2\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003eModel performance\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"6\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003ePrecision\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eRecall\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003eF1 Score\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eAUC\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eDecision Tree\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.698\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.434\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.525\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.469\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.642\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.730\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.518\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.326\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.396\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.723\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eNaive Bayes-Bernoulli\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.743\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.625\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.079\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.672\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRandom Forest\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.773\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.669\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.461\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.544\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.802\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.800\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.674\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.530\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.591\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.835\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGBDT\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.795\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.668\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e\u003cb\u003e0.576\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e\u003cb\u003e0.612\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.817\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eExtra Trees\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e\u003cb\u003e0.822\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e\u003cb\u003e0.845\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.479\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.604\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e\u003cb\u003e0.846\u003c/b\u003e\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.798\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.687\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.555\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.606\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.827\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003ctfoot\u003e \u003ctr\u003e\u003ctd colspan=\"6\"\u003eGBDT: Gradient boosting decision tree\u003c/td\u003e\u003c/tr\u003e \u003c/tfoot\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eSHAP value calculation for features in the extra trees model\u003c/h2\u003e \u003cp\u003eFigure \u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e displays the top 20 features and their significance in the feature importance analysis. According to the feature ranking and value, a high preoperative total bilirubin level was a significant risk factor for the development of HB. Other potential risk factors include the number of blood transfusions, the volume of blood loss, the international normalized ratio (INR), and the use of dexmedetomidine (DEX). The SHAP summary diagram (Fig.\u0026nbsp;6) and SHAP example diagram (Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e7\u003c/span\u003e) of the HB model were developed using SHAP interpretability analysis.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eExplanatory analysis\u003c/h2\u003e \u003cp\u003eSHAP helps reveal how feature impacts differ from person to person. Although all of the samples in Fig.\u0026nbsp;7abc have predictive values larger than the base value (positive sample), the positive drivers are not the same. For example, at 7a, the main drivers of positive contribution are PLTT\u0026thinsp;=\u0026thinsp;3, RBCT\u0026thinsp;=\u0026thinsp;800, DEX\u0026thinsp;=\u0026thinsp;0, and INR\u0026thinsp;=\u0026thinsp;1.44; at 7b, the positive drivers are Hb\u0026thinsp;=\u0026thinsp;177, UA\u0026thinsp;=\u0026thinsp;556, PH\u0026thinsp;=\u0026thinsp;0, and TB\u0026thinsp;=\u0026thinsp;28.2; and at 7c, the superimposed contribution of more features causes the predicted value to deviate from the baseline value toward the target category. Similarly, in the negative samples (which tended to be predicted as HB\u0026thinsp;=\u0026thinsp;0), the negative drivers were distinct (7d, 7e, and 7f).\u003c/p\u003e \u003c/div\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eA machine learning methodology was employed to develop a predictive model for the risk of HB in patients following cardiac surgery. The extra trees model surpassed other strategies in predicting and classifying this dataset, and the model evaluation demonstrated superior performance. The four primary parameters in the model were preoperative total bilirubin level, preoperative INR, intraoperative erythrocyte transfusion, and DEX utilization. The primary risk factors for postoperative HB included elevated preoperative total bilirubin and INR levels, increased intraoperative erythrocyte infusion volume, and the absence of DEX administration.\u003c/p\u003e \u003cp\u003eClassical prediction models can estimate the likelihood of postoperative hyperbilirubinemia; however, their risk factor ranking is less effective than that of machine learning, and they are affected by variable collinearity and sample size\u003csup\u003e\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e\u003c/sup\u003e. In contrast to the prior outcomes derived from our center's conventional predictive model, preoperative total bilirubin and intraoperative red blood cell infusion volume emerged as significant predictive indicators, whereas traditional factors, including operation time and aortic occlusion time, displayed minimal influence on the prediction. The intraoperative application of DEX may be acknowledged as a crucial component affecting HB. Utilizing the training set data, we developed eight prevalent supervised machine learning models, evaluated their performance with the test set data, and ultimately selected extra trees based on the ROC curve, calibration curve, decision curve, confusion matrix, and model performance evaluation metrics, including accuracy and sensitivity. The efficacy of the AdaBoost model is likewise substantial.\u003c/p\u003e \u003cp\u003eTwo risk factors, increased preoperative total bilirubin and intraoperative blood transfusion, were also discovered when used traditional modeling approaches\u003csup\u003e\u003cspan additionalcitationids=\"CR4\" citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e,\u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e\u003c/sup\u003e. This aligns with the findings of earlier research on conventional risk modeling and suggests that the results will remain consistent. Elevated preoperative total bilirubin levels suggested a potential loss in liver reserve function, resulting in the onset of postoperative HB\u003csup\u003e\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e\u003c/sup\u003e. While the volume of blood transfused is recognized as a risk factor for postoperative hyperbilirubinemia, the storage period of transfused red blood cells may also be an overlooked component. A NEJM study indicated that the long-term red cell storage group (exceeding 21 days) exhibited elevated bilirubin levels compared to the short-term storage group (under 10 days)\u003csup\u003e\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u003c/sup\u003e. This transpired due to the hemolysis of red blood cells during storage. Post-transfusion, preserved red blood cells have an elevated incidence of hemolysis in vivo owing to increased fragility. Nevertheless, we excluded this indicator due to the retrospective nature of the data collection.T\u003c/p\u003e \u003cp\u003eLimited findings indicate that increased preoperative INR is recognized as a risk factor for liver injury. The standard INR range at our institution was 0.8\u0026ndash;1.15, but the INR in the HB group was marginally increased at 1.2\u0026thinsp;\u0026plusmn;\u0026thinsp;0.3 in comparison to the other groups. Our prior research demonstrated that even a minor alteration in the INR can result in inappropriate postoperative coagulation. These alterations may enhance drainage during the initial postoperative stage, resulting in increased blood transfusions and subsequent hyperbilirubinaemia, primarily observed in the early postoperative period. This observation coincides with our findings in another article, which indicate that postoperative hyperbilirubinaemia is predominantly evident within the first three days following surgery.\u003c/p\u003e \u003cp\u003eIt is noteworthy that the intraoperative administration of DEX can be identified as a significant factor affecting HB. Numerous studies have indicated that DEX may enhance hepatic recovery through the following mechanisms: 1. Inhibition of the NLRP3 inflammasome: Wu et al. in LO2 hepatocytes demonstrated that DEX can reduce hepatic ischemia/reperfusion (I/R)damage through the PI3K/AKT/Nrf2-NLRP3 pathway\u003csup\u003e\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e\u003c/sup\u003e. 2. Oxidative stress and endoplasmic reticulum stress: In a rat model, Zhang et al. reported that DEX could reduce hepatic I/R injury in vivo and in vitro\u003csup\u003e\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u003c/sup\u003e. Ferroptosis: Dexamethasone (DEX) elevates iron concentration in hepatic I/R injury through the upregulation of the Nrf2/GPx4 signaling pathway\u003csup\u003e\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u003c/sup\u003e. 4. NO pathway: Lee's investigations demonstrated that serum aminotransferase levels exhibited a reduction in a dose-dependent manner following administration of DEX\u003csup\u003e\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eThe limitations of the investigation are outlined as follows: First, given that the patients included in the study predominantly originated from Asian populations, the applicability of the model to our demographic may not extend to other nations and regions. Consequently, it is imperative that future research encompasses a more extensive and diverse population. Second, this was a retrospective study conducted at a single location, which may have resulted in selection bias. Third, all participants in this study were sourced from a single institution; therefore, there is an immediate necessity for a more extensive external validation cohort. Fourth, given that the existing criteria for postoperative hyperbilirubinemia have not been standardized, the findings cannot be generalized to other studies employing varying criteria.\u003csup\u003e\u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e\u003c/sup\u003e.\u003c/p\u003e"},{"header":"CONCLUSIONS","content":"\u003cp\u003eThe Extra Trees model is a proficient machine learning framework for predicting the progression of hyperbilirubinemia subsequent to cardiac surgery. The most significant risk factors for postoperative HB include elevated preoperative total bilirubin and INR levels, increased intraoperative red blood cell transfusions, and nonuse of DEX.\u003c/p\u003e"},{"header":"Abbreviations","content":"\u003cp\u003eArea Under Curve (AUC)\u003c/p\u003e\n\u003cp\u003eCardiopulmonary bypass time (CPBT)\u003c/p\u003e\n\u003cp\u003eDexmedetomidine (DEX)\u003c/p\u003e\n\u003cp\u003eEjection fraction (EF)\u003c/p\u003e\n\u003cp\u003eHyperbilirubinemia (HB)\u003c/p\u003e\n\u003cp\u003eInternational normalized ratio (INR)\u003c/p\u003e\n\u003cp\u003eShapley additive explanation\u0026nbsp;（SHAP）\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eTotal bilirubin (TBIL)\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe thank Zhongkai Wu and Liting Kuang for their strong support in the study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthor contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eW.L. projected management, validation and editing. H.Z. and W.L. wrote the\u003c/p\u003e\n\u003cp\u003emain manuscript text. All authors had read and approved the manuscript.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThere is no funding support for this study.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eData availability\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe datasets used and analyzed during the current study are available from the corresponding author upon reasonable request\u003c/p\u003e\n\u003cp\u003eEthics approval and consent to participate\u003c/p\u003e\n\u003cp\u003eThis study was approved by the ethics committee of the First Affiliated Hospital of Sun Yat-sen University (No. [2023]493). Owing to the retrospective nature of the study, the need for informed consent was waived by the ethics committee of the First Affiliated Hospital of Sun Yat-sen University. The study complied with all regulations and complied with the Helsinki Declaration.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eNot applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors declare no competing interests\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003eMundth ED, Keller AR, Austen WG. Progressive hepatic and renal failure associated with low cardiac output following open-heart surgery. J Thorac Cardiovasc Surg. 1967;53(2):275\u0026ndash;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0022-5223(19)43231-5\u003c/span\u003e\u003cspan address=\"10.1016/S0022-5223(19)43231-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang MJ, Chao A, Huang CH, et al. Hyperbilirubinemia after cardiac operation. J Thorac Cardiovasc Surg. 1994;108(3):429\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/S0022-5223(94)70252-7\u003c/span\u003e\u003cspan address=\"10.1016/S0022-5223(94)70252-7\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang Z, Xu J, Cheng X, Zhang L, Wang D, Pan J. Hyperbilirubinemia after surgical repair for acute type a aortic dissection: A propensity score-matched analysis. Front Physiol. 2022;13:1009007. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3389/fphys.2022.1009007\u003c/span\u003e\u003cspan address=\"10.3389/fphys.2022.1009007\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao Y, Li D, Dong H, et al. Risk factors analysis of hyperbilirubinemia after off-pump coronary artery bypass grafting: a retrospective observational study. J Cardiothorac Surg. 2021;16(1):294. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s13019-021-01678-6\u003c/span\u003e\u003cspan address=\"10.1186/s13019-021-01678-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen X, Bai M, Zhang W, Sun S, Chen X. The incidence, risk factors, and prognosis of postoperative hyperbilirubinemia after cardiac surgery: a systematic review and meta-analysis. Annals Palliat Med. 2021;10(7):7247257\u0026ndash;7247257. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.21037/apm-21-410\u003c/span\u003e\u003cspan address=\"10.21037/apm-21-410\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSong X, Liu X, Liu F, Wang C. Comparison of machine learning and logistic regression models in predicting acute kidney injury: A systematic review and meta-analysis. Int J Med Informatics. 2021;151:104484. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.ijmedinf.2021.104484\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2021.104484\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChoi H, Lee JY, Sul Y, et al. Comparing machine learning and logistic regression for acute kidney injury prediction in trauma patients: A retrospective observational study at a single tertiary medical center. Medicine. 2023;102(33):e34847. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1097/MD.0000000000034847\u003c/span\u003e\u003cspan address=\"10.1097/MD.0000000000034847\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSoni J, Ansari U, Sharma D, Soni S. Predictive data mining for medical diagnosis: An overview of heart disease prediction. IJCA. 2011;17(8):43\u0026ndash;8. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.5120/2237-2860\u003c/span\u003e\u003cspan address=\"10.5120/2237-2860\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUddin S, Khan A, Hossain ME, Moni MA. Comparing different supervised machine learning algorithms for disease prediction. BMC Med Inf Decis Mak. 2019;19(1):281. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1186/s12911-019-1004-8\u003c/span\u003e\u003cspan address=\"10.1186/s12911-019-1004-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAn Y. Hyperbilirubinemia after extracorporeal circulation surgery: A recent and prospective study. World J Gastroenterol. 2006;12(41):6722. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3748/wjg.v12.i41.6722\u003c/span\u003e\u003cspan address=\"10.3748/wjg.v12.i41.6722\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChen JW, Chen HA, Liu TC, Wu TE, Lu CJ. The potential of SHAP and machine learning for personalized explanations of influencing factors in myopic treatment for children. Medicina. 2024;61(1):16. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.3390/medicina61010016\u003c/span\u003e\u003cspan address=\"10.3390/medicina61010016\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eNorman G, Monteiro S, Salama S. Sample size calculations: should the emperor\u0026rsquo;s clothes be off the peg or made to measure? BMJ. 2012;345(aug23 1):e5278\u0026ndash;5278. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1136/bmj.e5278\u003c/span\u003e\u003cspan address=\"10.1136/bmj.e5278\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSharma P, Ananthanarayanan C, Vaidhya N, Malhotra A, Shah K, Sharma R. Hyperbilirubinemia after cardiac surgery: An observational study. Asian Cardiovasc Thorac Annals. 2015;23(9):1039\u0026ndash;43. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1177/0218492315607149\u003c/span\u003e\u003cspan address=\"10.1177/0218492315607149\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSteiner ME, Ness PM, Assmann SF, et al. Effects of red-cell storage duration on patients undergoing cardiac surgery. N Engl J Med. 2015;372(15):1419\u0026ndash;29. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1056/NEJMoa1414219\u003c/span\u003e\u003cspan address=\"10.1056/NEJMoa1414219\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWu Y, Qiu G, Zhang H, et al. Dexmedetomidine alleviates hepatic ischaemia-reperfusion injury via the PI3K/AKT/Nrf2‐NLRP3 pathway. J Cell Mol Medi. 2021;25(21):9983\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1111/jcmm.16871\u003c/span\u003e\u003cspan address=\"10.1111/jcmm.16871\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang S, Tang J, Sun C, et al. Dexmedetomidine attenuates hepatic ischemia-reperfusion injury-induced apoptosis via reducing oxidative stress and endoplasmic reticulum stress. Int Immunopharmacol. 2023;117:109959. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.intimp.2023.109959\u003c/span\u003e\u003cspan address=\"10.1016/j.intimp.2023.109959\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang Y, Wei H, Wang M, et al. Dexmedetomidine alleviates ferroptosis following hepatic ischemia-reperfusion injury by upregulating Nrf2/GPx4-dependent antioxidant responses. Biomed Pharmacother. 2023;169:115915. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.biopha.2023.115915\u003c/span\u003e\u003cspan address=\"10.1016/j.biopha.2023.115915\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLee JE, Jung H, Cho JD et al. The Role of Dexmedetomidine in Hepatic Ischemia-Reperfusion Injury Via a Nitric Oxide-Dependent Mechanism in Rats. \u003cem\u003eTransplantation Proceedings\u003c/em\u003e. 2021;53(6):2060\u0026ndash;2069. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003e10.1016/j.transproceed.2021.05.008\u003c/span\u003e\u003cspan address=\"10.1016/j.transproceed.2021.05.008\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"LASSO, Extra Trees, SHAP, Transfusion, Dexmedetomidine","lastPublishedDoi":"10.21203/rs.3.rs-6456253/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-6456253/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e \u003cb\u003eObjective\u003c/b\u003e This study aimed to develop a machine learning-based model to predict hyperbilirubinemia after cardiac surgery.\u003c/p\u003e \u003cp\u003e \u003cb\u003eMethods\u003c/b\u003e A retrospective analysis was performed on 554 patients who underwent cardiopulmonary bypass surgery at the First Hospital of Sun Yat-sen University in 2021. The predictors of HB were determined by Least Absolute Shrinkage and Selection Operator (LASSO)regression, and eight different machine learning algorithms were constructed, including naive Bayes (NB), support vector machine (SVM), and decision tree (DT), random forest (RF), light gradient boosting machine (LightGBM), gradient boosting decision tree (GBDT), extra trees, and adaptive boosting (AdaBoost). Shapley additive explanation (SHAP) is used for interpretability analysis of the model.\u003c/p\u003e \u003cp\u003e \u003cb\u003eResults\u003c/b\u003e A total of 401 patients were enrolled, and 20 features were identified via LASSO regression. Among the 8 algorithms, the Area Under Curve (AUC)value of extra trees was 0.846, which was superior to those of the other models. The top 4 features of SHAP analysis were the preoperative total bilirubin level and international normalized ratio (INR). Other important risk factors included intraoperative red blood cell infusion and dexmedetomidine (DEX) use.\u003c/p\u003e \u003cp\u003e \u003cb\u003eConclusions\u003c/b\u003e The extra-trees model is a good model for predicting the occurrence of hyperbilirubinemia after cardiac surgery. The most important risk factors for postoperative HB were an increase in total bilirubin and the INR before the operation, an increase in red blood cell transfusion during the operation and no use of DEX.\u003c/p\u003e","manuscriptTitle":"Application of machine learning to develop a predictive model for hyperbilirubinemia after cardiac surgery","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2025-05-07 06:21:52","doi":"10.21203/rs.3.rs-6456253/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"72c56a37-8f76-4dee-a7cc-49a0a8d57da3","owner":[],"postedDate":"May 7th, 2025","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2025-12-11T08:54:25+00:00","versionOfRecord":[],"versionCreatedAt":"2025-05-07 06:21:52","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-6456253","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-6456253","identity":"rs-6456253","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}