Machine learning-based prediction of 1-year mortality risk after off-pump coronary artery bypass grafting

Introduction

Cardiovascular diseases are the leading cause of death in low- and middle-income countries worldwide, accounting for an estimated 17 million deaths annually, a number projected to rise to 24 million by 2030 (1,2). These diseases have substantially increased the global healthcare burden (3), with coronary heart disease (CHD) being one of the most prevalent forms. Currently, coronary artery bypass grafting (CABG) remains the primary surgical treatment for CHD (4-6). Compared with CABG, off-pump coronary artery bypass grafting (OPCABG) significantly reduces the risk of intraoperative transfusion and postoperative acute kidney injury (7). Previous studies have reported that the 1-year postoperative mortality rate following CABG is approximately 7.0% (8,9), whereas the 3-month all-cause mortality rate after OPCABG is around 3.3% (10). The markedly lower postoperative mortality in OPCABG patients compared with those undergoing CABG may be attributed to the avoidance of cardiopulmonary bypass, which substantially reduces systemic inflammatory responses and ischemia-reperfusion injury (11,12).

Machine learning (ML) has been widely applied in clinical medicine (13-15), playing an increasingly important role in disease diagnosis and prognosis. Previous studies have shown that ML performs exceptionally well in predicting all-cause mortality among patients with cardiovascular diseases (16). Compared with traditional predictive models, ML can effectively reduce data redundancy and uncover complex nonlinear relationships between predictors and outcomes (17). In recent years, the development of multimodal ML has further expanded research from single-modality data to include textual, visual, and temporal information (18,19). Although several studies have explored ML-based prediction of postoperative survival outcomes in patients undergoing CABG (20), no prior research has specifically modeled postoperative survival in OPCABG patients. In the study, we extracted patient data from the Medical Information Mart for Intensive Care (MIMIC)-IV database, conducted feature dimensionality reduction, and developed five ML-based survival models—Gradient Boosting Machine (GBM), least absolute shrinkage and selection operator-Cox regression (Lasso-Cox), Cox Boosting (CoxBoost), eXtreme Gradient Boosting (XGBoost), and partial least squares regression-Cox (PLSRCox)—to predict 1-year postoperative survival in OPCABG patients. The optimal model was further externally validated using patient data from the electronic intensive care unit Collaborative Research Database (eICU-CRD) to assess its generalization capability. To our knowledge, this is the first study to apply ML for postoperative survival prediction in OPCABG patients, providing a valuable reference value for clinical risk assessment and decision-making. We present this article in accordance with the TRIPOD reporting checklist (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2271/rc).

Methods

Data sources

Data for patients who underwent OPCABG were obtained from the MIMIC-IV (version 3.1) and eICU-CRD (version 2.0). Prior to data extraction, the required institutional training was completed and data access authorization was obtained. Structured Query Language (SQL) and Navicat Premium 16.0 (PremiumSoft Cyber Tech, Hong Kong, China) were used for data retrieval. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments; however, as all data in these public databases are anonymized and time-shifted, this study was exempt from ethical review.

Data collection

Patients who underwent OPCABG were identified using the corresponding International Classification of Diseases, 9^th and 10^th revision (ICD-9/10) codes. The exclusion criteria were as follows: (I) duplicate records; (II) intensive care unit (ICU) stay <24 hours; (III) concurrent cardiac surgeries performed with OPCABG; (IV) age <18 or >80 years; and (V) history of malignancy. A total of 2,280 eligible patients from the MIMIC-IV database and 2,547 patients from the eICU-CRD were included.

The following categories of data were extracted and all the data were extracted from the first measurement of the patient in the ICU before surgery to ensure the rationality of the prediction time: (I) demographic characteristics: sex, age, height, weight, and body mass index (BMI); (II) vital signs: heart rate, respiratory rate, systolic blood pressure (SBP), and diastolic blood pressure (DBP); (III) clinical scores: Acute Physiology Score III (APSIII), Sequential Organ Failure Assessment (SOFA) score, Glasgow Coma Scale (GCS) score, and Charlson Comorbidity Index; (IV) laboratory variables: red blood cells (RBC), hemoglobin, platelets, white blood cells (WBC), red cell distribution width (RDW), creatinine, aspartate aminotransferase (AST), alanine aminotransferase (ALT), anion gap, bicarbonate, total bilirubin (TBIL), blood urea nitrogen (BUN), chloride, glucose, calcium, sodium, potassium, creatine kinase (CK), creatine kinase-myocardial band (CKMB), international normalized ratio (INR), prothrombin time (PT), and partial thromboplastin time (PTT); (V) medical history: chronic pulmonary disease, chronic kidney disease (CKD), myocardial infarction, heart failure, and cerebrovascular disease.

Model construction and evaluation

To identify the most predictive variables, univariate Cox regression analysis was first performed on all features, and statistically significant features were further included in multivariate Cox regression as potential predictors, and finally features were determined to construct the ML model. The MIMIC-IV dataset was randomly divided into training and testing sets at a ratio of 7:3. Univariate and multivariate Cox regression analyses were performed on the training cohort to identify risk factors and independent predictors associated with 1-year all-cause mortality following OPCABG. Based on these identified features, five survival ML models—GBM, Lasso-Cox, CoxBoost, XGBoost, and PLSRCox—were constructed. Five-fold cross-validation was applied to enhance model performance. Model efficacy was evaluated at 3 months, 6 months, and 12 months postoperatively across the training, testing, and validation cohorts. Key evaluation metrics included the receiver operating characteristic (ROC) curve, decision curve analysis (DCA), time-dependent concordance index (C-index), and time-dependent area under the receiver operating characteristic curve (AUC). After identifying the optimal model, Shapley Additive Explanations (SHAP) visualization was used to interpret the contribution of individual features to model predictions.

Outcome measure

The primary outcome was 1-year mortality risk following OPCABG in patients with CHD.

Statistical analysis

Data preprocessing was performed prior to analysis. Variables with missing values <20% were imputed using the survival-predictive mean matching (Survival-PMM) method, whereas variables with >20% missing data or samples with missing values exceeding this threshold were excluded. Duplicate and outlier samples were also removed. Because 1-year mortality accounted for only 1.27% of cases, a significant class imbalance was present. To address this, the survival Synthetic Minority Oversampling Technique (SMOTE) was applied to balance the dataset. Comparisons between survival and death groups were conducted as follows: continuous variables were expressed as mean ± standard deviation and analyzed using the t-test for normally distributed data or the Mann-Whitney U test for non-normally distributed data; categorical variables were expressed as frequency (percentage) and compared using the chi-square test. All statistical analyses and figure generation were performed using R software version 4.3.2 (R Foundation for Statistical Computing, Vienna, Austria). A P value <0.05 was considered statistically significant.

Results

Feature selection for ML models

After applying the predefined inclusion and exclusion criteria, a total of 2,280 patients from the MIMIC-IV database and 2,547 patients from eICU-CRD were included in the study. OPCABG cases from MIMIC-IV were randomly split into training and testing cohorts at a 7:3 ratio. In the training cohort, univariate and multivariate Cox regression analyses were performed to identify risk factors and independent predictors of 1-year survival. The final independent predictors included CK, RDW, TBIL, ALT, CKD, anion gap, and AST (Table 1). The patient selection process is shown in Figure 1.

Figure 1 Flow chart of patient screening. CoxBoost, Cox Boosting; ICU, intensive care unit; MIMIC, Medical Information Mart for Intensive Care; OPCABG, off-pump coronary artery bypass grafting; SHAP, Shapley Additive Explanations.

ML models in the training cohort

Using the above features, the ML models were trained in the training cohort. Model performance, evaluated at 3 months post-OPCABG, yielded the following AUCs: CoxBoost, 0.955 [95% confidence interval (CI): 0.919–0.991]; GBM, 0.920 (95% CI: 0.818–1.000); Lasso-Cox, 0.931 (95% CI: 0.866–0.996); XGBoost, 0.951 (95% CI: 0.918–0.984); and PLSRCox, 0.937 (95% CI: 0.881–0.993). At 6 months, the AUCs were: CoxBoost, 0.958 (95% CI: 0.929–0.987); GBM, 0.944 (95% CI: 0.876–1.000); Lasso-Cox, 0.945 (95% CI: 0.900–0.989); XGBoost, 0.948 (95% CI: 0.918–0.979); and PLSRCox, 0.951 (95% CI: 0.913–0.989). At 12 months, the AUCs were: CoxBoost, 0.961 (95% CI: 0.935–0.987); GBM, 0.950 (95% CI: 0.889–1.000); Lasso-Cox, 0.944 (95% CI: 0.905–0.983); XGBoost, 0.948 (95% CI: 0.919–0.976); and PLSRCox, 0.954 (95% CI: 0.919–0.988). Across all time points, CoxBoost consistently demonstrated the highest predictive performance. ROC curves at these time points are shown in Figures 2-4, while DCA curves are presented in Figures 5-7. Time-dependent C-index visualizations also favored CoxBoost (Figure 8), with time-dependent AUCs shown in Figure 9. The calibration curve of the training set is shown in Figure 10, and the time-dependent Brier score is presented in Figure 11.

Figure 2 ROC curve of the training set at 3 months after surgery. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 3 ROC curve of the training set at 6 months after surgery. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 4 ROC curve of the training set at 1-year after surgery. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 5 DCA curve of the training set at 3 months after surgery. CoxBoost, Cox Boosting; DCA, decision curve analysis; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 6 DCA curve of the training set at 6 months after surgery. CoxBoost, Cox Boosting; DCA, decision curve analysis; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 7 DCA curve of the training set at 1-year after surgery. CoxBoost, Cox Boosting; DCA, decision curve analysis; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 8 Training set time-dependent C-index. C-index, concordance index; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 9 Training set time-dependent AUC. AUC, area under the receiver operating characteristic curve; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 10 Calibration curve of training set. CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; OS, overall survival; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 11 Time-dependent Brier score of the training set. CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

ML models in the testing and validation cohorts

The performance of CoxBoost, GBM, Lasso-Cox, XGBoost, and PLSRCox was evaluated in the testing cohort, with external validation conducted using the eICU-CRD dataset. In the test cohort, the AUCs for CoxBoost at 3, 6, and 12 months were 0.864 (95% CI: 0.763–0.965), 0.897 (95% CI: 0.813–0.981), and 0.906 (95% CI: 0.830–0.982), respectively. The corresponding AUCs for the other models were: GBM: 0.817 (95% CI: 0.629–1.000), 0.860 (95% CI: 0.716–1.000), and 0.871 (95% CI: 0.742–1.000); Lasso-Cox: 0.848 (95% CI: 0.738–0.959), 0.879 (95% CI: 0.789–0.969), and 0.893 (95% CI: 0.805–0.980); XGBoost: 0.822 (95% CI: 0.630–1.000), 0.838 (95% CI: 0.685–0.991), and 0.844 (95% CI: 0.709–0.978); PLSRCox: 0.851 (95% CI: 0.752–0.950), 0.884 (95% CI: 0.801–0.967), and 0.888 (95% CI: 0.814–0.962). Across all time points, CoxBoost consistently demonstrated the highest predictive performance (Figures 12-14). Time-dependent C-index and time-dependent AUC visualizations in the testing cohort also favored CoxBoost (Figures 15,16). The time-dependent Brier scores of the test set and the external validation set are shown in Figures 17,18. External validation using the eICU-CRD cohort further confirmed the robustness of CoxBoost, with a 1-year AUC of 0.897 (Figure 19) and time-dependent C-index and AUC in the validation cohort (Figures 20,21). To further test the stability of the model, we conducted a sensitivity analysis on the survival-smote algorithm. The detailed results are shown in Table 2, indicating that the survival-smote algorithm effectively enhanced the performance of the model.

Figure 12 ROC curve at 3 months after surgery in the test set. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 13 ROC curve at 6 months after surgery in the test set. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 14 ROC curve at 1-year after surgery in the test set. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; ROC, receiver operating characteristic; XGBoost, eXtreme Gradient Boosting.

Figure 15 Time-dependent C-index of each model in the test set. C-index, concordance index; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 16 Time-dependent AUC of each model for the test set. AUC, area under the receiver operating characteristic curve; CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 17 Time-dependent Brier score of the test set. CoxBoost, Cox Boosting; GBM, Gradient Boosting Machine; Lasso-Cox, least absolute shrinkage and selection operator-Cox regression; PLSRCox, partial least squares regression-Cox; XGBoost, eXtreme Gradient Boosting.

Figure 18 Time-dependent Brier score of the validation set. CoxBoost, Cox Boosting.

Figure 19 ROC curve of validation set for CoxBoost model at 1 year after surgery. AUC, area under the receiver operating characteristic curve; CI, confidence interval; CoxBoost, Cox Boosting; ROC, receiver operating characteristic.

Figure 20 Time-dependent C-index for validation set of CoxBoost model. C-index, concordance index; CoxBoost, Cox Boosting.

Figure 21 Time-dependent AUC of validation set for CoxBoost model. AUC, area under the receiver operating characteristic curve; CoxBoost: Cox Boosting.

Model interpretation

Following evaluation of model performance across the training, testing, and validation cohorts, CoxBoost was identified as the optimal model and subsequently interpreted using SHAP. In the training cohort, force plots were generated for representative high-, medium-, and low-risk samples to illustrate relationships between internal features and predicted risk (Figure 22). A SHAP heatmap was generated for the first 50 samples (Figure 23). Feature importance, ranked by SHAP values, indicated the following order of contributions: CK, RDW, TBIL, ALT, CKD, anion gap, and AST (Figure 24), with a SHAP summary plot shown in Figure 25. SHAP analyses were also performed in the testing and validation cohorts to confirm feature importance rankings and facilitate cross-cohort comparison (Figure 26).

Figure 22 Visualization of SHAP force plots for different risk samples in the training set. ALT, alanine aminotransferase; AST, aspartate aminotransferase; RDW, red cell distribution width; SHAP, Shapley Additive Explanations; TBIL, total bilirubin.

Figure 23 Visualization of SHAP force maps for some samples of the training set. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CoxBoost, Cox Boosting; RDW, red cell distribution width; SHAP, Shapley Additive Explanations; TBIL, total bilirubin.

Figure 24 SHAP feature importance ranking of training set. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CoxBoost, Cox Boosting; RDW, red cell distribution width; SHAP, Shapley Additive Explanations; TBIL, total bilirubin.

Figure 25 SHAP summary diagram of the training set. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CoxBoost, Cox Boosting; RDW, red cell distribution width; SHAP, Shapley Additive Explanations; TBIL, total bilirubin.

Figure 26 Comparison of SHAP importance ranking between training set, test set, and validation set. ALT, alanine aminotransferase; AST, aspartate aminotransferase; CoxBoost, Cox Boosting; RDW, red cell distribution width; SHAP, Shapley Additive Explanations; TBIL, total bilirubin.

Discussion

We identified 2,280 patients from the MIMIC-IV database of whom 29 patients died within 1 year following OPCABG, corresponding to an all-cause mortality rate of 1.27%, consistent with previous studies (21). Multivariate Cox regression analysis identified that CK, RDW, TBIL, ALT, CKD, anion gap, and AST were independent predictors of 1-year postoperative survival. Among the five trained ML models, CoxBoost consistently demonstrated the highest predictive performance at 3, 6, and 12 months in both the training and testing cohorts. External validation using the eICU-CRD patient cohort further confirmed the model’s generalizability, with a 1-year AUC of 0.897 for OPCABG patients.

In 2021, Huang et al. (20) investigated postoperative survival risk factors among CABG patients aged ≥65 years and developed a predictive model. However, their approach was limited to a simple binary classification, did not incorporate time-to-event variables, and lacked external validation. In 2025, Jafarkhani et al. (1) conducted a systematic review of ML models predicting postoperative survival after CABG, identifying age, ejection fraction, and renal function as key risk factors. To date, no study has specifically modeled postoperative survival outcomes in OPCABG patients. In this study, we were the first to apply ML to predict 1-year postoperative survival in OPCABG patients and to explore the relationship between independent risk factors and 1-year survival outcomes. We found that OPCABG patients with pre-existing CKD had a significantly higher risk of adverse 1-year survival outcomes. This may be explained by the generally reduced preoperative renal reserve in CKD patients, who are more susceptible to severe acute kidney injury during intraoperative stress and postoperative medication exposure. These patients may subsequently develop multi-organ dysfunction, markedly worsening prognosis (22), consistent with previous research findings (23-25). RDW has been associated with prognosis in various diseases, such as diffuse large B-cell lymphoma and esophageal cancer (26). Elevated RDW often indicates chronic inflammation or pre-existing cardiac dysfunction, correlating with a higher risk of adverse 1-year survival outcomes in OPCABG patients. Inflammation can directly induce cardiomyocyte injury (27). The release of inflammatory cytokines can trigger overexpression of mitochondrial function–related genes and activation of apoptotic pathways, leading to a significant increase in CK levels (28), which is associated with poor prognosis in OPCABG patients. Abnormally elevated AST and ALT levels often indicate hepatic injury and inflammatory responses, which may contribute to higher postoperative mortality (29), consistent with prior studies (30). Previous research has also shown that hypoproteinemia is associated with poor postoperative wound healing (31) and impaired immune function (32,33). Accordingly, decreased TBIL levels may be linked to a higher risk of adverse 1-year survival outcomes. Additionally, abnormal variations in the anion gap often suggest metabolic disturbances and heightened systemic inflammatory responses, mechanisms that can lead to worse postoperative outcomes, therefore, in actual clinical decision-making, when significant changes occur in the above-mentioned risk factors of the patient, clinicians should intervene as early as possible to reduce the patient’s medium and long-term postoperative mortality risk. Although we have effectively predicted the 1-year mortality risk for patients undergoing OPCABG, we have not obtained valid data for verification for those not treated in the ICU. We also hope that future multicenter studies will conduct more in-depth exploration of this issue.

Limitations

Despite being the first study to predict 1-year mortality risk in OPCABG patients, several limitations should be acknowledged. First, due to data availability, predictions were limited to 1-year outcomes, and longer-term follow-up was not assessed. Second, the 1-year mortality rate was 1.27%, resulting in a class imbalance. Although the survival-specific SMOTE was applied, residual effects on model performance cannot be fully excluded, and the original data might be distorted and lead to overfitting. Therefore, we also used sensitivity analysis to further determine the stability of model performance. Third, all data were derived solely from the MIMIC-IV and eICU-CRD databases, which may introduce selection bias; therefore, future multicenter studies are warranted to validate and generalize these findings. Fourthly, due to the limitations of the variables provided by the database, we were unable to incorporate more clinical variables, such as left ventricular ejection fraction. The absence of certain features also prevented our survival ML model from being compared with previous cardiac surgery risk scoring systems (such as the European System for Cardiac Operative Risk Evaluation II and Society of Thoracic Surgeons Risk Score).

Conclusions

Using clinical data from the MIMIC-IV database for OPCABG patients, we identified independent predictors of 1-year mortality risk, including CK, RDW, TBIL, ALT, CKD, anion gap, and AST. Five ML models were developed and evaluated at 3, 6, and 12 months, with CoxBoost consistently demonstrating superior performance and robust external validity across multicenter datasets.

Acknowledgments

None.

Footnote

Reporting Checklist: The authors have completed the TRIPOD reporting checklist. Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2271/rc

Peer Review File: Available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2271/prf

Funding: None.

Conflicts of Interest: All authors have completed the ICMJE uniform disclosure form (available at https://jtd.amegroups.com/article/view/10.21037/jtd-2025-aw-2271/coif). The authors have no conflicts of interest to declare.

Ethical Statement: The authors are accountable for all aspects of the work in ensuring that questions related to the accuracy or integrity of any part of the work are appropriately investigated and resolved. The study was conducted in accordance with the Declaration of Helsinki and its subsequent amendments.

Open Access Statement: This is an Open Access article distributed in accordance with the Creative Commons Attribution-NonCommercial-NoDerivs 4.0 International License (CC BY-NC-ND 4.0), which permits the non-commercial replication and distribution of the article with the strict proviso that no changes or edits are made and the original work is properly cited (including links to both the formal publication through the relevant DOI and the license). See: https://creativecommons.org/licenses/by-nc-nd/4.0/.

References

1. Jafarkhani A, Imani B, Saeedi S, et al. Predicting Factors Affecting Survival Rate in Patients Undergoing On-Pump Coronary Artery Bypass Graft Surgery Using Machine Learning Methods: A Systematic Review. Health Sci Rep 2025;8:e70336. [Crossref] [PubMed]
2. Kannel WB, McGee D, Gordon T. A general cardiovascular risk profile: the Framingham Study. Am J Cardiol 1976;38:46-51. [Crossref] [PubMed]
3. Malmberg M, Gunn J, Rautava P, et al. Outcome of acute myocardial infarction versus stable coronary artery disease patients treated with coronary bypass surgery. Ann Med 2021;53:70-7. [Crossref] [PubMed]
4. Chang YC, Chiang JH, Lay IS, et al. Increased Risk of Coronary Artery Disease in People with a Previous Diagnosis of Carpal Tunnel Syndrome: A Nationwide Retrospective Population-Based Case-Control Study. Biomed Res Int 2019;2019:3171925. [Crossref] [PubMed]
5. Benjamin EJ, Virani SS, Callaway CW, et al. Heart Disease and Stroke Statistics-2018 Update: A Report From the American Heart Association. Circulation 2018;137:e67-e492. [Crossref] [PubMed]
6. Wang M, Earley M, Chen L, et al. Clinical outcomes and bacterial characteristics of carbapenem-resistant Klebsiella pneumoniae complex among patients from different global regions (CRACKLE-2): a prospective, multicentre, cohort study. Lancet Infect Dis 2022;22:401-12. [Crossref] [PubMed]
7. Zhang Y, Li L, Li Y, et al. Machine learning model-based risk prediction of severe complications after off-pump coronary artery bypass grafting. Adv Clin Exp Med 2023;32:185-94. [Crossref] [PubMed]
8. Graham MM, Ghali WA, Faris PD, et al. Survival after coronary revascularization in the elderly. Circulation 2002;105:2378-84. [Crossref] [PubMed]
9. Soltani MH, Rasti M, Namayandeh SM, et al. Short and long-term outcomes of patients with coronary artery bypass surgery. ARYA Atheroscler 2021;17:1-6. [Crossref] [PubMed]
10. Sergeant P, Wouters P, Meyns B, et al. OPCAB versus early mortality and morbidity: an issue between clinical relevance and statistical significance. Eur J Cardiothorac Surg 2004;25:779-85. [Crossref] [PubMed]
11. Sawant AC, Adhikari P, Narra SR, et al. Neutrophil to lymphocyte ratio predicts short- and long-term mortality following revascularization therapy for ST elevation myocardial infarction. Cardiol J 2014;21:500-8. [Crossref] [PubMed]
12. Wirtz PH, von Känel R. Psychological Stress, Inflammation, and Coronary Heart Disease. Curr Cardiol Rep 2017;19:111. [Crossref] [PubMed]
13. Caetano Dos Santos FL, Michalek IM, Laurila K, et al. Automatic classification of IgA endomysial antibody test for celiac disease: a new method deploying machine learning. Sci Rep 2019;9:9217. [Crossref] [PubMed]
14. Handelman GS, Kok HK, Chandra RV, et al. eDoctor: machine learning and the future of medicine. J Intern Med 2018;284:603-19. [Crossref] [PubMed]
15. Lee A, Taylor P, Kalpathy-Cramer J, et al. Machine Learning Has Arrived! Ophthalmology 2017;124:1726-8. [Crossref] [PubMed]
16. Glass C, Lafata KJ, Jeck W, et al. The Role of Machine Learning in Cardiovascular Pathology. Can J Cardiol 2022;38:234-45. [Crossref] [PubMed]
17. Si Y, Abdollahi A, Ashrafi N, et al. Optimized feature selection and advanced machine learning for stroke risk prediction in revascularized coronary artery disease patients. BMC Med Inform Decis Mak 2025;25:276. [Crossref] [PubMed]
18. Hua L, Han J, Zhang S, et al. Machine learning prediction of post-CABG atrial fibrillation using clinical and pharmacogenomic biomarkers. Front Med (Lausanne) 2025;12:1650700. [Crossref] [PubMed]
19. Wang K, Tian J, Zheng C, et al. Interpretable prediction of 3-year all-cause mortality in patients with heart failure caused by coronary heart disease based on machine learning and SHAP. Comput Biol Med 2021;137:104813. [Crossref] [PubMed]
20. Huang YC, Li SJ, Chen M, et al. Machine-Learning Techniques for Feature Selection and Prediction of Mortality in Elderly CABG Patients. Healthcare (Basel) 2021;9:547. [Crossref] [PubMed]
21. Carr BM, Romeiser J, Ruan J, et al. Long-Term Post-CABG Survival: Performance of Clinical Risk Models Versus Actuarial Predictions. J Card Surg 2016;31:23-30. [Crossref] [PubMed]
22. Chua TKT, Gao F, Chia SY, et al. Long-term mortality after isolated coronary artery bypass grafting and risk factors for mortality. J Cardiothorac Surg 2024;19:429. [Crossref] [PubMed]
23. Cooper WA, O'Brien SM, Thourani VH, et al. Impact of renal dysfunction on outcomes of coronary artery bypass surgery: results from the Society of Thoracic Surgeons National Adult Cardiac Database. Circulation 2006;113:1063-70. [Crossref] [PubMed]
24. Lin Y, Zheng Z, Li Y, et al. Impact of renal dysfunction on long-term survival after isolated coronary artery bypass surgery. Ann Thorac Surg 2009;87:1079-84. [Crossref] [PubMed]
25. Gallagher S, Kapur A, Lovell MJ, et al. Impact of diabetes mellitus and renal insufficiency on 5-year mortality following coronary artery bypass graft surgery: a cohort study of 4869 UK patients. Eur J Cardiothorac Surg 2014;45:1075-81. [Crossref] [PubMed]
26. Turcato G, Cervellin G, Cappellari M, et al. Early function decline after ischemic stroke can be predicted by a nomogram based on age, use of thrombolysis, RDW and NIHSS score at admission. J Thromb Thrombolysis 2017;43:394-400. [Crossref] [PubMed]
27. Garcia MA, Rucci JM, Thai KK, et al. Association between Troponin I Levels during Sepsis and Postsepsis Cardiovascular Complications. Am J Respir Crit Care Med 2021;204:557-65. [Crossref] [PubMed]
28. Zeng XM, Liu DH, Han Y, et al. Assessment of inflammatory markers and mitochondrial factors in a rat model of sepsis-induced myocardial dysfunction. Am J Transl Res 2020;12:901-11.
29. Wen M, Han Y, Ye J, et al. Peri-operative risk factors for in-hospital mortality in acute type A aortic dissection. J Thorac Dis 2019;11:3887-95. [Crossref] [PubMed]
30. van Boxtel AG, Bramer S, Soliman Hamad MA, et al. Perioperative serum aspartate aminotransferase level as a predictor of survival after coronary artery bypass grafting. Ann Thorac Surg 2012;94:1492-8. [Crossref] [PubMed]
31. Estrella EP, Lee EY. A retrospective, descriptive study of sacral ulcer flap coverage in nonambulatory patients with hypoalbuminemia. Ostomy Wound Manage 2010;56:52-9.
32. Borthakur G, Lin E, Faderl S, et al. Low serum albumin level is associated with cytomegalovirus reactivation in patients with chronic lymphoproliferative diseases treated with alemtuzumab (Campath-1H)-based therapies. Cancer 2007;110:2478-83. [Crossref] [PubMed]
33. Takagi K, Yamamori H, Morishima Y, et al. Preoperative immunosuppression: its relationship with high morbidity and mortality in patients receiving thoracic esophagectomy. Nutrition 2001;17:13-7. [Crossref] [PubMed]