Multimodal Data‐Driven Prognostic Model for Predicting Long‐Term Prognosis in Patients With Ischemic Cardiomyopathy and Heart Failure With Preserved Ejection Fraction After Coronary Artery Bypass Grafting: A Multicenter Cohort Study Authors: Jun Wang, Yijun Wang, Shoupeng Duan, Li Xu, Yanan Xu, Wenyuan Yin, Yi Yang, Bing Wu, Jinjun Liu Published: 2024 ABSTRACT AbstractBackgroundLimited data from the literature are available to assess the efficacy of coronary artery bypass grafting in patients with ischemic cardiomyopathy and heart failure with preserved ejection fraction. Therefore, our objective was to use machine learning techniques integrating clinical features, biomarker data, and echocardiography data to enhance comprehension and risk stratification in patients diagnosed with ischemic cardiomyopathy and heart failure with preserved ejection fraction who have undergone coronary artery bypass grafting surgery.Methods and ResultsFor this study, 294 patients with ischemic cardiomyopathy and heart failure with preserved ejection fraction who underwent coronary artery bypass grafting surgery were assigned to the development cohort (n=176) and the independent validation cohort (n=118). A total of 52 clinical variables were extracted for each patient. The principal clinical end point was the incidence of major adverse cardiovascular events, encompassing cardiac mortality, acute myocardial infarction, acute heart failure, and graft failure. From least absolute shrinkage and selection operator regression, 4 predictors were selected for the final prediction nomogram: diabetes, hypertension, the systemic immune‐inflammation index, and NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide). The prediction nomogram achieved satisfactory prediction performance in both the development cohort (C index, 0.768 [95% CI, 0.701–0.835]) and independent validation cohort (C index, 0.633 [95% CI, 0.521–0.745]). Adequate calibration was noted for the likelihood of major adverse cardiovascular events in both the development and independent validation cohorts. Decision curve analysis confirmed the clinical usefulness of the established prediction nomogram.ConclusionsA clinically feasible prognostic model, based on preoperative multimodal data, was developed for risk stratification of patients with ischemic heart and heart failure with preserved ejection fraction who receive coronary artery bypass grafting surgery.Registration https://www.chictr.org.cn; Unique identifier: ChiCTR2300074439.Keywords: coronary artery bypass grafting, heart failure with preserved ejection fraction, multimodal data, prediction nomogramSubject Categories: Heart Failure Nonstandard Abbreviations and Acronyms heart failure with preserved ejection fraction major adverse cardiovascular event Clinical Perspective What Is New? Our study presents a novel prognostic model that integrates multimodal data to accurately predict the risk of major adverse cardiovascular events in patients with ischemic cardiomyopathy and heart failure with preserved ejection fraction who have undergone coronary artery bypass grafting. Our study presents a novel prognostic model that integrates multimodal data to accurately predict the risk of major adverse cardiovascular events in patients with ischemic cardiomyopathy and heart failure with preserved ejection fraction who have undergone coronary artery bypass grafting. What Are the Clinical Implications? The incorporation of easily replicable and straightforward biomarkers, such as the systemic immune‐inflammation index and NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide), alongside conventional risk factors, enhances the precision of risk stratification, enabling more individualized and efficacious patient management strategies following coronary artery bypass grafting. The incorporation of easily replicable and straightforward biomarkers, such as the systemic immune‐inflammation index and NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide), alongside conventional risk factors, enhances the precision of risk stratification, enabling more individualized and efficacious patient management strategies following coronary artery bypass grafting. By identifying individuals at high risk, the model supports tailoring therapeutic strategies, facilitating prompt implementation of intensified surveillance, and aggressive management of modifiable risk factors. By identifying individuals at high risk, the model supports tailoring therapeutic strategies, facilitating prompt implementation of intensified surveillance, and aggressive management of modifiable risk factors. The prevalence of heart failure with preserved ejection fraction (HFpEF) is increasing, particularly in the context of aging populations, a systemic proinflammatory state, and rising rates of metabolic disorders such as diabetes. 1 , 2 , 3 , 4 Unlike heart failure with reduced ejection fraction, which benefits from effective evidence‐based treatments, HFpEF lacks targeted therapies that have been proven to enhance outcomes. 1 , 2 The complexity of HFpEF is underscored by this gap, because it encompasses multiple pathophysiological pathways and is frequently complicated by comorbid conditions such as coronary heart disease. 1 , 2 In particular, the coexistence of HFpEF and ischemic heart disease presents a significant challenge, because it not only complicates the clinical management of patients but also exacerbates their prognosis. Current treatment strategies, such as coronary artery bypass grafting (CABG), have well‐established advantages in the general population for alleviating ischemic symptoms and enhancing survival outcomes. 5 However, mounting evidence suggests that up to 40% to 50% of patients may experience treatment failure within 10 years after CABG. 6 , 7 Notably, the effectiveness of CABG in the context of ischemic cardiomyopathy and HFpEF remains unclear, because the distinctive pathophysiology of HFpEF and its interplay with ischemic mechanisms pose challenges to understanding its impact. This discrepancy highlights the imperative need for multimodal data predictive tools to predict outcomes in patients with HFpEF undergoing CABG, which would facilitate tailored treatment strategies to optimize results. Moreover, the precise and prompt identification of patients at high risk for long‐term adverse outcomes, based on modifiable risk factors, represents a pivotal step beyond the conventional approach of initiating guideline‐directed medical therapy and managing postoperative risk factors. 8 , 9 , 10 , 11 , 12 , 13 , 14 The present multicenter study produced a longitudinal prognostic model using multimodal data to forecast the post‐CABG risk level in patients with ischemic cardiomyopathy and HFpEF. Furthermore, we evaluated the effectiveness of a prediction model based on the identified factors for individualized risk stratification of these patients. METHODS Data Sharing Agreement Individual participant data that underlie the results reported in this article, after deidentification, can be obtained from the corresponding author upon reasonable request. Ethics Approval and Consent to Participate Ethical approval for this study was granted by the local research ethics committee (Xinjiang Medical University Affiliated Hospital of Traditional Chinese Medicine, number 2023XE0144 and People's Hospital of Xinjiang Uygur Autonomous Region, number KY2023111001). Because of the retrospective design of the study, the need to obtain informed consent from eligible patients was waived by the ethics committee. Participants This current study identified cases of new‐onset ischemic cardiomyopathy and HFpEF in patients who underwent CABG from January 2018 to December 2019. Detailed information on the criteria for inclusion and exclusion in this study can be found in Table S1. The patients were included in a development cohort (n=176; data sets from the Xinjiang Medical University Affiliated Hospital of Traditional Chinese Medicine) and an independent validation cohort (n=118; data sets from the People's Hospital of Xinjiang Uygur Autonomous Region). The diagnostic criteria for HFpEF aligned with the recommendations of the China Heart Failure Guideline, 15 , 16 encompassing the manifestation of clinical symptoms and/or signs of heart failure, identification of cardiac structural and/or functional anomalies through echocardiographic examination, and elevation of serum NT‐proBNP (N‐terminal pro‐B‐type natriuretic peptide) levels. Our work is reported in line with the transparent reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD) guidelines (Table S1). 17 The flowchart of patient enrollment is presented in Figure 1. Flow diagram for patient selection. Clinical Data For this study, we recorded the clinical characteristics, comorbidities, laboratory results, and echocardiographic findings for all patients. Before venous blood collection, the participants were instructed to fast for a minimum of 8 hours, and blood specimens were typically obtained in the morning following hospital admission and before CABG. The systemic immune‐inflammation index (SII), a novel biomarker, was used to quantify systemic inflammation and its association with unfavorable outcomes in patients across all types of coronary heart disease. 18 The SII was calculated as follows: SII = platelet count × neutrophil count/lymphocyte count. Coronary Artery Bypass Grafting According to the prevailing best practices, CABG was scheduled for all patients. The recommended approach for arterial revascularization of the left anterior descending coronary artery used the left internal mammary artery. Additionally, alternative options such as saphenous venous grafts, free arterial grafts, or the right internal mammary artery were considered. The management of antiplatelet aggregation after CABG is individualized based on patient‐specific risk profiles and specific clinical circumstances. The duration and necessity of antiplatelet therapy are determined by conducting a comprehensive assessment of thrombotic versus bleeding risks, while adhering to the latest guidelines from relevant cardiovascular societies. 19 Clinical End Point Clinical follow‐up was conducted through outpatient visits or telephone interviews, with a median duration of 34.98 months among the study population. The primary clinical end point in this study was the occurrence of major adverse cardiovascular events (MACEs). MACEs encompassed a range of cardiac events, such as cardiac mortality, acute myocardial infarction, acute heart failure, and graft failure. Data Collection and Evaluation In this study, we implemented a comprehensive data quality review and verification process to ensure the accuracy and reliability of the data used for analysis. First, we established standardized protocols for data collection to minimize variability across different sites and ensure consistency in recording clinical characteristics, laboratory results, and echocardiographic findings. Data entry was conducted using a secure centralized database with built‐in validation checks to prevent entry errors. The accuracy of all records was cross‐verified through double data entry, using exclusively the data from routine clinical examinations to ensure comprehensive data collection. This approach was intentionally designed to enhance the integrity and robustness of the data set, ensuring that our findings are directly applicable to clinical practice. Moreover, based on the use of routinely collected clinical data, we circumvented the common issue of missing data that occurs in many studies. Quality control audits were conducted periodically to review data integrity and adherence to the study protocol, involving cross‐checking data entries and ensuring compliance with predefined data quality standards. To maintain high data quality standards, data collectors and research staff received training on data collection procedures and the importance of data quality, with ongoing supervision and feedback provided throughout the process. Through these measures, we aimed to ensure the robustness and validity of the data used in our analyses, thereby enhancing the credibility of our study's conclusions. Furthermore, experienced cardiovascular physicians collected the follow‐up data through clinical visits or telephone interviews conducted with either the patients themselves or their family members. Predictive Model Performance Three methods (discrimination, calibration, and clinical usefulness) were used to evaluate the performance of the prediction nomogram. 20 First, its discriminatory ability was analyzed by receiver operating characteristic curve analysis. In addition, the C indexes were adjusted using bootstrap resampling to estimate the anticipated predictive precision for future scenarios (500 iterations). Second, the model's calibration was evaluated through calibration plots (500‐bootstrap resampling) comparing the predicted and observed probabilities of MACEs at 1, 2, and 3 years. Third, decision curve analysis was performed to analyze the net benefit of the model at different threshold probabilities. The observed median incidence of MACEs was considered to determine these time points. In each case, a risk score was derived from the nomogram for each patient. The tertile point total was used to stratify patients in the development and independent validation cohorts into low‐risk, moderate‐risk and high‐risk categories. The survival distributions among the high‐risk, moderate‐risk, and low‐risk groups were compared using Kaplan‐Meier analysis. Statistical Analysis Least absolute shrinkage and selection operator regression analysis serves as a regularization technique that prevents overfitting by incorporating a penalty term into the loss function. This penalty term effectively shrinks coefficients associated with less significant features toward 0, facilitating feature selection and mitigating multicollinearity concerns. Although alternative algorithms exist, we found least absolute shrinkage and selection operator regression analysis particularly suitable for handling high‐dimensional data due to its inherent feature selection mechanism. 21 We used 10‐fold cross‐validation to select the optimal λ value, specifically choosing λmin=0.033 as the one that achieved the minimum mean squared error. This selection ensures an optimal balance between model complexity and error minimization, enabling us to construct a robust and interpretable predictive model. By focusing on λmin, our aim was to enhance the performance of the model while avoiding overfitting, thereby maintaining a straightforward and effective set of predictors. Moreover, using a 10‐fold cross‐validation technique can yield a more robust evaluation of the significance of patient parameters through iterative data partitioning into distinct training and test sets, enabling comprehensive assessment of the model's performance in each iteration. Additionally, we used R software (version 4.2.2) to perform the least absolute shrinkage and selection operator regression analysis, specifically using the glmnet package. To enhance the accuracy of variable selection and consider potential influencing factors, we performed multivariate analysis using backward stepwise Cox regression. We performed additional analyses, including interaction validation, calculation of the variance inflation factor, Schoenfeld residuals test, and proportional hazards assumption test, to offer a more comprehensive understanding of potential interactions among influencing factors and their collective predictive power. Statistical analysis of the study results was performed using SPSS 26 and R software 4.2.2. We used simple imputation methods facilitated by R's built‐in functions, primarily due to their straightforward implementation and the initial assessment of our data's missingness pattern, which appeared random and minimal. The strategy and procedure are detailed in Data S1. Categorical variables were summarized using descriptive statistics, including frequencies and proportions, whereas continuous variables were summarized using either medians and interquartile ranges or means±SDs. The nonparametric Mann‐Whitney U test was used to compare continuous variables exhibiting nonnormal distributions, and either the exact Fisher test or χ2 test was applied to compare the distributions of categorical variables between the development and independent validation cohorts. The Kaplan‐Meier method was used to determine the incidence of MACEs. A significance level of P<0.05 considered for all comparisons. RESULTS Patient Characteristics Our retrospective screening identified a total of 294 patients with new‐onset ischemic cardiomyopathy and HFpEF who underwent CABG and satisfied the eligibility criteria for this study. These patients were included in the development cohort (n=176; data sets from the People's Hospital of Xinjiang Uygur Autonomous Region) and the independent validation cohort (n=118; data sets from the Xinjiang Medical University Affiliated Hospital of Traditional Chinese Medicine). In our study, involving a total of 294 patients, 59 patients (20.07%) experienced MACEs. More specifically, in the training cohort, there were 35 cases of MACEs, accounting for 19.89% of this group. In the validation cohort, MACEs were observed in 24 patients, representing 20.34% of this cohort. All demographic data of the patients were thoroughly examined before modeling. No notable disparities were observed in terms of demographics, cardiovascular risk factors, physiological parameters, pertinent medical histories, and laboratory findings. The development and independent validation cohorts were found to be well‐matched in terms of baseline characteristics and clinical outcomes (Table 1). Table 1Baseline Patient Characteristics and MACEs (Clinical Outcome) in the Development and Independent Validation Cohorts* Characteristic or outcomeAll patients (n=294)Development cohort (n=176)Independent validation cohort (n=118) P valueASDMale sex, n (%)231 (78.57%)136 (77.27%)95 (80.51%)0.6050.079Age, y62.00 [55.00–68.00]63.00 [56.00–69.00]61.00 [55.00–66.00]0.1200.018Hypertension, n (%)154 (52.38%)90 (51.14%)64 (54.24%)0.6870.062Diabetes, n (%)112 (38.10%)66 (37.50%)46 (38.98%)0.8930.030Current smoking, n (%)146 (49.66%)91 (51.70%)55 (46.61%)0.4610.102Current drinking, n (%)86 (29.25%)55 (31.25%)31 (26.27%)0.4300.110Family history of CHD, n (%)8 (2.72%)4 (2.27%)4 (3.39%)0.7180.068Family history of hypertension, n (%)6 (2.04%)3 (1.70%)3 (2.54%)0.6870.058BMI, kg/m2 26.89 [24.82–29.22]26.44 [24.74–29.11]27.26 [25.15–29.30]0.2970.078Systolic blood pressure, mm Hg130.00 [116.00–142.00]127.00 [115.00–142.00]133.00 [120.00–144.00]0.0650.189Diastolic blood pressure, mm Hg74.46±12.4174.03±12.4875.10±12.340.4670.087HDL‐C, mmol/L0.98 [0.84–1.14]0.98 [0.83–1.15]0.98 [0.86–1.13]0.9150.048LDL‐C, mmol/L1.79 [1.40–2.24]1.83 [1.46–2.25]1.70 [1.33–2.19]0.2760.104Total cholesterol, mmol/L3.33 [2.86–4.08]3.37 [2.90–4.14]3.25 [2.82–4.02]0.5910.089Triglyceride, mmol/L1.37 [1.05–1.97]1.35 [1.05–1.88]1.42 [1.07–2.02]0.6390.001Fasting blood glucose, mmol/L5.52 [4.68–7.04]5.43 [4.64–6.88]5.64 [4.78–7.26]0.1460.198White blood cell count, 109/L7.15 [5.85–8.82]7.23 [5.63–8.67]7.10 [5.90–9.18]0.5380.100Neutrophil count, 109/L4.18 [3.20–5.62]4.15 [3.29–5.24]4.20 [3.16–6.20]0.4880.143Lymphocyte count, 109/L1.83 [1.44–2.32]1.81 [1.39–2.31]1.86 [1.52–2.34]0.3800.079Neutrophil count/lymphocyte count2.18 [1.61–3.31]2.18 [1.66–3.31]2.20 [1.60–3.32]0.9210.052Platelets count, 109/L221.50 [177.50–265.00]213.50 [172.00–264.00]228.00 [189.25–269.75]0.3570.053Platelets/lymphocyte count117.60 [88.01–155.43]117.42 [88.59–152.42]118.89 [86.35–161.35]0.7850.026SII490.38 [338.13–765.45]489.64 [332.97–739.75]492.55 [340.28–782.55]0.7980.095Hemoglobin, g/L139.00 [125.25–149.75]138.00 [122.00–148.00]140.00 [128.25–150.00]0.1290.226Hematocrit, %0.41 [0.38–0.45]0.41 [0.37–0.44]0.41 [0.39–0.45]0.7900.038Mean corpuscular volume, fL90.30±4.1390.53±4.0789.96±4.220.2530.137Mean corpuscular hemoglobin, pg30.20 [29.02–31.30]30.20 [29.10–31.40]30.10 [28.92–31.08]0.3530.127Mean platelet volume, fL10.20 [9.60–11.00]10.20 [9.60–10.93]10.30 [9.70–11.00]0.5230.004Platelet distribution width, %11.65 [10.50–13.28]11.60 [10.50–13.30]11.75 [10.43–13.17]0.9770.013Plateletcrit, %0.23 [0.19–0.27]0.22 [0.19–0.27]0.24 [0.20–0.28]0.0830.173Platelet large cell ratio, %26.50 [21.60–32.90]26.20 [21.17–32.42]26.80 [22.38–33.22]0.4120.154Red blood cell distribution width, %42.65 [40.52–44.80]42.45 [40.40–44.60]43.15 [41.00–45.18]0.1610.173Creatinine, μmol/L77.00 [66.43–91.99]77.45 [66.38–94.65]76.00 [66.55–89.75]0.5460.110Urea nitrogen, mg/dL5.70 [4.80–6.70]5.60 [4.77–6.65]5.80 [4.80–6.70]0.7400.017Uric acid, μmol/L329.55 [270.50–383.40]325.75 [275.70–380.22]334.50 [264.78–390.00]0.9750.046BUN/CREA0.07 [0.06–0.09]0.07 [0.06–0.09]0.08 [0.06–0.09]0.5590.093Fibrinogen, g/L3.13 [2.61–3.81]3.19 [2.66–3.81]2.96 [2.54–3.79]0.2980.077D‐dimer, μg/L0.84 [0.54–1.43]0.84 [0.54–1.44]0.85 [0.59–1.41]0.7800.031ALT, U/L27.90 [19.70–41.20]27.75 [19.78–42.02]27.95 [18.70–40.73]0.8550.075AST, U/L22.10 [18.90–29.80]21.90 [18.88–29.47]22.80 [19.02–30.62]0.4850.110ALT/AST1.19 [0.91–1.54]1.19 [0.91–1.56]1.14 [0.92–1.51]0.6070.008γ‐glutamyl transferase, U/L28.95 [18.90–45.08]28.95 [19.67–45.37]28.65 [18.02–44.20]0.7910.012Total protein, g/L65.50 [61.35–69.75]65.40 [61.70–69.40]65.53 [61.30–70.21]0.9140.035Total bilirubin, μmol/L12.70 [10.20–16.58]12.50 [10.00–15.83]13.30 [10.22–17.10]0.3560.074Creatine kinase MB, U/L12.54 [9.73–16.03]11.93 [9.50–15.29]13.07 [10.66–16.46]0.0530.081Troponin T, ng/L0.02 [0.01–0.10]0.02 [0.01–0.08]0.03 [0.01–0.11]0.3120.073NT‐proBNP, ng/L244.25 [142.75–689.50]212.90 [146.90–512.25]299.90 [138.90–918.25]0.2370.252Left atrial diameter, mm37.00 [35.00–39.00]37.00 [35.00–39.00]37.00 [35.00–39.75]0.6260.115Left ventricular interventricular septum thickness, mm9.00 [9.00–10.00]9.00 [9.00–9.00]9.00 [9.00–10.00]0.5500.083Left ventricular end‐diastolic diameter, mm51.00 [48.00–54.00]50.00 [48.00–53.00]52.00 [48.00–54.00]0.2510.090Left ventricular end‐systolic diameter, mm34.00 [32.00–37.00]33.00 [32.00–36.00]35.00 [32.00–38.00]0.3320.085Left ventricular ejection fraction, %60.00 [55.00–63.00]60.00 [56.00–63.00]60.00 [54.00–63.75]0.2000.172MACEs, n (%)59 (20.07%)35 (19.89%)24 (20.34%)1.0000.011Cardiac mortality, n (%)14 (4.76%)7 (3.98%)7 (5.93%)0.6230.090Acute myocardial infarction, n (%)4 (1.36%)2 (1.14%)2 (1.69%)1.0000.047Graft failure, n (%)33 (11.22%)22 (12.50%)11 (9.32%)0.5110.102Acute heart failure, n (%)8 (2.72%)4 (2.27%)4 (3.39%)0.7180.067*Data are presented as mean±SD for normally distributed data and median [interquartile range] for nonnormally distributed data. Categorical variables are presented as frequency (percentage). ALT indicates alanine aminotransferase; ASD, absolute standard difference; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; CHD, coronary heart disease; CREA, creatinine; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; MACEs, major adverse cardiovascular events; MB, muscle and brain; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; and SII, systemic immune‐inflammation index. Baseline Patient Characteristics and MACEs (Clinical Outcome) in the Development and Independent Validation Cohorts* Data are presented as mean±SD for normally distributed data and median [interquartile range] for nonnormally distributed data. Categorical variables are presented as frequency (percentage). ALT indicates alanine aminotransferase; ASD, absolute standard difference; AST, aspartate aminotransferase; BMI, body mass index; BUN, blood urea nitrogen; CHD, coronary heart disease; CREA, creatinine; HDL‐C, high‐density lipoprotein cholesterol; LDL‐C, low‐density lipoprotein cholesterol; MACEs, major adverse cardiovascular events; MB, muscle and brain; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; and SII, systemic immune‐inflammation index. Construction of the Nomogram According to the least absolute shrinkage and selection operator regression model, diabetes, hypertension, SII, and NT‐proBNP were identified as potential MACE predictors among a pool of 52 variables, as presented in the Table (Figures S1 and S2). Further multivariate Cox regression analysis additionally identified diabetes (hazard ratio [HR], 2.115 [95% CI, 1.011–4.424]; P=0.047), hypertension (HR, 2.431 [95% CI, 1.165–5.075]; P=0.018), SII (HR, 1.001 [95% CI, 1–1.001]; P=0.001), and NT‐proBNP (HR, 3.685 [95% CI, 1–1.001]; P=0.019) as independent risk factors for MACEs (Figure 2). The global Schoenfeld test P value of 0.0579 suggests a marginally nonsignificant outcome for the overall model, encompassing diabetes, hypertension, SII, and NT‐proBNP (Figure S3). Subsequently, these autonomous variables were integrated into a prognostic nomogram designed to predict the likelihood of MACEs in patients with ischemic cardiomyopathy and HFpEF who underwent CABG (Figure 3). The variance inflation factor values for the predictors were as follows (Table S2): diabetes, 1.1548; hypertension, 1.0040; SII, 1.0688; and NT‐proBNP, 1.1074. These variance inflation factor values are all below the commonly accepted threshold of 10, indicating that multicollinearity is not a significant concern in our model (Table S3). The proportional hazards assumption was assessed for each predictor, yielding the following results: diabetes (χ2=3.191, df=1, P=0.074), hypertension (χ2=1.475, df=1, P=0.225), SII (χ2=0.666, df=1, P=0.415), NT‐proBNP (χ2=2.200, df=1, P=0.138), and a global test of the combined prediction model (χ2=9.128, df=4, P=0.058). These findings suggest that our predictors adhere reasonably well to the proportional hazards assumption without significant violations. The individual P values for diabetes (0.074), hypertension (0.2245), SII (0.4145), and NT‐proBNP (0.138) indicate no significant deviations from the proportional hazard assumption for these predictors. The global test P value of 0.05798 is in close proximity to the conventional threshold of 0.05, suggesting a marginally nonsignificant outcome and indicating that the proportional hazards assumption holds reasonably well for the prediction model. Forest plot for analysis of HRs for independent prognostic variables identified by multivariate stepwise Cox regression analysis. HR indicates hazard ratio; NT‐proBNP, N‐terminal pro‐B‐type natriuretic peptide; and SII, systemic immune‐inflammation index. Prediction nomogram developed from variables identified by multivariate Cox regression analysis. To calculate the point total for each patient, a vertical line is traced from the respective characteristic value on the top row, and the scores for all 4 characteristics are summed in a single matrix as shown in the middle row. The probability of major adverse cardiac events is assessed by drawing a vertical line from the total point value to the bottom 3 rows. NT‐proBNP indicates N‐terminal pro‐B‐type natriuretic peptide; Pr, probability; and SII, systemic immune‐inflammation index. Nomogram Performance Model performance was assessed using the C index, which was obtained through bootstrap cross‐validation (500 iterations) in the development cohort (C index, 0.768 [95% CI, 0.701–0.835]) and independent validation cohort (C index, 0.633 [95% CI, 0.521–0.745]). In regard to the area under the curve, receiver operating characteristic curve analysis of the predictive ability of the nomogram for 1‐, 2‐, and 3‐year MACE risks produced area under the curve values of 0.869 (95% CI, 0.81–0.929), 0.774 (95% CI, 0.684–0.863), and 0.788 (95% CI, 0.711–0.866), respectively, in the development cohort and 0.573 (95% CI, 0.298–0.848), 0.561 (95% CI, 0.394–0.728), and 0.643 (95% CI, 0.518–0.767), respectively, in the independent validation cohort (Figure 4A and 4B). According to the calibration plots (500‐bootstrap resampling), the predicted probabilities from the nomogram were close to the observed probabilities for 1‐, 2‐, and 3‐year MACEs (Figure 5A and 5B). Moreover, decision curve analysis in the development cohort indicated that the threshold probability ranged from 1% to 47% for predicting MACEs over a 1‐year period, from 1% to 26% for predicting MACEs over a 2‐year period, and from 4% to 80% for predicting MACEs over a 3‐year period (Figure 6A through 6C). Decision curve analysis produced similar threshold probability ranges in the independent validation cohort of 1% to 15% for 1‐year MACE risk, 5% to 20% for 2‐year MACE risk, and 5% to 70% for 3‐year MACE risk (Figure 6D through 6F). The net benefit rate of the prediction nomogram was found to be higher than that of separate models run for each of the 4 variables (diabetes, hypertension, SII, and NT‐proBNP) for 1‐, 2‐ and 3‐year MACE risk in both the development and independent validation cohorts (Figure S4A–F). Finally, Kaplan‐Meier analysis showed that MACEs were more prevalent in the high‐risk and moderate‐risk groups than in the low‐risk group in both the development and independent validation cohorts (both P<0.05; Figure S5A and B). ROC curves for the prediction nomogram for 1‐, 2‐, and 3‐year MACEs in the development (A) and validation (B) groups. AUC indicates area under the curve; MACEs, major adverse cardiovascular events; and ROC, receiver operating characteristic. Calibration curves for the performance of the prediction nomogram in the development (A) and validation (B) cohorts. MACEs indicates major adverse cardiovascular events. DCA of the precision of the prediction of 1‐year, 2‐year, and 3‐year MACE risks in the development (A through C) and validation (D through F) cohorts. The DCA graph presents the expected net benefit of using the nomogram model to predict the risk of MACEs for individual patients. The net benefit was found to increase proportionally with the length of the model curve. DCA indicates decision curve analysis; and MACE, major adverse cardiovascular event. DISCUSSION The prognostic model presented in this study uses machine learning techniques to vertically visualize and predict the risk of MACEs in patients with HFpEF following CABG. This model integrates traditional cardiovascular risk factors, such as diabetes and hypertension, along with biomarkers including SII and NT‐proBNP. Integration of multimodal data into a model and visualization of the different variables through a column line chart can help physicians better understand a patient's condition in an intuitive way. By stratifying patients into higher and lower risk categories, our model supports the customization of therapeutic strategies, facilitating prompt implementation of intensified surveillance and aggressive management of modifiable risk factors post‐CABG for high‐risk individuals. Although coronary heart disease is more frequently observed in individuals with HFpEF, 22 less attention has been given to HfpEF among patients with coronary heart disease in clinical settings, especially in cases with ischemic cardiomyopathy. Moreover, current management strategies for patients with ischemic cardiomyopathy and HFpEF undergoing CABG often rely on generalized risk assessments that may not sufficiently consider the multifaceted nature of individual patient risk profiles. Notably, multiple evidence‐based approaches for managing cardiovascular disease suggest that biomarkers are indispensable for the rational evaluation and management of cardiovascular disease. 23 , 24 , 25 In our study, by incorporating a comprehensive panel of biomarkers that accurately reflect the underlying pathophysiology of ischemic cardiomyopathy and HFpEF, our model presents a more personalized approach to risk stratification. This is particularly crucial given the dynamic and progressive nature of ischemic cardiomyopathy and HFpEF, which is influenced by inadequate natriuretic peptide function and inflammation, 1 , 2 both of which are effectively captured by the biomarkers (SII and NT‐proBNP) used in our model. Meanwhile, among the numerous intricate and critical pathophysiological pathways associated with cardiovascular disease, 2 potent biomarkers, NT‐proBNP and SII, have previously been identified as having significant roles in acute illness. 26 , 27 , 28 , 29 The use of validated cardiac biomarkers also has been extended to long‐term monitoring and risk stratification in cases of nonacute illness. 26 , 27 , 28 , 29 Nevertheless, the specificity of these cardiac biomarkers is often inadequate due to individual variation. Hence, integration of multiple cardiac biomarkers and significant clinical factors is imperative for devising a preoperative evaluation instrument for individuals afflicted with ischemic cardiomyopathy and HFpEF and treated with CABG. Currently, clinical prediction nomograms could provide patients and their doctors with a quantitative risk value (probability) for the future likelihood they will develop a particular disease based on their current disease status and thus offer an intuitive and powerful scientific tool for personalized precision medicine. 9 , 10 , 11 , 12 , 13 , 14 , 20 , 30 Such prediction nomograms have been developed for patient outcomes following CABG. A previous study devised a clinical prediction nomogram to predict improvement of moderate ischemic mitral regurgitation after CABG that incorporated 5 preoperative clinical and echocardiographic predictors. 31 That prediction model was found to provide reliable preoperative assessment of moderate ischemic mitral regurgitation improvement after surgery, exhibiting superior discriminative ability and predictive accuracy compared with individual factors. 31 In another study, clinical data were combined to construct a nomogram for predicting the likelihood of in‐hospital mortality after CABG in patients with heart failure and reduced ejection fraction. 32 However, evidence from the literature is scant on reliable prognostic markers for the risk stratification of patients with ischemic cardiomyopathy and HFpEF who have been treated with CABG. Taken together, the available evidence collectively suggests that by identifying patients at higher risk of poor outcomes, these prediction models can be used by clinicians to optimize preoperative planning and postoperative management, thereby enhancing overall patient care. Such patients may need more frequent follow‐ups, tailored pharmacological treatments, and targeted lifestyle interventions that are not uniformly used among all patients after CABG. To our knowledge, the present study represents the first investigation of the incorporation of multimodal biomarker data in a longitudinal prognostic model for new‐onset ischemic cardiomyopathy and symptomatic HFpEF in patients who underwent CABG. The inclusion criteria ensured that all patients exhibited echocardiographic evidence of cardiac abnormalities and clinical symptoms indicative of heart failure. This rigorous selection process aimed to establish a homogeneous study population, thereby enhancing the model's applicability to symptomatic patients with HFpEF. Although this approach ensures the model's relevance for those currently experiencing symptoms, it inherently excludes asymptomatic patients who may have underlying structural abnormalities. The exclusion of asymptomatic individuals could potentially limit the applicability of our model in broader clinical settings where early detection and intervention are crucial. Incorporating asymptomatic patients with structural heart abnormalities in future studies would enhance the predictive power and generalizability of our model. Therefore, future research should aim to include both symptomatic and asymptomatic patients to provide a more comprehensive risk assessment tool that not only improves early intervention strategies but also optimizes resource allocation by identifying at‐risk individuals before symptom onset. Moreover, the current study evaluated multiple types of preoperative data to identify patients who are most likely to benefit from surgical intervention, thereby optimizing the allocation of resources and improving patient outcomes. The application of our predictive model has the potential to optimize health care resource allocation by strategically allocating intensive management efforts toward individuals with the highest likelihood of benefiting. This targeted approach not only holds promise for enhancing outcomes in high‐risk patients but also serves as a preventive measure against unnecessary interventions in those at lower risk. Additionally, this model serves as a valuable tool for facilitating discussions on informed consent. By providing patients and their families with a more comprehensive comprehension of the potential risks and benefits associated with undergoing CABG, the model promotes a heightened level of transparency and informed decision‐making process. This is particularly crucial for patient populations in which the assessment of risks and benefits can be intricate and nuanced. Still, additional research is needed to better characterize the intricate pathophysiological mechanisms of cardiac diseases and to devise more refined appraisal tools that yield precise biomarker‐based outcome prediction in a prompt and consistent manner with optimal precision and well‐defined individual variation. We acknowledge that the model's generalization ability may be compromised, as indicated by a significant decline in receiver operating characteristic curves for the predictive nomogram's performance in the development cohort compared with the independent validation cohort. Potential factors contributing to this discrepancy include the fact that the development and validation cohorts were derived from distinct hospitals. Variations in clinical practices, diagnostic protocols, and treatment strategies between these distinct hospitals may contribute to disparities in patient outcomes and model performance. Moreover, heterogeneity in comorbidities and disease severity within the patient populations from these 2 hospitals can potentially impact the generalizability of the model across diverse patient groups. Additionally, the model's performance assessment may have been limited by the relatively small sample size, particularly in the validation cohort, which could compromise its robustness. To address these challenges and enhance the model's generalization capability, we propose the following strategies. First, to improve performance estimation reliability and mitigate overfitting risks, a 10‐fold cross‐validation was implemented within the development cohort. Second, 500‐bootstrap resampling techniques were used to refine the model's performance metrics and assess their stability and robustness across diverse subsets of data. Last, to further evaluate the model's generalizability and reliability, it is crucial to validate it in additional independent cohorts from various geographic regions or health care settings. Notably, despite conducting independent external validation, the generalizability and practical applicability of the proposed model in real‐world practice may be influenced by various external factors, including variations in health care settings and clinical practices. 33 , 34 Discrepancies in resource availability, such as disparities in access to advanced diagnostic tools and specialized medical personnel, can significantly impact the implementation and performance of the model across diverse health care settings. Variations in health care infrastructure, including hospital facilities and patient management systems, may also influence the model's applicability and effectiveness in predicting outcomes. Inconsistencies observed among institutions on diagnostic protocols and criteria for ischemic cardiomyopathy and HFpEF can affect both the consistency and accuracy of model predictions. Additionally, revisions in treatment strategies, encompassing medication regimens and surgical techniques, can lead to disparities in patient outcomes and model performance. Therefore, variations in the prevalence and severity of ischemic cardiomyopathy and HFpEF across different regions and populations may impact the model's applicability. In a follow‐up study, the refinement of the prediction model's architecture, implementation of more robust validation methods, and establishment of prospective, observational multicenter studies with large sample sizes are imperative for future investigations. Study Limitations The present study has several limitations that must be acknowledged. First, the sample size was relatively small. Thus, a large‐scale prospective cohort study is needed to validate the accuracy and effectiveness of the nomogram in a more extensive population sample. Second, confounding variables that were not included in the analysis may have impacted the results. This may limit the extent to which the conclusions of the study can be generalized and applied to other patient populations. Third, the nomogram's predictions assumed that the relationships between the variables remain stable over time, which may not be accurate in real‐world scenarios where disease states fluctuate due to a variety of factors, such as medical advancements or lifestyle changes. Clinical practitioners should exercise caution when interpreting and applying the findings of this study. Further research is warranted to confirm and enhance the accuracy and efficacy of the developed prediction model and to explore potential sources of bias that could impact the results. Fourth, our cohort included only patients with symptomatic HFpEF, and we did not evaluate predictive values separately for asymptomatic patients with HFpEF. This focus on symptomatic patients may limit the generalizability of our findings to the broader HFpEF population, including those who are asymptomatic. In future investigations, our objective will be to incorporate asymptomatic patients in our study cohort to enhance the generalizability and comprehensiveness of our predictive model across a broader spectrum of patients with heart failure. Fifth, the C index in the validation cohort was 0.633, indicating a relatively weaker discriminatory ability of the model in this independent cohort. This observation could be attributed to the limited sample size, which might have compromised both the robustness and generalizability of our findings. Sixth, the small sample size and heterogeneity in the external validation cohort may contribute to deviations in the survival curves. Finally, the exclusion of certain comorbidities, such as overweight/obesity, atrial fibrillation, anemia, and chronic kidney disease, was driven by the statistical methods used and the specific inclusion criteria. Although these comorbidities are common in patients with HFpEF, their exclusion may have impacted the comprehensiveness of our predictive model. Future studies with larger and more heterogeneous cohorts are needed to refine these predictors and potentially incorporate additional comorbidities. CONCLUSIONS In this study, a vertical visualization risk stratification was devised to predict the 1‐, 2‐, and 3‐year MACE risks in patients diagnosed with ischemic cardiomyopathy and HFpEF after treatment with CABG. The machine‐learning‐driven vertical visualization prognostic model incorporates routinely collected data, including multimodal data (diabetes, hypertension, SII, and NT‐proBNP), and showed good predictive ability and clinical benefit. Further research is required to validate the nomogram in a larger prospective cohort and conduct long‐term follow‐up. Sources of Funding This work was supported by the grants from the Key Project of Natural Science Research of the Anhui Provincial Department of Education (2024AH051187). Disclosures None. Supporting information Data S1 Tables S1–S3 Figures S1–S5 Acknowledgments Author contributions: Y.Y., L.X., Y.W., B.W., and J.L. conceived and performed the study. J.W. and Y.X. performed the clinical study and wrote the article. S.D., J.W., and Y.X. performed data collection and analysis. All authors contributed to the article and approved the submitted version. This article was sent to Sula Mazimba, MD, MPH, Associate Editor, for review by expert referees, editorial decision, and final disposition. Supplemental Material is available at https://www.ahajournals.org/doi/suppl/10.1161/JAHA.124.036970 For Sources of Funding and Disclosures, see page 13. Contributor Information Yi Yang, Email: 644956539@qq.com. Bing Wu, Email: wubingstf@hbmu.edu.cn. Jinjun Liu, Email: byyfyliujinjun@163.com. References Associated Data Supplementary Materials Data S1 Tables S1–S3 Figures S1–S5