Machine learning algorithms to predict heart failure with preserved ejection fraction among patients with premature myocardial infarction

Background: Heart Failure with Preserved Ejection Fraction (HFpEF) in patients with Premature Myocardial Infarction (PMI) is a crucial factor affecting long-term prognosis. This study aims to develop a model based on a machine learning algorithm that can predict the risk of in-hospital HFpEF in patients with PMI early and quickly.

Methods: This prospective study consecutively included PMI patients from January 2017 to December 2022. Lasso-Logistic, XGBoost, Random Forest, K-Nearest Neighbor, and Support Vector Machine models were constructed. The prediction performance of the models was compared through AUC, Accuracy, Precision, F1 score, and Brier score. Shapley Additive exPlanations is used to explain the model. A prediction system was developed to identify high-risk patients.

Results: The study finally included 840 PMI patients. 268 (31.90%) developed in-hospital HFpEF. The XGBoost model has the best prediction performance (AUC 0.854; Accuracy 0.798; Precision 0.686; F1 score 0.586; Brier score 0.143). The final model included ten variables, which were Brain natriuretic peptide (BNP) > 100pg/ml, SYNTAX Score > 14.5, Age, Monocyte to Lymphocyte Ratio (MLR) > 0.3, Hematocrit (HCT) < 45%, Heart rate (HR) > 75 bpm, Body Mass Index (BMI) ≥ 24 kg/m², C-reactive Protein to Lymphocyte Ratio (CLR) > 2.83, Hypertension and Fibrinogen (Fg) > 4 g/L.

Conclusions: The explainable prediction model established based on the XGBoost algorithm can accurately predict the risk of in-hospital HFpEF in PMI patients and is available at https://hfpefpmi.shinyapps.io/apppredict/. This system is expected to assist clinicians in decision-making by providing timely, prioritized, and precise interventions for PMI patients, ultimately reducing the incidence of HFpEF and improving long-term prognosis.

Introduction

The risk of heart failure (HF) remains after premature myocardial infarction (PMI) and is a determinant of poor prognosis (1). Heart Failure with Preserved Ejection Fraction (HFpEF) is defined as HF with a left ventricular ejection fraction (LVEF) ≥50%. As a result of improved treatments and increased public awareness of HFpEF, the proportion of HFpEF in HF has increased to approximately 50%, affecting up to 32 million people worldwide (2). Previous studies have shown that 30%–60% of patients with HFpEF have a history of acute myocardial infarction (AMI) (3). The in-hospital occurrence of HFpEF after AMI significantly affects the long-term prognosis, especially in the younger population, where patients suffer from decreased or even loss of their labor capacity, which in turn triggers a significant increase in the cost of medical care and creates a massive burden on the family and society (4). Therefore, a rapid, accurate, and reliable comprehensive algorithm to assess the risk in this specific group is needed for the early identification of high-risk patients with poor prognosis and for the timely improvement of long-term prognosis and quality of survival through personalized treatment.

While logistic regression (LR) remains computationally efficient and highly interpretable, it is constrained by its linear assumption, making it difficult to capture complex interactions. Additionally, LR struggles with high-dimensional data, relies heavily on manual feature selection, and is sensitive to outliers and missing values. In contrast, machine learning (ML) has gained increasing attention in cardiovascular disease prediction due to its ability to process large-scale clinical data and uncover intricate patterns (5, 6). Various ML algorithms have been widely applied to classification and predictive tasks. Among them, Extreme Gradient Boosting (XGBoost) has gained prominence for its ability to integrate multiple weak learners through a gradient boosting tree framework, achieving superior predictive performance while effectively mitigating overfitting risk (7). This ensemble approach performs exceptionally well in both binary and multiclass classification tasks and often outperforms traditional logistic regression models in clinical prediction settings (8). Random Forest (RF) is another ensemble learning method that constructs multiple decision trees and combines their outputs through averaging or majority voting, thereby enhancing model generalizability and reducing variance (9). K-Nearest Neighbors (KNN) is a simple yet effective classification algorithm that assigns class labels based on the proximity of data points (10, 11). Support Vector Machine (SVM) constructs an optimal hyperplane to maximize the margin between different classes, making it well-suited for binary classification tasks in structured datasets (12). Previous studies have demonstrated that XGBoost excels in predicting 30-day readmission in HF patients (13), AMI outcomes (14), ICU mortality (15), and acute kidney injury (16), whereas Random Forest has shown superior performance in postoperative delirium prediction (17).

Despite numerous advances, few applications of machine learning have been addressed in predicting HFpEF events during hospitalisation in patients with PMI. Given that PMI patients may exhibit unique risk factors, this study leverages the strengths of multiple ML algorithms to develop a predictive model integrating multidimensional clinical features of PMI patients. Specifically, we aim to identify key predictive factors, quantify their contributions, and provide a foundation for risk stratification and early intervention in this high-risk cohort.

Material and methods

Study population

The flow of the study is shown in Figure 1. This is a single-center, prospective, observational cohort study. Consecutive patients admitted to Tianjin Chest Hospital for AMI between January 2017 and December 2022, meeting the PMI age threshold, were included in the PMI cohort. The inclusion criteria were as follows: (a) Age > 18 years, with male ≤ 50 years and female ≤ 55 years; (b) First diagnosis of AMI upon admission, meeting the universal definition including clinical symptoms, typical changes in the electrocardiogram, and elevated cardiac biomarkers (18, 19). (c) No occurrence of HF upon admission; (d) Undergoing coronary angiography (CAG) and primary percutaneous coronary intervention (PPCI). CAG and PPCI were performed by two or more cardiologists qualified in coronary diagnosis and treatment at our center.

Figure 1

Figure 1. Flow chart of the study. AMI, acute myocardial infarction; CA. G, coronary angiography; DCA, decision curve analysis; HF, heart failure; HFpEF, heart failure with preserved ejection fraction; PPCI, primary percutaneous coronary intervention; PMI, premature myocardial infarction; SHAP: shapley additive exPlanations; NRI, net reclassification improvement; IDI, integrated discriminant improvement.

The exclusion criteria were as follows: (a) Patients with severe liver and/or renal failure; (b) Patients with congenital heart disease and/or valvular heart disease; (c) Patients with severe inflammatory diseases and/or malignant tumors; (d) Patients with missing transthoracic echocardiography and/or other data; (e) Patients without signed informed consent. This study was conducted in accordance to the Declaration of Helsinki and was approved by the Ethics Committee of Tianjin Chest Hospital (No. 2017KY-007-01).

Data collection

We developed a structured electronic medical record database specifically for PMI patients at our hospital. This database was designed to standardize data collection and facilitate analysis. The collected data includes age, sex, body mass index (BMI), smoking and alcohol habits, family history of coronary artery disease (CAD), previous history, type of AMI; admission vital signs; laboratory examinations, CAG, and transthoracic echocardiography (TTE) parameters. TTE was conducted by certified sonographers following a standardized image acquisition protocol and utilizing consistent imaging equipment to capture the final result before discharge. Inflammatory biomarkers were determined using the following formulas, CLR (C-reactive Protein (mg/L) to Lymphocyte (*10⁹/L) ratio) and MLR (Monocyte (*10⁹/L) to Lymphocyte (*10⁹/L) ratio). The SYNTAX Score (20) was utilized to evaluate the severity of CAD and to aid in risk stratification and the planning of revascularization strategies among patients with CAD. It was calculated using software available online (https://syntaxscore.org/). In addition, we recorded the patient's medication status during hospitalization, including antiplatelet drugs, statins, diuretics, angiotensin-converting enzyme inhibitors (ACEI), angiotensin-receptor blockers (ARB), angiotensin receptor-neprilysin inhibitor (ARNI), and beta-blockers.

Study endpoint

All patients had a TTE assessment before discharged. HF was defined according to the criteria established by the guidelines for HF (21, 22). The diagnostic criteria for HFpEF are as follows:

(1) Presence of at least one typical heart failure symptom (e.g., exertional dyspnea, orthopnea) AND/OR one clinical sign of fluid retention (e.g., jugular venous distension, pulmonary rales, Leg edema).

(2) Objective evidence of cardiac congestion demonstrated by either:

(a) Hemodynamic confirmation via right heart catheterization [resting pulmonary capillary wedge pressure (PCWP)>15 mmHg],

(b) tructural/functional abnormalities through imaging: Thoracic radiography demonstrating pulmonary vascular redistribution or interstitial edema; Echocardiographic indices elevated E/e′ ratio (>13).

(3) Echocardiography indicates that the LVEF is ≥50%.

Model construction and validation

The minimum sample size required to construct the predictive model based on the pmsampsize criterion proposed by Riley et al. (23) requires the inclusion of at least 298 subjects. The missing rate of each variable was less than 5%; we used the median to fill in the missing values for continuous variables and conducted multiple imputations for categorical variables. The parameters for multiple imputation were set as m = 5, method = NULL, maxit = 50, and seed = 123. The number of missing values for each variable and the analysis results of the dataset before and after imputation are presented in Supplementary Table S1. To construct the model, the dataset is first randomly divided into: 80% as a training set for model training and cross-validation, and 20% as a testing set for final model evaluation. Within the training set, a five-fold cross-validation is further employed. The training data is randomly partitioned into five equal folds. For each iteration, four folds are selected for training, and one fold is used for validation. This process is repeated five times to optimize hyperparameters and reduce the risk of overfitting. The final model is evaluated using the 20% testing set to ensure a true assessment of its generalization performance. Considering the wide range of intervals for the different variables, as well as achieving the goal of rapid clinical application of the prediction system, we converted continuous variables other than age into categorical variables for analysis based on reference ranges, guidelines, or medians, and normalized age by a min-max scaling. Specifically, BMI was classified according to the Chinese population's overweight standard (24 kg/m²) (24). Heart rate (HR), systolic blood pressure (SBP), diastolic blood pressure (DBP), white blood cell count (WBC), neutrophils percentage (NEU%), hematocrit (HCT), hemoglobin (Hb), platelet count (PLT), total bilirubin (TBil), creatinine (Cr), uric acid (UA), homocysteine (HCY), glucose (Glu), glycated hemoglobin (HbA1c), total cholesterol (TC), triglycerides (TG), low-density lipoprotein cholesterol (LDL-C), high-density lipoprotein cholesterol (HDL-C), free fatty acids (FFA), C-reactive protein (CRP), D-dimer, fibrinogen (Fg), and brain natriuretic peptide (BNP) were classified according to their normal value reference ranges or guideline recommendations for classification. We classified total bile acids (TBA), CLR, MLR, creatine kinase isoenzyme (CK-MB), cardiac troponin T (cTnT), and SYNTAX scores according to the median of the dataset due to the lack of clear classification criteria. The number of HFpEF patients in this study is lower than that of non-HF patients. This imbalance in the data samples could significantly impact the performance of the prediction model. Therefore, we use the Synthetic Minority Over-sampling Technique (SMOTE) to balance the data set in the training set. The k-Nearest Neighbor (k-NN) threshold used in SMOTE is 5, and the over-sampling ratio is 1. This technique involves increasing the number of samples in the minority class by synthesizing new samples. Achieving a balanced number of samples in the data set is crucial for improving the model's training effectiveness (25).

Five ML models, Least absolute shrinkage and selection operator (Lasso)-Logistic, RF, XGBoost, KNN, and SVM, were constructed using the data from the training set with whether or not an in-hospital HFpEF occurred as the outcome event. Gradient boosting technology is based on decision trees. Variance inflation factor (VIF) was used to assess multicollinearity among predictor variables, and variables with VIF greater than 10 were excluded. Lasso is a regularization technique that performs variable selection and coefficient estimation by imposing constraints on the sum of the absolute values of the model parameters. This process causes some coefficients to be reduced to zero, effectively excluding them from the final model. Using the variables screened by Lasso, a two-category logistic risk prediction model is constructed. The modeling process of the other four models (RF, XGBoost, KNN, and SVM) is as follows. First, all variables are trained. To optimize the model's performance, the automatic parameter adjustment strategy is adopted, and a 5-fold cross-validation is used as the re-sampling strategy to evaluate the model's generalization ability. To further optimize model performance, we rank the variables based on their importance and select the top 10 features for retraining, following the same procedure as described above.

Statistical analysis

Shapiro–Wilk test was used to evaluate the normality of continuous variables. Mean ± standard deviation (SD) or median (interquartile) was used to describe variables according to the evaluation results. T-test or Mann–Whitney test was used for inter-group comparison. The categorical variables were expressed as numbers and percentages, and the chi-square test or Fisher's exact test was used for intergroup comparisons.

To evaluate the models, the area under the curve (AUC), Precision, Accuracy, F1 score, and Brier score were calculated in the testing set. Net Reclassification Improvement (NRI) and Integrated Discriminant Improvement (IDI) are used to assess the enhancement of the predictive model. According to the comparison results, the model with the best comprehensive performance was selected to construct the in-hospital HFpEF prediction model of PMI patients. The final model is explained using Shapley Additive exPlanations values (SHAP). Calibration curves determine how close the model's predicted probabilities are to the actual observed probabilities. Decision Curve Analysis (DCA) quantifies the net benefit of using the model at different thresholds and assesses the utility of the model in decision-making. Finally, a visual online prediction system was constructed to calculate the prediction probability for clinical application.

All statistical analyses were performed in SPSS Statistics 26.0 (IBM, Chicago, USA) and R software (version 4.3.1). Specifically, SPSS was used for descriptive statistics (normality tests, means, standard deviations and frequency distributions), and basic inferential analyses (t-tests, one-way ANOVA, and chi-square tests). R was employed for model construction, validation, performance evaluation and visualization analysis. A two-sided P < 0.05 was considered statistically significant.

Results

Baseline characteristics

The baseline characteristics of the study population, stratified into the HFpEF group (n = 268) and the Non-HF group (n = 572), are presented in Table 1. The total cohort comprised 840 participants, with a median age of 42[38–44] years and a pronounced male predominance (91.0% males vs. 9.0% females). Notably, significant differences were observed between the two groups. The HFpEF group had a higher proportion of females compared to the Non-HF group (15.7% vs. 5.9%, P < 0.001) and a slightly older median age (42 [39–45] vs. 41 [37–44] years, P = 0.001). Additionally, participants with HFpEF exhibited a higher prevalence of obesity (BMI ≥ 24 kg/m²: 82.5% vs. 74.5%, P = 0.013) and hypertension (56.3% vs. 45.8%, P = 0.006). Laboratory and clinical findings further distinguished the groups. The HFpEF group demonstrated elevated markers of inflammation (CRP >5.0 mg/L: 63.4% vs. 43.5%, P < 0.001; MLR >0.3: 65.7% vs. 42.7%, P < 0.001) and cardiac dysfunction (BNP >100 pg/ml: 91.0% vs. 62.4%, P < 0.001). Coronary angiography revealed more severe coronary artery disease in HFpEF patients, as evidenced by a higher SYNTAX Score >14.5 (64.2% vs. 43.0%, P < 0.001).

Table 1

Table 1. Baseline characteristics of the hFpEF group and Non-HF group.

In summary, the HFpEF cohort was characterized by older age, a higher burden of comorbidities, and more pronounced cardiovascular dysfunction compared to the Non-HF group.

Supplementary Table S2 shows the baseline characteristics of the training set and testing set. Supplementary Table S3 shows the descriptive statistics of continuous variables.

Model construction

According to the ratio of 8:2, 840 patients were randomly divided into a training set and a testing set, with 672 in the training set and 168 in the testing set, and five different models were constructed in the training set. The VIF values are all below 5, indicating that multicollinearity is not a significant issue in our model.

Lasso regression screening (Figures 2A,B) was performed on the full set of variables, identifying 12 key predictors: female sex, hypertension, STEMI, BMI ≥24 kg/m², SYNTAX Score >14.5, admission HR >75 bpm, NEU% >75%, HCT <45%, cTnT >1.44 ng/L, MLR >0.3, CLR >2.83, and BNP >100 pg/ml. Multivariate logistic regression modeling was carried out using the 12 variables from the screening. The final predictive model obtained consisted of 8 variables, which were female, BMI ≥24 kg/m², SYNTAX Score >14.5, admission HR >75 bpm, HCT <45%, MLR >0.3, CLR >2.83, and BNP >100 pg/ml.

Figure 2

Figure 2. (A) Lasso regression coefficients vs. λ values; (B) cross-validated λ and classification error plot relationships. Ranking of important variables based on Lasso-Logistic (C), XGBoost (D), RF (E), KNN (F), and SVM (G) models. BMI, body mass index; BNP, brain natriuretic peptide; CLR, C-reactive protein to lymphocyte ratio; cTnT, cardiac troponin T; Cr, creatinine; Fg, fibrinogen; Hb, hemoglobin; HbA1c, glycated hemoglobin; HCT, hematocrit; HDL-C, high-density lipoprotein cholesterol; HR, heart rate; MLR, monocyte to lymphocyte ratio; NEU%, neutrophil percentage; SBP, systolic blood pressure; TC, total cholesterol; UA, uric acid; WBC, white blood cell.

Following model training using RF, XGBoost, KNN, and SVM, the top 10 most important variables were identified and used for subsequent modeling. The ordering of important variables based on the Lasso-Logistic, XGBoost, RF, KNN, and SVM models are shown in Figures 2C–G, respectively.

Model performance

The classification performance of different models is compared in the testing set. (Table 2, Figures 3A,B). Regarding discrimination, the AUC values of the five models range from 0.517–0.854, with the highest being XGBoost and the lowest being SVM. The F1 score for combined precision and recall was the highest among the XGBoost models (0.586). A radar plot was made based on precision, accuracy, AUC, F1 score, and Brier score, which reflects the performance of each model (Figure 3B). Although the XGBoost model predicts a lower accuracy than Lasso-Logistic (0.686 vs. 0.762), its AUC, accuracy, and F1 score are the highest, and its Brier score is comparable to Lasso-Logistic (0.143 vs. 0.143). Supplementary Figure S1 illustrates the class distribution of the training set before and after applying the SMOTE. The figure demonstrates that SMOTE effectively addresses class imbalance by generating synthetic samples for the minority class, ensuring a more balanced distribution for model training. We noticed improvements in the performance of most models, with the exception of SVM. In particular, the F1 score of the combined precision and recall, and the AUC of the XGBoost and KNN models also improved. Even after balancing the classes with SMOTE, XGBoost remained the best-performing model (Supplementary Table S4). Supplementary Table S5 shows the NRI and IDI calculated for the XGBoost and four other models. The NRI calculated by the XGBoost model and Lasso-logistic was 0.149 (95% CI: 0.008–0.290, P = 0.039), and the IDI was 0.049 (95% CI: 0.008–0.091, P = 0.019). The NRI calculated by the XGBoost model and RF model was 0.222 (95% CI: 0.067–0.377, P = 0.005), and the IDI was 0.051 (95% CI: 0.015–0.086, p = 0.005). The NRI calculated by the XGBoost model and KNN model was 0.212 (95% CI: 0.028–0.396, P = 0.025), and the IDI was 0.067 (95% CI: 0.004–0.130, p = 0.038). The NRI calculated with the XGBoost model and the SVM model was 0.428 (95%CI: 0.277–0.579, P < 0.001), and the IDI was 0.227 (95% CI: 0.175–0.279, P < 0.001). These results indicate that the XGBoost model demonstrates stronger predictive ability compared to the other models.

Table 2

Table 2. Classification performance of five models on the testing set.

Figure 3

Figure 3. (A) ROC curve of five algorithms; (B) Radar plot of precision, accuracy, AUC, F1 score, and Brier score of the five models; Calibration curves of the XGBoost model in the training set (C) and testing set (D) DCA of the XGBoost model in the training set (E) and testing set (F).

Calibration curves show that the predicted probabilities of the XGBoost model are in good agreement with the actual probabilities (Figures 3C,D). The DCA plot (Figures 3E,F) demonstrates that the XGBoost model provides positive net benefits across a range of threshold probabilities in both the training and testing sets. This suggests that the model has strong clinical utility in aiding clinicians with accurate risk assessment and decision-making for PMI patients. In conclusion, the XGBoost model was selected as the final model based on its superior performance.

Importance ranking

The XGBoost model's variables' importance ranking showed that BNP >100 pg/ml is the most important characteristic influencing the occurrence of in-hospital HFpEF in patients with PMI. In addition, SYNTAX Score >14.5, Age, MLR >0.3, HCT <45%, HR >75 bpm, BMI ≥24 kg/m², CLR >2.83, Hypertension, and Fg ≥4 g/L were also important variables in predicting HFpEF (Figure 2D). Figure 4A and Supplementary Table S6 shows that the above variables were risk factors for in-hospital HFpEF in patients with PMI.

Figure 4

Figure 4. (A) Forest plot with 10 important variables based on the XGBoost model. (B) Visual online prediction system based on the XGBoost model. The input interface is on the left, and the output interface is on the right. The upper part of the right represents the prediction probability, and the lower part of the right shows an individualized SHAP diagram. The red feature makes the model recognize the sample as class 1, while the blue feature makes the model identify the sample as class 0. BMI, body mass index; BNP, brain natriuretic peptide; CLR, C-reactive protein to lymphocyte ratio; Fg, fibrinogen; HCT, hematocrit; HF, heart failure; HFpEF, heart failure with preserved ejection fraction; HR, heart rate; MLR, monocyte to lymphocyte ratio.

Visual prediction system

The visual prediction system includes an input interface (left) for entering patient variables (e.g., age, hypertension status) and an output interface (right) displaying predicted probabilities (top) and individualized SHAP charts explaining model decisions (bottom). In Figure 4B, the system predicts a 65.5% probability of HFpEF for a PMI patient based on clinical inputs. The system is publicly accessible at https://hfpefpmi.shinyapps.io/apppredict/.

Discussion

In this study, we explored five ML methods to build a prediction model for HFpEF in PMI patients based on clinically accessible data. The model developed by XGBoost has superior prediction performance and can better predict whether HFpEF occurs in the hospital in patients with PMI. Based on the ranking of variable importance, the final model includes ten predictive variables: BNP > 100 pg/ml, SYNTAX Score >14.5, Age, MLR >0.3, HCT <45%, HR >75 bpm, BMI ≥24 kg/m², CLR >2.83, Hypertension, and Fg >4 g/L. In addition, establishing a visual online prediction system will facilitate a more convenient clinical application, which can calculate the prediction probability of in-hospital HFpEF in PMI patients. As a model explanation, SHAP can show how each variable affects the model prediction and the degree of influence.

To our knowledge, this is the first study to develop an in-hospital HFpEF prediction model for PMI patients using ML algorithms. Previous studies, such as Liang et al. (26) and Xu et al. (27), primarily relied on logistic regression models to identify risk factors for HF in AMI patients. While logistic regression offers advantages in model interpretability and statistical rigor, it struggles to capture complex nonlinear interactions, potentially underestimating the influence of key variables. Given the highly heterogeneous pathophysiology of HFpEF, this study employs the XGBoost algorithm, which excels in identifying nonlinear relationships and intricate variable interactions, making it particularly suitable for HFpEF prediction. Our model incorporates multidimensional predictive factors, including SYNTAX score to assess coronary complexity, inflammatory markers (CLR, MLR), metabolic risk factors (BMI, hypertension), and dynamic biomarker thresholds (e.g., BNP >100 pg/ml), enabling more refined risk stratification. Unlike previous methods that primarily relied on static OR values, we implemented SHAP-based visualization analysis, which enhances model transparency and mitigates the “black-box” issue commonly associated with ML applications in clinical practice.

In recent years, Li et al. (28) demonstrated the potential of machine learning in predicting HF after AMI. Li et al. compared seven ML algorithms and identified XGBoost as the best-performing model (AUC = 0.966). However, their study primarily focused on general HF prediction rather than specifically addressing HFpEF as a distinct subtype. In contrast, our study is dedicated to HFpEF risk prediction, filling a critical gap in early risk stratification for PMI patients. Additionally, we developed an interpretable and real-time visualization tool, which can assist clinicians in rapidly assessing HFpEF risk in clinical practice. However, unlike Li et al. (28) which conducted external validation using an independent patient cohort, our study has not yet undergone multi-center external validation. Future research should focus on evaluating the model's stability and generalizability across different hospitals and populations, ensuring its clinical applicability and scalability.

The epidemiology of HF after AMI has undergone significant changes over the past decades, with an increased proportion of HFpEF and a mortality rate comparable to that of patients with Heart Failure with Reduced Ejection Fraction (HFrEF) (29). HFpEF is a heterogeneous systemic disease in which risk factors such as aging, obesity, hypertension, and systemic metabolic disorders may impair cardiac, pulmonary, vascular, and peripheral reserve capacity, abnormalities not apparent in the resting state (30). Damage and necrosis of cardiomyocytes after AMI can lead to impaired systolic reserve, and changes in cardiomyocyte and interstitial structure, such as cardiomyocyte hypertrophy and increased collagen fiber content, can lead to impaired diastolic reserve (31, 32).

Doshi et al. (33) studied the characteristics of HFpEF occurring during hospitalization in STEMI patients undergoing PCI and showed that older age, a higher proportion of females, and more comorbidities were risk factors for HFpEF. The SYNTAX Score, which reflects the extent of coronary artery disease, is also an essential factor in HFpEF. Complex coronary artery disease indicates the presence of a more significant atherosclerotic burden. The prevalence of HFpEF is strongly associated with metabolic syndrome epidemics, and obesity may be a significant driver of HFpEF in young people (34). The prevalence of subclinical left ventricular dysfunction has been confirmed to occur in Asians at lower BMI thresholds (35). In this study, HFpEF was predicted in young AMI patients with a BMI ≥24 kg/m². In addition to the traditional belief that hypertension leads to increased left ventricular afterload, causing left ventricular hypertrophy and subsequent left ventricular diastolic dysfunction, the pro-inflammatory state resulting from systemic hypertension also promotes the development of HFpEF (36). HFpEF is also manifested by elevated circulating inflammatory biomarkers, and the results of the present study suggest that inflammatory markers such as CLR, MLR, and Fg. On the other hand, the increase of Fg indicates increased clotting ability and impaired blood flow. Additionally, our study demonstrates a significant gender disparity, with a markedly higher prevalence of HFpEF among PMI patients compared to males (15.7% vs. 5.9%, p < 0.001). This finding is consistent with existing literature on sex-specific comorbidity burden and adverse outcomes following AMI in young women (37, 38). It highlights the need for early identification and gender-specific management strategies to improve outcomes in this vulnerable population.

Combined with the predictive factors discussed in this study, for high-risk patients, first-line treatment includes early use of sodium-glucose cotransporter type 2 (SGLT2) inhibitors for patients without contraindications and diuretics for patients with evident hyperemia to maintain average blood volume (2). Revascularization treatment should be carried out for patients with complex vascular lesions as soon as possible, and related risk factors should be controlled. Patients should insist on taking medicine, limit sodium, calorie, and fluid intake, control blood pressure, and monitor BNP and inflammatory markers.

In this study, we investigated for the first time the risk factors and characteristics of in-hospital HFpEF in young patients with AMI and developed a prediction model. With the prevalence of metabolic syndrome in young people, the incidence of HFpEF is increasing, and the prognosis is poor; more attention should be given to the young population. In addition, the XGBoost algorithm can enhance the performance of the prediction model, and there has not been a previous attempt to apply the ML algorithm to predict the risk of in-hospital HFpEF in patients with PMI. This study has several limitations. First, the single-center design may limit generalizability despite internal cross-validation. Second, although the sample size was sufficient for preliminary model development, a larger cohort is necessary to validate feature stability and improve model robustness. The limited sample size might affect model robustness. Third, future studies should incorporate additional clinical dimensions including lifestyle factors (e.g., diet, physical activity), medication histories, socioeconomic parameters, and genetic profiles to improve predictive performance. Finally, external validation across diverse populations remains essential before clinical implementation.

Conclusions

In this study, a new model was developed to predict the risk of in-hospital HFpEF in PMI patients. In a comprehensive comparison, the XGBoost model had the best predictive ability. The XGBoost-based visual prediction system shows clinical decision-support potential for PMI management, pending rigorous external validation in diverse clinical settings. Following successful validation, it could provide early and precise intervention guidance for PMI patients to reduce HFpEF incidence and improve long-term prognosis.

Data availability statement

The original contributions presented in the study are included in the article/Supplementary Material, further inquiries can be directed to the corresponding authors.

Ethics statement

The studies involving humans were approved by The Ethics Committee of Tianjin Chest Hospital (No. 2017KY-007-01). The studies were conducted in accordance with the local legislation and institutional requirements. The participants provided their written informed consent to participate in this study.

Author contributions

JW: Conceptualization, Data curation, Formal analysis, Investigation, Methodology, Software, Validation, Visualization, Writing – original draft, Writing – review & editing. CL: Conceptualization, Methodology, Supervision, Writing – review & editing. ZC: Methodology, Writing – review & editing. YLia: Data curation, Investigation, Writing – original draft. YW: Data curation, Investigation, Writing – original draft. YZ: Data curation, Visualization, Writing – original draft. YLiu: Methodology, Supervision, Writing – review & editing. JG: Conceptualization, Funding acquisition, Investigation, Methodology, Validation, Visualization, Writing – review & editing.

Funding

The author(s) declare that financial support was received for the research and/or publication of this article. This work was supported by the Key Project of Scientific and Technological Support Plan of Tianjin in 2020 [No. 20YFZCSY00820] and the Tianjin Health Research Project [No. TJWJ2022XK032].

Acknowledgments

The authors especially thank all the selected patients for their participation.

Conflict of interest

The authors declare that the research was conducted in the absence of any commercial or financial relationships that could be construed as a potential conflict of interest.

Generative AI statement

The author(s) declare that no Generative AI was used in the creation of this manuscript.

Publisher's note

All claims expressed in this article are solely those of the authors and do not necessarily represent those of their affiliated organizations, or those of the publisher, the editors and the reviewers. Any product that may be evaluated in this article, or claim that may be made by its manufacturer, is not guaranteed or endorsed by the publisher.

Supplementary material

The Supplementary Material for this article can be found online at: https://www.frontiersin.org/articles/10.3389/fcvm.2025.1571185/full#supplementary-material

References

1. Jenča D, Melenovský V, Stehlik J, Staněk V, Kettner J, Kautzner J, et al. Heart failure after myocardial infarction: incidence and predictors. ESC Heart Failure. (2021) 8:222–37. doi: 10.1002/ehf2.13144

PubMed Abstract | Crossref Full Text | Google Scholar

2. Redfield MM, Borlaug BA. Heart failure with preserved ejection fraction: a review. JAMA. (2023) 329:827–38. doi: 10.1001/jama.2023.2020

PubMed Abstract | Crossref Full Text | Google Scholar

3. Cunningham JW, Vaduganathan M, Claggett BL, John JE, Desai AS, Lewis EF, et al. Myocardial infarction in heart failure with preserved ejection fraction. JACC Heart Fail. (2020) 8:618–26. doi: 10.1016/j.jchf.2020.02.007

PubMed Abstract | Crossref Full Text | Google Scholar

4. Elgendy IY, Mahtta D, Pepine CJ. Medical therapy for heart failure caused by ischemic heart disease. Circ Res. (2019) 124:1520–35. doi: 10.1161/CIRCRESAHA.118.313568

PubMed Abstract | Crossref Full Text | Google Scholar

5. Popat A, Yadav S, Patel SK, Baddevolu S, Adusumilli S, Rao Dasari N, et al. Artificial intelligence in the early prediction of cardiogenic shock in acute heart failure or myocardial infarction patients: a systematic review and meta-analysis. Cureus. (2023) 15:e50395. doi: 10.7759/cureus.50395

PubMed Abstract | Crossref Full Text | Google Scholar

6. Kresoja K-P, Unterhuber M, Wachter R, Thiele H, Lurz P. A cardiologist’s guide to machine learning in cardiovascular disease prognosis prediction. Basic Res Cardiol. (2023) 118:10. doi: 10.1007/s00395-023-00982-7

PubMed Abstract | Crossref Full Text | Google Scholar

7. Chen T, Guestrin C. XGBoost: A Scalable Tree Boosting System. Edited by. ACM. 785–794 (2016).

Google Scholar

8. Wang X, Zhu T, Xia M, Liu Y, Wang Y, Wang X, et al. Predicting the prognosis of patients in the coronary care unit: a novel multi-category machine learning model using XGBoost. Front Cardiovasc Med. (2022) 9:764629. doi: 10.3389/fcvm.2022.764629

PubMed Abstract | Crossref Full Text | Google Scholar

9. Scrutinio D, Amitrano F, Guida P, Coccia A, Pagano G, D'addio G, et al. Prediction of mortality in heart failure by machine learning. Comparison with statistical modeling. Eur J Intern Med. (2025) 133:106–12. doi: 10.1016/j.ejim.2025.01.020

PubMed Abstract | Crossref Full Text | Google Scholar

10. Al Younis SM, Hadjileontiadis LJ, Khandoker AH, Stefanini C, Soulaidopoulos S, Arsenos P, et al. Prediction of heart failure patients with distinct left ventricular ejection fraction levels using circadian ECG features and machine learning. PLoS One. (2024) 19:e302639. doi: 10.1371/journal.pone.0302639

PubMed Abstract | Crossref Full Text | Google Scholar

11. Yan F-J, Chen X-H, Quan X-Q, Wang L-L, Wei X-Y, Zhu J-L. Development and validation of an interpretable machine learning model—predicting mild cognitive impairment in a high-risk stroke population. Front Aging Neurosci. (2023) 15:1180351. doi: 10.3389/fnagi.2023.1180351

PubMed Abstract | Crossref Full Text | Google Scholar

12. Liu J, Xu H, Chen Q, Zhang T, Sheng W, Huang Q, et al. Prediction of hematoma expansion in spontaneous intracerebral hemorrhage using support vector machine. EBioMedicine. (2019) 43:454–9. doi: 10.1016/j.ebiom.2019.04.040

PubMed Abstract | Crossref Full Text | Google Scholar

13. Golas SB, Shibahara T, Agboola S, Otaki H, Sato J, Nakae T, et al. A machine learning model to predict the risk of 30-day readmissions in patients with heart failure: a retrospective analysis of electronic medical records data. BMC Med Inform Decis Mak. (2018) 18:44. doi: 10.1186/s12911-018-0620-z

PubMed Abstract | Crossref Full Text | Google Scholar

14. Gupta S, Ko DT, Azizi P, Bouadjenek MR, Koh M, Chong A, et al. Evaluation of machine learning algorithms for predicting readmission after acute myocardial infarction using routinely collected clinical data. Can J Cardiol. (2020) 36:878–85. doi: 10.1016/j.cjca.2019.10.023

PubMed Abstract | Crossref Full Text | Google Scholar

15. Li J, Liu S, Hu Y, Zhu L, Mao Y, Liu J. Predicting mortality in intensive care unit patients with heart failure using an interpretable machine learning model: retrospective cohort study. J Med Internet Res. (2022) 24:e38082. doi: 10.2196/38082

PubMed Abstract | Crossref Full Text | Google Scholar

16. Zhang Y, Wang Y, Xu J, Zhu B, Chen X, Ding X, et al. Comparison of prediction models for acute kidney injury among patients with hepatobiliary malignancies based on XGBoost and LASSO-logistic algorithms. Int J Gen Med. (2021) 14:1325–35. doi: 10.2147/IJGM.S302795

PubMed Abstract | Crossref Full Text | Google Scholar

17. Yang T, Yang H, Liu Y, Liu X, Ding Y-J, Li R, et al. Postoperative delirium prediction after cardiac surgery using machine learning models. Comput Biol Med. (2024) 169:107818. doi: 10.1016/j.compbiomed.2023.107818

PubMed Abstract | Crossref Full Text | Google Scholar

18. Thygesen K, Alpert JS, Jaffe AS, Chaitman BR, Bax JJ, Morrow DA, et al. Fourth universal definition of myocardial infarction (2018). Eur Heart J. (2019) 40:237–69. doi: 10.1093/eurheartj/ehy462

PubMed Abstract | Crossref Full Text | Google Scholar

19. Thygesen K, Alpert JS, Jaffe AS, Simoons ML, Chaitman BR, White HD, et al. Third universal definition of myocardial infarction. J Am Coll Cardiol. (2012) 60:1581–98. doi: 10.1016/j.jacc.2012.08.001

PubMed Abstract | Crossref Full Text | Google Scholar

20. Sianos G, Morel MA, Kappetein AP, Morice MC, Colombo A, Dawkins K, et al. The SYNTAX score: an angiographic tool grading the complexity of coronary artery disease. EuroIntervention. (2005) 1:219–27. doi: 10.4244/EIJV1I2A36

PubMed Abstract | Crossref Full Text | Google Scholar

21. McDonagh TA, Metra M, Adamo M, Gardner RS, Baumbach A, Böhm M, et al. 2021 ESC guidelines for the diagnosis and treatment of acute and chronic heart failure. Eur Heart J. (2021) 42:3599–726. doi: 10.1093/eurheartj/ehab368

PubMed Abstract | Crossref Full Text | Google Scholar

22. Ponikowski P, Voors AA, Anker SD, Bueno H, Cleland JGF, Coats AJS, et al. 2016 ESC guidelines for the diagnosis and treatment of acute and chronic heart failure: the task force for the diagnosis and treatment of acute and chronic heart failure of the European Society of Cardiology (ESC) Developed with the special contribution of the heart failure association (HFA) of the ESC. Eur Heart J. (2016) 37:2129–200. doi: 10.1093/eurheartj/ehw128

PubMed Abstract | Crossref Full Text | Google Scholar

23. Riley RD, Snell KI, Ensor J, Burke DL, Harrell FE Jr, Moons KG, et al. Minimum sample size for developing a multivariable prediction model: PART II—binary and time-to-event outcomes. Stat Med. (2019) 38:1276–96. doi: 10.1002/sim.7992

PubMed Abstract | Crossref Full Text | Google Scholar

24. Zhou BF. Predictive values of body mass index and waist circumference for risk factors of certain related diseases in Chinese adults–study on optimal cut-off points of body mass index and waist circumference in Chinese adults. Biomed Environ Sci. (2002) 15:83–96.12046553

PubMed Abstract | Google Scholar

25. Chawla NV, Bowyer KW, Hall LO, Kegelmeyer WP. SMOTE: synthetic minority over-sampling technique. J Artif Intell Res. (2002) 16:321–57. doi: 10.1613/jair.953

Crossref Full Text | Google Scholar

26. Liang J, Zhang Z. Predictors of in-hospital heart failure in patients with acute anterior wall ST-segment elevation myocardial infarction. Int J Cardiol. (2023) 375:104–9. doi: 10.1016/j.ijcard.2023.01.002

PubMed Abstract | Crossref Full Text | Google Scholar

27. Xu M, Yan L, Xu J, Yang X, Jiang T. Predictors and prognosis for incident in-hospital heart failure in patients with preserved ejection fraction after first acute myocardial infarction. Medicine (Baltimore). (2018) 97:e11093. doi: 10.1097/MD.0000000000011093

PubMed Abstract | Crossref Full Text | Google Scholar

28. Li X, Shang C, Xu C, Wang Y, Xu J, Zhou Q. Development and comparison of machine learning-based models for predicting heart failure after acute myocardial infarction. BMC Med Inform Decis Mak. (2023) 23:165. doi: 10.1186/s12911-023-02240-1

PubMed Abstract | Crossref Full Text | Google Scholar

29. Gerber Y, Weston SA, Berardi C, McNallan SM, Jiang R, Redfield MM, et al. Contemporary trends in heart failure with reduced and preserved ejection fraction after myocardial infarction: a community study. Am J Epidemiol. (2013) 178:1272–80. doi: 10.1093/aje/kwt109

PubMed Abstract | Crossref Full Text | Google Scholar

30. Del Buono MG, Iannaccone G, Scacciavillani R, Carbone S, Camilli M, Niccoli G, et al. Heart failure with preserved ejection fraction diagnosis and treatment: an updated review of the evidence. Prog Cardiovasc Dis. (2020) 63:570–84. doi: 10.1016/j.pcad.2020.04.011

PubMed Abstract | Crossref Full Text | Google Scholar

31. Del Buono MG, Garmendia CM, Seropian IM, Gonzalez G, Berrocal DH, Biondi-Zoccai G, et al. Heart failure after ST-elevation myocardial infarction: beyond left ventricular adverse remodeling. Curr Probl Cardiol. (2023) 48:101215. doi: 10.1016/j.cpcardiol.2022.101215

PubMed Abstract | Crossref Full Text | Google Scholar

32. Antonelli L, Katz M, Bacal F, Makdisse MR, Correa AG, Pereira C, et al. Heart failure with preserved left ventricular ejection fraction in patients with acute myocardial infarction. Arq Bras Cardiol. (2015) 105:145–50. doi: 10.5935/abc.20150055

PubMed Abstract | Crossref Full Text | Google Scholar

33. Doshi R, Patel K, Gupta N, Gupta R, Meraj P. Characteristics and in-hospital outcomes of hospitalisations with heart failure with reduced or preserved ejection fraction undergoing percutaneous coronary intervention. Ir J Med Sci. (2019) 188:791–9. doi: 10.1007/s11845-018-1910-2

PubMed Abstract | Crossref Full Text | Google Scholar

34. Tromp J, MacDonald MR, Tay WT, Teng T-HK, Hung C-L, Narasimhan C, et al. Heart failure with preserved ejection fraction in the young. Circulation. (2018) 138:2763–73. doi: 10.1161/CIRCULATIONAHA.118.034720

PubMed Abstract | Crossref Full Text | Google Scholar

35. Lai YH, Liu ME, Su CH, Yun C-H, Liu C-Y, Hou CJ-Y, et al. Obesity-related changes in cardiac structure and function among Asian men and women. J Am Coll Cardiol. (2017) 69:2876–8. doi: 10.1016/j.jacc.2017.04.016

PubMed Abstract | Crossref Full Text | Google Scholar

36. Clemenza F, Citarrella R, Patti A, Rizzo M. Obesity and HFpEF. J Clin Med. (2022) 11:3858. doi: 10.3390/jcm11133858

PubMed Abstract | Crossref Full Text | Google Scholar

37. Sawano M, Lu Y, Caraballo C, Mahajan S, Dreyer R, Lichtman JH, et al. Sex difference in outcomes of acute myocardial infarction in young patients. J Am Coll Cardiol. (2023) 81:1797–806. doi: 10.1016/j.jacc.2023.03.383

PubMed Abstract | Crossref Full Text | Google Scholar

38. Izadnegahdar M, Singer J, Lee MK, Gao M, Thompson CR, Kopec J, et al. Do younger women fare worse? Sex differences in acute myocardial infarction hospitalization and early mortality rates over ten years. J Women’s Health. (2014) 23:10–7. doi: 10.1089/jwh.2013.4507

PubMed Abstract | Crossref Full Text | Google Scholar