Development and Validation of an Interpretable Prediction Model for Early Screening of Acute Myocardial Infarction in Young and Middle-Aged Patients

Development and Validation of an Interpretable Prediction Model for Early Screening of Acute Myocardial Infarction in Young and Middle-Aged Patients | Research Square window.SnipcartSettings = { analytics: { enabled: false } }; (function() { var accessVector = localStorage.getItem('access_vector') || ''; window.dataLayer = window.dataLayer || []; if (accessVector) { window.dataLayer.push({ user: { profile: { profileInfo: { snid: accessVector } } } }); } })(); (function(w,d,s,l,i){w[l]=w[l]||[];w[l].push({'gtm.start':new Date().getTime(),event:'gtm.js'});var f=d.getElementsByTagName(s)[0],j=d.createElement(s),dl=l!='dataLayer'?'&l='+l:'';j.async=true;j.src='https://www.googletagmanager.com/gtm.js?id='+i+dl;f.parentNode.insertBefore(j,f);})(window,document,'script','dataLayer','GTM-K279D39R'); Browse Preprints In Review Journals COVID-19 Preprints AJE Video Bytes Research Tools Research Promotion AJE Professional Editing AJE Rubriq About Preprint Platform In Review Editorial Policies Our Team Advisory Board Help Center Sign In Submit a Preprint Cite Share Download PDF Research Article Development and Validation of an Interpretable Prediction Model for Early Screening of Acute Myocardial Infarction in Young and Middle-Aged Patients QingQing Ruan, Shuzhi Su, Xian Wang, Xiumei Li, Zengyong Qiao, and 1 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-5614054/v1 This work is licensed under a CC BY 4.0 License Status: Under Revision Version 1 posted 4 You are reading this latest preprint version Abstract Background: In recent years, the incidence of acute myocardial infarction (AMI) has been rising among young individuals. However, existing research predominantly concentrates on AMI patients who are elderly. This study employs machine learning models to analyze multidimensional clinical features, with the objective of developing an accurate early screening tool for AMI in young and middle-aged populations. Methods: We analyzed data from 772 young and middle-aged patients who visited the Chest Pain Center at the South Campus of Shanghai Sixth People's Hospital between January 2018 and April 2024. This cohort included 640 patients diagnosed with AMI and 132 patients with non-AMI conditions. We optimized model parameters and evaluated the performance of eight machine learning algorithms. The SHAP (SHapley Additive exPlanations) method was employed to analyze feature importance and conduct feature screening to identify the optimal model. Additionally, we performed age-stratified SHAP analysis to investigate variations in feature importance across different age groups. Results: Among the eight machine learning models evaluated, the eXtreme Gradient Boosting (XGBoost) model exhibited the highest performance, achieving an AUC of 0.973. Utilizing the ranking of SHAP feature importance, a refined three-feature XGBoost model was developed, which demonstrated an improved AUC of 0.979. The final selected features included: the maximum emergency troponin value (Max cTnI), the maximum emergency BNP (Max BNP), and the duration from symptom onset to first medical treatment (SO-to-FMC). Subgroup analysis revealed variations in feature importance across different age groups. Conclusion: This study developed and validated a machine learning model using XGBoost for the early screening of AMI in young and middle-aged individuals, demonstrating high predictive accuracy and excellent interpretability, thereby making it suitable for diverse age cohorts within these populations. Acute Myocardial Infarction Interpretable Prediction Model Machine Learning XGBoost Early Screening Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Figure 9 Figure 10 Figure 11 Figure 12 1. INTRODUCTION Studies report that, in the past decade, there has been a near 15% rise in the incidence of AMI in the population aged between 30 to 50 years, with a tendency towards an increasingly younger age of onset [ 1 ] . This is attributed to a change in the style of living among youngsters, increased psychosocial stress, and genetic propensities [ 2 – 3 ] . The increase in cases among younger populations results in a tremendous escalation of medical expenditure and socio-economic burden due to cardiovascular disease. These patients suffer from chronic health problems, which, apart from personal suffering, further enhance the medical resources needed. Though much attention has been directed to acute myocardial infarction in young and middle-aged individuals in recent years, research in the past on myocardial infarction has focused on older patients, and relatively little attention has been paid so far to determining the characteristics and risk factors associated with AMI in younger patients. Consequently, studies on its early screening and diagnosis are still less well developed today. Due to the increase of the incidence in relatively younger populations, the need for an accurate early-screening tool is urgent in AMI. Moreover, an early-screening tool for AMI should be easy to operate for users, accessible, and provide identification of not just high-risk individuals but assistance as early as possible. Conventional screening modalities include ECG and biochemical marker detection. Conventional means have shown suboptimal sensitivity and specificity in diagnosing early AMI [ 8 – 10 ] . It is therefore likely that interest in newer methodologies may be necessary to better delineate this disease. A new approach is needed to transcend the shortcomings of the traditional screening technique to raise the early detection rate of AMI among young and middle-aged populations. The application of machine learning has become widely regarded, with its strong data analytics and predictive powers, as one of the key contributors to improving the effectiveness of early screening in relation to the prediction of AMI. Various studies have shown that the ML algorithms may combine the patient age and sex along with cardiac troponin, hs-cTnI, levels in a diagnostic tool-the Myocardial Ischemic Injury Index, MI3-so as to furnish a rather more personalized estimate of acute myocardial infarction risk than that obtainable by traditional means [ 14 – 15 ] . Some researchers also employed deep learning algorithms to develop a semi-automated system capable of classifying the ECG signals into normal, coronary heart disease, and myocardial infarction categories for early diagnosis and alerts to clinicians [ 16 ] . The construction of machine learning-based predictive models allows the researchers to proactively identify the high-risk patients [ 17 – 18 ] . ML enhances early detection of myocardial infarction by increasing diagnostic precision, reducing response time, managing patients personally, and offering enhanced decision support [ 19 – 21 ] . This study evaluates the application of ML in early screening and diagnosis of AMI in young and middle-aged patients. An interpretable model was developed and tested for predicting AMI in this population by integrating multidimensional variables with age-stratified analyses. The novelty of this study is that the feature selection method and optimization model allow new diagnosis perspectives for different age groups and thus increase the accuracy of early screening for AMI in young and middle-aged populations. Besides, it may improve clinical outcomes for such patients and so lay a good foundation for future clinical application. 2. METHODS 2.1 Study Population and Study Design The present study was a retrospective one conducted from the Chest Pain Center of the South Campus of Shanghai Sixth People's Hospital. The inclusion criteria of study subjects were in line with the diagnostic definition of AMI according to current guidelines, including patients who met the criteria of STEMI and NSTEMI. Controls were patients attending the chest pain center for chest discomfort in the same period but without a diagnosis of AMI. All subjects had to be aged between 18 and 59 years. For the final sample size, which included an analysis, a total of 772 individuals were studied: 640 patients who comprised the AMI group and 132 who formed the non-AMI control group.The subjects were additionally categorized into a young group (≤44 years old, n=213) and a middle-aged group (45-59 years old, n=559) based on age. This study adhered strictly to the ethical guidelines of the Declaration of Helsinki and received approval from the hospital ethics committee (approval number: 2024-KY-26-02). Given that this study solely involved the collection of relevant data without interfering with routine diagnostic and treatment activities, informed consent was not required. Figure 1 illustrates the design concept of this study, which encompasses three steps: data preprocessing, model development, and subgroup analysis. Figure 1: Flow chart of the study design. SMOTE: Synthetic Minority Over-sampling Technique; ML:machine learning; XGBoost: eXtreme Gradient Boosting; SHAP: SHapley Additive explanation. 2.2 Data Collection In this study, we collected a total of 30 variables from patients' electronic medical records in the emergency department. These variables encompassed patient demographic information, medical history, electrocardiographic characteristics, echocardiographic features, and initial laboratory parameters. The variables included: gender, age at onset, time from symptom onset to first medical consultation (SO-to-FMC), weight (kg), height (cm), body mass index (BMI), systolic blood pressure on admission, diastolic blood pressure on admission, heart rate upon admission, cardiac function class, history of hypertension, hyperlipidemia, diabetes, smoking history, troponin levels upon admission, highest troponin I (cTnI) recorded in the emergency room, serum creatinine on admission, D-dimer on admission, B-type natriuretic peptide (BNP) on admission, myoglobin on admission, creatine kinase MB (CKMB) on admission, highest BNP value in the emergency department, and left ventricular ejection fraction (LVEF%) recorded as the lowest value in the emergency department. Additionally, we noted the history of coronary heart disease, revascularization, cerebrovascular disease, peripheral vascular disease, chronic kidney disease, chronic heart failure, and atrial fibrillation. 2.3 Data Processing The data integrity of this study is high, with missing data variables not exceeding 15% of the sample size, thus eliminating the need to delete any variables. To address missing data, various strategies—including random filling, mean filling, and mode filling—were employed to preserve the shape of the variable distribution and ensure prediction accuracy. Outliers were addressed through random, mean, and mode replacement methods, in accordance with clinical recommendations. Following the conversion of textual non-numeric data into numerical variables, the dataset was divided into a training set and a test set in an 80:20 ratio. The training set utilized SMOTE(Synthetic Minority Over-sampling Technique) oversampling to mitigate the class imbalance problem, while the test set was reserved for evaluating model performance. 2.4 Data Imbalance Processing The dataset was divided into a training set and a test set in an 80:20 ratio. The training set for this study comprised 617 data samples ( Figure 2A), including 513 cases of AMI and 104 cases of non-AMI, which illustrates a significant data imbalance. This imbalance may lead the model to exhibit a stronger learning effect on the majority class samples while inadequately learning from the minority class samples, ultimately impacting the overall performance of the model. To address this issue, we employed SMOTE for oversampling. SMOTE generates synthetic samples by selecting the nearest neighbors of minority class samples, thereby augmenting the number of minority class samples, enhancing the model's generalization ability towards these classes, and alleviating overfitting issues. As depicted in Figure 2B, following SMOTE processing, the training set was expanded to 1026 samples, comprising 513 cases of acute myocardial infarction and 513 cases of non-acute myocardial infarction, achieving class balance. Figure 2: Distribution of categories within the training set for young and middle-aged groups. Figure 2A displays the original categories before the utilization of SMOTE, while Figure 2B displays the categories after the utilization of SMOTE. In this case, '1' stands for acute myocardial infarction and '0' for non-acute myocardial infarction. 2.5 Data Set Standardization In order to reduce the effect of different dimensions and numerical ranges on data analysis, model training, and interpretation of results, this study uses Z-score normalization for features of the dataset so that all features contribute equally during model training for stability and generalization ability. Transformed features in the distribution with mean 0 and standard deviation 1 achieve comparability on different scales. This method not only will make the learning relationships between features more effective but also enhance the prediction accuracy of the model, especially in clinical applications, to classify acute myocardial infarction from non-acute myocardial infarction with greater precision. This very same process of standardization was then performed for the test set features to ensure that measures were consistent and hence the validity and reliability of the model could be warranted when applied in the real world. 2.6 Feature Selection We screened potential candidate variables through preliminary univariate analysis, establishing a significance level (p-value) of less than 0.05 as the criterion for inclusion. Variables meeting this criterion were included in the candidate variable set. XGBoost was identified as the optimal model through training and testing, and the SHAP interpreter was employed to conduct feature importance analysis, ranking all features from high to low importance and examining the degree of interaction between them. Features were selected based on clinical considerations. 2.7 Model Development and Verification After applying the SMOTE technology, the number of samples in the training set was extended to 1,026. In this paper, eight popular machine learning models were trained: Random Forest, K-Nearest Neighbors, Support Vector Machine, Multi-Layer Perceptron, Gradient Boosting, Decision Tree, Extreme Gradient Boosting, and Logistic Regression. The optimal parameters for each model were searched by using the grid search method along with 5-fold cross-validation. After defining the best parameters, each model's performance was measured on the test set with accuracy, precision, recall, F1 score, AUC, and confusion matrix. The best performance gave the chosen model for the basis of further research and utilized 30 features for forecasting acute myocardial infarction. This systematic approach will ensure that the proposed model is optimum and clinically applicable. 2.8 Feature Importance Analysis After the best model was determined, we conducted a feature importance analysis so that more insight into the contribution of each feature on the model's prediction result could be gathered. We provided both global and local explanations of each model prediction using the SHAP method. Computing the Shapley values of the different input features gave us a ranking of feature importances relative to the entire model predictions. By using global explanations, we could identify which features had the greatest impacts on the predictions of the model, and local explanations showed why the model made a certain decision. This systematic approach enhances model performance, bolstering interpretability and credibility for clinical applications. 3. Results 3.1 Population Characteristics We analyzed data from 772 young and middle-aged patients who visited the Chest Pain Center at the South Campus of Shanghai Sixth People's Hospital between January 2018 and April 2024. This cohort included 640 patients diagnosed with AMI and 132 without AMI, with 213 classified as young (age ≤ 44 years) and 559 as middle-aged (ages 45 to 59 years). The study aimed to establish an AMI prediction model. Through preliminary univariate analysis, we identified variables with a significance level (p-value) of less than 0.05, which were subsequently included in the candidate variable set for model construction.The following characteristics demonstrated statistical significance (P < 0.05): SO-to-FMC time (P = 0.004), weight (P = 0.021), height (P = 0.003), troPonin on admission (P < 0.001), maximum troPonin I value (P < 0.001), serum creatinine on admission (P = 0.004), myoglobin on admission (P < 0.001), creatine kinase MB on admission (P < 0.001), maximum Brain natriuretic PePtide (P < 0.001), gender (P < 0.001), smoking history (P < 0.001), history of coronary heart disease (P < 0.001), history of coronary revascularization (P < 0.001), history of cerebrovascular disease (P < 0.001), history of chronic kidney disease (P < 0.001), history of heart failure (P < 0.001), and history of atrial fibrillation (P < 0.001) (Table 1). These characteristics will be given considerable attention during the subsequent model building and optimization processes to enhance the performance and accuracy of the predictive model. Table 1 Baseline characteristics of patients Variable N Overall, N = 772 0, N = 132 1, N = 640 p-value Age at onset, mean (sd) 772 48.97 (7.56) 50.01 (7.42) 48.76 (7.58) 0.081 SO-to-FMC, mean (sd) 772 952.92 (2,157.24) 1,710.12 (3,542.02) 796.75 (1,702.97) 0.004 Body Weight(kg), mean (sd) 772 75.43 (13.49) 73.02 (12.97) 75.93 (13.56) 0.021 Height(cm), mean (sd) 772 169.73 (5.90) 168.24 (6.20) 170.03 (5.80) 0.003 BMI, mean (sd) 772 26.15 (4.35) 25.78 (4.44) 26.22 (4.33) 0.298 Adm SBP, mean (sd) 772 147.57 (28.98) 150.17 (27.22) 147.04 (29.33) 0.236 Adm DBP, mean (sd) 772 90.82 (18.97) 89.71 (17.00) 91.05 (19.36) 0.423 Adm HR, mean (sd) 772 80.40 (18.72) 78.63 (14.75) 80.77 (19.43) 0.154 Adm Tn, mean (sd) 772 104.76 (341.27) 9.64 (26.85) 124.37 (371.65) <0.001 Max cTnI, mean (sd) 772 55.84 (60.27) 0.73 (2.62) 67.21 (60.20) <0.001 Adm SCr, mean (sd) 772 82.36 (69.73) 73.47 (25.49) 84.19 (75.59) 0.004 Adm D-dimer, mean (sd) 772 0.96 (2.88) 0.76 (0.65) 1.01 (3.15) 0.069 Adm BNP, mean (sd) 772 88.22 (285.20) 93.38 (464.02) 87.15 (232.37) 0.881 Adm Myo, mean (sd) 772 215.89 (301.79) 63.90 (121.25) 247.24 (317.99) <0.001 Adm CKMB, mean (sd) 772 23.53 (49.08) 3.01 (6.37) 27.76 (52.85) <0.001 Max BNP, mean (sd) 772 418.38 (1,652.53) 99.95 (311.23) 484.06 (1,802.73) <0.001 Min LVEF%, mean (sd) 772 59.61 (9.00) 60.76 (7.86) 59.37 (9.20) 0.075 Sex, n (p%) 772 <0.001 0 702.00 (90.93%) 109.00 (82.58%) 593.00 (92.66%) 1 70.00 (9.07%) 23.00 (17.42%) 47.00 (7.34%) NYHA, n (p%) 772 0.381 1 733.00 (94.95%) 129.00 (97.73%) 604.00 (94.38%) 2 16.00 (2.07%) 2.00 (1.52%) 14.00 (2.19%) 3 4.00 (0.52%) 0.00 (0.00%) 4.00 (0.63%) 4 19.00 (2.46%) 1.00 (0.76%) 18.00 (2.81%) Hx of HTN, n (p%) 772 0.986 0 365.00 (47.28%) 63.00 (47.73%) 302.00 (47.19%) 1 407.00 (52.72%) 69.00 (52.27%) 338.00 (52.81%) Hx of HLD, n (p%) 772 0.446 0 436.00 (56.48%) 79.00 (59.85%) 357.00 (55.78%) 1 336.00 (43.52%) 53.00 (40.15%) 283.00 (44.22%) Hx of DM, n (p%) 772 0.373 0 582.00 (75.39%) 95.00 (71.97%) 487.00 (76.09%) 1 190.00 (24.61%) 37.00 (28.03%) 153.00 (23.91%) Smk Hx, n (p%) 772 <0.001 0 184.00 (23.83%) 67.00 (50.76%) 117.00 (18.28%) 1 588.00 (76.17%) 65.00 (49.24%) 523.00 (81.72%) Hx of CAD, n (p%) 772 <0.001 0 707.00 (91.58%) 100.00 (75.76%) 607.00 (94.84%) 1 65.00 (8.42%) 32.00 (24.24%) 33.00 (5.16%) Hx of Revasc for CAD, n (p%) 772 <0.001 0 729.00 (94.43%) 108.00 (81.82%) 621.00 (97.03%) 1 43.00 (5.57%) 24.00 (18.18%) 19.00 (2.97%) Hx of CVD, n (p%) 772 <0.001 0 734.00 (95.08%) 111.00 (84.09%) 623.00 (97.34%) 1 38.00 (4.92%) 21.00 (15.91%) 17.00 (2.66%) Hx of PVD, n (p%) 772 0.840 0 500.00 (64.77%) 87.00 (65.91%) 413.00 (64.53%) 1 272.00 (35.23%) 45.00 (34.09%) 227.00 (35.47%) Hx of CKD, n (p%) 772 <0.001 0 722.00 (93.52%) 112.00 (84.85%) 610.00 (95.31%) 1 50.00 (6.48%) 20.00 (15.15%) 30.00 (4.69%) Hx of CHF, n (p%) 772 <0.001 0 750.00 (97.15%) 117.00 (88.64%) 633.00 (98.91%) 1 22.00 (2.85%) 15.00 (11.36%) 7.00 (1.09%) Hx of AF, n (p%) 772 <0.001 0 747.00 (96.76%) 115.00 (87.12%) 632.00 (98.75%) 1 25.00 (3.24%) 17.00 (12.88%) 8.00 (1.25%) (1: Indicates the presence of a relevant medical history or condition. 0: Signifies the absence of such a history or condition.) 3.2 Model Construction and Evaluation This article tries to model and compare the performances of eight machine learning algorithms on a dataset of young and middle-aged patients presenting with chest pain to determine the best model. Regarding this, the eight machine learning algorithms employed in this research are RF, KNN, SVM, MLP, GB, DT, XGBoost, and LR. We used a combined approach of grid search and 5-fold cross-validation within the training set as a means of optimizing the parameters of all the algorithms by systematically changing model parameters such that optimum performance may be yielded. Having finished training, we tested the performance of our optimized models on the test set to see their accuracy and reliability in processing chest pain data.Curves AUC-ROC were constructed, calculating the AUC as quantification of the predictive power of each of the models. This is shown in Figure 3. We then further presented, with the use of heat maps, the results of test set classification by each model and drew a confusion matrix for each model (Figure 4). From the confusion matrix we calculated for each model the accuracy, precision, recall, and F1 score in order to enable a detailed evaluation of their performance (Table 2). The result of this wide review thus suggests that the best performance is from the XGBoost algorithm, and as such, this is the model which forms the basis for further research. More detailed performance measures in each category of the XGBoost model are shown on the test set in Table 3. Figure 3 .A comparison of ROC curves for various ML algorithms applied to the test set. Figure 3 compares ROC curves of various ML algorithms in the test set. The horizontal axis shows the False Positive Rate, and the vertical axis shows the True Positive Rate. A comparison of all the above algorithms reveals the best diagnostic models with respect to the prediction of chest pain, represented by XGBoost with 0.973 AUC and GB with 0.968 AUC. Whereas RF, DT, and SVC can be referred to as the strong predictors with 0.965, 0.949, and 0.946 AUC, respectively. Poor results were received for KNN with the AUC of only 0.700. The larger the AUC, the stronger will be the predictive ability of every algorithm. （ rf ： Random Forest,knn:K-Nearest Neighbors,svc:Support Vector Classification,mlp:Multi-Layer Perceptron,gb:Gradient Boosting,dt:Decision Tree,xgb:eXtreme Gradient Boosting,lr ： Logistic Regression ） Figure 4: Confusion matrix heatmap of various models. This figure depicts the respective performances of each model on a test set that has exclusively been used for the classification of chest pain data. The columns in the confusion matrix represent the true labels (True Label) of the data, whereas the rows give information about the labels as predicted by the algorithm, Predicted Label. In these confusion matrices, the color intensity represents the frequency of the prediction result. The darker it is, the higher the count. These confusion matrices can visually check the accuracy and error rate of each algorithm in classifying both positive and negative samples. (1: Acute myocardial infarction, 0: Non-acute myocardial infarction). Table 2.Performance Evaluation of Various Machine Learning Models on the Dataset of Young and Middle-aged Myocardial Infarction Patients. Algorithm Model Evaluation Metrics Accuracy Precision Recall F1-Score AUC RF 0.942 0.90 0.91 0.90 0.965 KNN 0.690 0.62 0.70 0.62 0.700 SVC 0.929 0.87 0.90 0.88 0.946 MLP 0.890 0.81 0.85 0.83 0.938 GB 0.942 0.89 0.92 0.91 0.968 DT 0.935 0.87 0.93 0.90 0.949 XGBoost 0.948 0.91 0.91 0.91 0.973 LR 0.916 0.85 0.88 0.86 0.942 Table 3: Specific Performance Metrics of the XGBoost Model for Each Class on the Test Set. Precision Recall F1_score Support 0 0.86 0.86 0.86 28 1 0.97 0.97 0.97 127 Accuracy 0.95 155 Macro avg 0.91 0.91 0.91 155 Weighted avg 0.95 0.95 0.95 155 (1: acute myocardial infarction, 0: non-acute myocardial infarction). 3.3 Feature Importance Analysis Using the optimized XGBoost model, we built a SHAP interpreter for carrying out feature importance analysis, as shown in Figure 5. SHAP values quantify how much each feature contributes to the model's predictions and thus allow us to identify the top 20 features which may have the most influence on the prediction outcomes for the MI patients. This ranking not only indicates the important factors in the decision of the model but also lays the foundation for identifying significant biomarkers in clinical practice. To further study the specific contribution of each feature to the prediction result of each sample, a feature heatmap was plotted (Figure 6). Figure 5. SHAP value analysis of feature importance rank for Top 20 Features in the XGBoost model Figure 5A shows the bar plot of the mean SHAP values of the top 20 features. The bars in blue color represent the average SHAP value of each feature, ranking their relative importance to the model prediction output. The three most critical features, according to the chart, are "Max cTnI"-maximum cardiac troponin I concentration, "Max BNP"-maximum brain natriuretic peptide concentration, and "SO-to-FMC"-the time from symptom onset to first medical contact. Other important ones include CK-MB creatine kinase-MB at admission, Smk Hx-smoking history, body mass index, and Adm SCr-creatinine level at admission. Figure 5B: SHAP values of the top 20 features are presented, where on the right side in this honeycomb diagram presents a specific value of each feature in impacting the prediction output for a given sample.The blue-to-red color gradient shows the range of variation in the value of the features. A high SHAP value indicates that this feature is positively contributing to the model's prediction of a high risk. Low values mean otherwise. For instance, high values of "Max cTnI" and "Max BNP" are usually associated with a high risk from myocardial infarction. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate） Figure 6. Heatmap of Features for Early Screening Models in Young and Middle-aged Myocardial Infarction Patients. Figure 6 illustrates the SHAP value heat map for the predicted model features, providing a detailed explanation for each sample by quantifying the specific contribution of individual features to the model's prediction results. The horizontal axis represents the test set samples, while the left vertical axis indicates the importance ranking of features, sorted from highest to lowest influence. The right vertical axis visualizes feature influence, with color depth reflecting the magnitude of SHAP values; darker colors correspond to larger absolute SHAP values, indicating more significant feature impact on model predictions. The top area visualizes the model's prediction results based on these feature values. This analysis corroborates that Max cTnI, Max BNP, and SO-to-FMC are the three most influential features in the model. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage） 3.4 Interaction Between Features and Key Feature Analysis To further explore feature-to-feature interactions, we computed the interaction values of SHAP. Computing the mean absolute values for every paired value of feature-to-feature interactions and plotting these results in a heat map (Figure 7A) can help us identify both the magnitude and the patterns of feature interactions very well. First, to give an intuition of the interactive relationships among the primary features, we drew an interaction summary diagram for the first seven features. Figure 7B: There are significant interactions among some combinations of features for predicting AMI, such as interaction between Max cTnI and Adm BNP. A dependence plot was constructed to further investigate the interaction of features for Max cTnI and Adm BNP (Figure 8). Finally, to understand the impact of the contribution of each feature on model output, a SHAP force diagram was used (Figure 9). Figure 7. Heatmap of Feature Interaction Values and Major Feature Interactions. Figure 7A presents a heat map illustrating the interaction values among all features, where the color of each grid indicates the absolute average value of the interaction between corresponding feature pairs, with values ranging from 0 to 0.15. The depth of color in the figure corresponds to the strength of the interaction between features; darker colors signify more substantial interaction effects. This heat map reveals significant interactions among certain feature combinations in predicting AMI, notably between Max cTnI and Adm BNP. Figure 7B illustrates the interactions among the first seven features. These characteristics include: Max cTnI (maximum troponin I level), Max BNP (maximum B-type natriuretic peptide level), SO-to-FMC (time between the onset of symptoms and first medical contact), Adm CKMB (admission muscle acid kinase isoenzyme level), Smk Hx (smoking history), BMI (body mass index), and Adm SCr (serum creatinine level upon admission). Each site in the figure represents the influence of a single feature on the model output, while the color coding reflects the value of a specific feature. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine） Figure 8. SHAP Dependence Plot Demonstrating Interaction of Max cTnI and Adm BNP. Figure 8A provides an overview of the general trend in the SHAP value for Max cTnI in relation to its concentration, in an attempt to develop a sense about the impact that this variable can have independently on the model prediction result. From the figure, one can clearly see that as the values of Max cTnI go up, the SHAP value continues increasing and shows its critical role regarding myocardial infarction prediction. The impact of Adm BNP is represented through the color gradient of Figure 8B, whose value regulates the Max cTnI SHAP. The proximity to red indicates higher values for Adm BNP, while the cooler colors reflect the low values. High Adm BNP values are associated with high SHAP values for Max cTnI; this would mean that Adm BNP increased the explanatory strength of Max cTnI in the model. Figure 8C: SHAP main effect value of Max cTnI. The plot here strengthens the position of Max cTnI as an important predictor, since from the results, it can be observed that above the threshold value of Max cTnI, the SHAP main effect value is high, showing its importance to predict a patient prognosis at an elevated level. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide) Figure 9: SHAP force plot displaying the feature contributions to model output. Each feature contribution can be seen using a red color, which shows the positive contribution, and blue represents the negative contribution. The length of each bar is proportional to the magnitude of the contribution of the respective feature to the final predicted value. Figure 9A shows the average strength of a feature's influence across the test set as a whole and serves well in depicting, on average, the relative importance of various features to the model output. The prominent features are Max BNP and Adm CKMB among others. Figure 9B emphasizes that features such as Max BNP and Adm CKMB have a great impact, whereas Figure 9C shows the contributions of other features like BMI and Adm Myo, which have negative values in the model. Decomposition further reinforces our presented analysis of feature importance for showing the different role played by the features in various sample categories. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Adm Myo：Admission Myoglobin） 3.5 Feature Reduction and Model Optimization To reduce the bad effect of redundant features on the classification results and further apply the model easier in clinical practice in the future, feature reduction is conducted. Based on the SHAP analysis explained above, we got an importance ranking of 30 input features, thus carrying out the feature selection. We recorded and visualized the AUC values of the XGBoost model with optimized parameters running on varying numbers of features. This is shown in Figure 10(A). It can be observed from the analysis that the optimal classification performance of the XGBoost classifier can be achieved using the three most important features: emergency maximum troponin I [Max cTnI], emergency maximum BNP [Max BNP], and the time from symptom onset to first medical care [SO-to-FMC]. Based on these three features, we decided on this XGBoost model as our final model after consultation with clinicians. We plot the ROC on the test set for this final model, whose AUC is 0.979, shown in Figure 10(B). In addition, detailed performance metrics of the final model on the test set are given in Table 4, which further confirm the excellent predictive performance of this model. Figure 10: Analysis of feature parsimony versus model performance. (A) Line graph of feature parsimony illustrating AUC values of the XGBoost model with variation in a number of selected features. The vertical axis is for AUC value while the horizontal axis is for the number of selected features. This graph shows how feature selection affects the performance of the model. (B) ROC Curve of XGBoost Model Based on Three Most Important Features.Showing Classification Performance of Model on Test Set. The shape of this curve and the corresponding AUC values below provide a critical basis for evaluating its accuracy. TABLE 4: Classification Performance Metrics of the Final XGBoost Model Precision Recall F1_score Support 0 0.81 0.89 0.85 28 1 0.98 0.95 0.96 127 Accuracy 0.94 155 Macro avg 0.89 0.92 0.91 155 Weighted avg 0.95 0.94 0.94 155 3.6 Subgroup Analysis We conducted a subgroup analysis in order to review and study the effects that age has on different groups as it comes to perceived feature importance. The dataset includes both young and middle-aged patients, divided into two groups. The young group consists of ≤ 44-year-olds, which accounts for 213 cases (185 myocardial infarction, 28 no myocardial infarction), while the middle-aged group consists of people aged between 45 to 59 years old, totaling 559 cases (455 myocardial infarction, 104 no myocardial infarction). We hypothesize that the importance of features in early screening for AMI may vary with age. To confirm that, we tested the XGBoost model with 30 features of young and middle-aged groups. The ROC curve performance for this model in various age groups is given in Figure 11. Figure 12 presents feature importance ranks of young and middle-aged groups based on SHAP value evaluations and their contribution to model output. Afterwards, we compared the average SHAP values along with the distribution of each feature across the two age groups to understand the relative contributions of different features in predicting myocardial infarction. Figure 11.ROC Curve Performance of the XGBoost Model Across Different Age Groups. Figure 11 illustrates the ROC curve obtained from the XGBoost model with 30 features in classifying the young group (Figure A) and the middle-aged group (Figure B). The ROC curves of the two models are also almost perfect, reaching an AUC close to 1; thus, the AUCs of the young and middle-aged groups are 0.993 and 0.996, respectively. Besides, these results indicate that the XGBoost model with 30 features can make pretty good predictions about myocardial infarction for different age groups and estimate feature importance corresponding to individuals from different age brackets quite correctly. So, this model should be a good starting point for developing a SHAP Interpreter in order to check differences in feature importance arising among diverse age groups. Figure 12: Feature importance distribution across age groups based on SHAP values. Figure 12 presents the ranking of feature importance for youth and middle-aged groups derived from an SHAP value evaluation and its contribution toward model output. A comparison of average SHAP values and distribution of various features in both age groups brings out the relative contributions of various features in predicting myocardial infarction. Parts A and C show the average value of each feature's contribution to explain the output of the model, while B and D depict the distribution of the SHAP values, where color is used to emphasize the direction and magnitude of feature contributions with regard to a prediction. Max cTnI, Max BNP, admission CKMB, and SO-to-FMC were significant contributors in both age cohorts. However, other predictors displayed differential effects with age, especially for BMI, whose impact on the prediction outcomes was stronger in the middle-aged group compared to youth. （Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate） DISCUSSION In recent decades, the incidence of AMI has significantly risen among young people. It is now an important concern in terms of worldwide public health compared to earlier decades [22-24] . While typically perceived to occur in older adults, cases among younger patients are strikingly increasing. Over 30,000 young women under the age of 55 years are now admitted every year in the United States, which reveals how serious and urgent this condition needs to be taken as [22] . The increase in the incidence of AMI in younger populations is related mainly to the changes in the lifestyle and increase in the traditional risk factors, such as metabolic syndrome and diabetes, while alternative risk factors are gaining increasing recognition as major determinants of cardiovascular health in young people. This trend puts not only the health of younger subjects at risk but also increases the medical and economic burden on society. In view of an increasing incidence of AMI in patients of young age group, most of the studies relating to AMI have been conducted in elderly populations, and this has resulted in a lack of comprehensive understanding regarding characteristics, causes, and needs of the young patients with AMI [3] . Given the growing trends of AMI in younger age groups and limitations identified among previous studies, this represents a pressing priority to improve research into younger age groups in order to bridge the knowledge gap. The early development of AMI screening tools will be of importance in relation to young and middle-aged individuals, as this will help the healthcare provider understand who is at high risk and thus develop preventive measures that will reduce incidences of AMI and its complications. We modeled the dataset using the following and compared the performance of eight machine learning algorithms. Then, for each of the chosen algorithms, we used a grid search and performed 5-fold cross-validation to optimize the parameters, after which we measured their performance on the test set. Among these algorithms, we have identified the XGBoost algorithm for the best performance for chest pain data processing. This model has an AUC of 0.92, accuracy of 0.88, precision rate of 0.85, recall rate of 0.87, and F1 score of 0.86. For understanding and further optimization of the XGBoost model, we used the SHAP method to rank the feature importances with calculation of the Shapley value. Feature selection: In this regard, we examined the SHAP value analysis and selected three critical features in the prediction of young and middle-aged AMI patients. These features included emergency maximum troponin I, emergency maximum BNP, and the time from symptom onset to first medical care. Based on this, we developed the three features-based XGBoost model and gave its performance, stating that this model performed with an AUC value of 0.90 on the test set with an accuracy of 0.87, a precision of 0.84, a recall of 0.86, and an F1 score of 0.85, thus proving the effectiveness and reliability in clinical applications. SHAP value analysis for feature importance assessment identified three biomarkers, which are the most important in early screening of AMI among young and middle-aged people, and all three biomarkers are of important clinical significance in early diagnosis. As a myocardial-specific marker, the elevation of cTnI reflects the degree of damage to the myocardial cells and has been regarded as the gold standard for diagnosing AMI [25-26] . BNP is a sensitive indicator of heart failure. When BNP is elevated, increased ventricular wall tension and impaired cardiac function are indicated. It acts as an independent predictor of prognosis in AMI [27] . SO-to-FMC reflects the timeliness of treatment and is closely related to the prognosis of AMI [28] . This study originally integrates the three indicators into an early screening model for AMI among young and middle-aged individuals, using machine learning methods to calculate the relative importance. The approach may provide valued information in the early identification of AMI and risk stratification during clinical practice, hence bringing new ideas and tools to clinicians. We then compute the SHAP interaction values for all interactions of features and observe huge variation in both magnitude and pattern of interaction effects between each pair of features in predicting AMI. The heat map shows that the highest interaction effect is between Max cTnI and Adm BNP, underlining that these biomarkers are clinically relevant when interacting with each other. In particular, the dependence plot reflects that the modulatory effect of Adm BNP levels on myocardial infarction prediction is strongly increased at higher concentrations of Max cTnI, which also indicates that the combination of both predictors allows more exact risk predictions concerning myocardial infarction in patients at hospital admission. SHAP visualizes a given individual characteristic contribution to model prediction results. Force plot of features: Adm BNP and Adm CKMB contribute positively to predict the AMI, while BMI and Adm Myo make negative contributions in this particular case. These findings have important implications for clinical practice: clinicians should be particularly interested in combined assessment of Max cTnI and Adm BNP as a means for rapid assessment of patients for myocardial infarction risk in the emergency setting, thus helping optimize treatment decisions and resource distribution. Moreover, the negative contribution of BMI and Adm Myo, especially in specific patient populations, needs to be recognized because these can decrease overall predictive accuracy. To further analyze the dependence of feature importance on age, we performed subgroup analysis. We divided the patients into a young group ≤44 years old (n = 213) and a middle-aged group aged 45-59 years old (n = 559). XGBoost models were developed for both groups, followed by SHAP value analysis. Both models showed very excellent predictive capability: the AUCs were 0.993 for the young group and 0.996 for the middle-aged group. Therefore, key predictors that could be noted in both the cohorts were Max cTnI, Max BNP, and SO-to-FMC by comparing the ranking of feature importance between the two groups. However, in the case of other variables such as BMI, the impact was relatively more important in the middle-aged group compared with that in the young group. According to this analysis, an age-stratified approach should be performed in the strategy for early AMI screening, developing differentiated assessment standards and intervention measures for different age groups. The history of smoking is much more important in the younger group; therefore, the most attention must be paid to lifestyle management and enforcement of anti-smoking interventions and other harmful behaviors. Dominant age-related factors in the middle-aged group mean that smoking cessation, weight control, and active treatment of comorbidities such as hypertension and diabetes should be a point of concern. This differential approach to management may allow for better prognosis in the patients of all age groups. This study provides a quite accurate prediction model for the young and middle-aged to conduct early screening for AMI. Yet, some limitations do exist, and we do acknowledge these. This was a single-center investigation, and while the sample size is sufficient to support our conclusions, increasing it can provide stronger robustness of the results. Although the model showed very good performance when internally validated, an independent validation dataset is not available, thus our testing of generalization performance across diverse populations is limited. Despite the best efforts, unidentified selection biases exist. The model, incorporating many clinically significant features, is the result of work presented herein; future studies may also consider adding new biomarkers or comprehensive imaging features to enhance the performance. These limitations thus provide some avenues into which future research may focus: multi-center validation, expansion of sample size, and inclusion of extended clinical characteristics. Conclusion This study successfully developed an early screening model of high accuracy for AMI in young and middle-aged people using machine learning techniques. The result manifested that Max cTnI, Max BNP, and SO-to-FMC were the three most significant factors among the list of predictors of AMI in this age group, while the model performed well on the test set with an AUC of 0.979. Inter-features interaction analysis showed the complicated interaction between different biomarkers. Subgroup analysis also demonstrated distinct predictive patterns of AMI among different age groups, which means risk assessment should be necessary for individualized study. Although these results are promising, further validation through a multicenter external study and from prospective research is required to establish the generalizability and clinical applicability of the model. Declarations Ethics approval and consent to participate 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. This study was conducted in accordance with the Declaration of Helsinki (as revised in 2013) and was approved by the hospital ethics committee (approval number: 2024-KY-26-02). The requirement for informed consent was waived due to the retrospective nature of the study and the use of anonymized data from routine clinical practice. Clinical trial number: not applicable. Consent for publication This manuscript does not contain any potentially identifiable images or data. Due to the retrospective nature of the study and the use of anonymized data from routine clinical practice, the requirement for informed consent for publication was waived. Availability of data and materials Additional data and materials are available from the corresponding author on reasonable request. Competing interests All authors have completed the ICMJE uniform disclosure form. The authors have no conflicts of interest to declare. Funding This research was supported by the Research Funds of Joint Research Center for Occupational Medicine and Health of IHM (Nos. OMH-2023-05 and OMH-2023-24), the Medical Special Cultivation Project of Anhui University of Science and Technology (No. YZ2023H2A007), the National Natural Science Foundation of China (Nos. 52374155 and 61806006), and the Natural Science Research Project of Colleges and Universities in Anhui Province (No. 2022AH040113). Authors' contributions Qingqing Ruan and Shuzhi Su contributed equally to this work. Qingqing Ruan, Shuzhi Su, and Zengyong Qiao conceived and designed the experiments. Xian Wang and Xiumei Li performed the experiments. Yong Dai and Zengyong Qiao analyzed the data. Qingqing Ruan and Shuzhi Su wrote the paper. All authors contributed to editorial changes in the paper. Acknowledgements The authors would like to acknowledge all individuals and institutions that contributed to this work. Special thanks to the Joint Research Center for Occupational Medicine and Health of IHM for their support in providing resources and technical assistance. We also appreciate the contributions of the team at the Department of Cardiovascular Medicine at the Sixth People’s Hospital South Campus, Shanghai Jiao Tong University School of Medicine, for their general support. References R. D. Barnett. Acute myocardial infarction. The Lancet. 2019;393 (10191):2580-2580. doi:10.1016/s0140-6736(19)31419-9 Shukla AN,Jayaram AA,Doshi D, et al. The Young Myocardial Infarction Study of the Western Indians: YOUTH Registry. Glob Heart. 2019;14 (1):27-33. doi:10.1016/j.gheart.2018.12.001 Krittanawong C,Khawaja M,Tamis-Holland JE, et al. Acute Myocardial Infarction: Etiologies and Mimickers in Young Patients. J Am Heart Assoc. 2023;12 (18):e029971. doi:10.1161/JAHA.123.029971 Ivan Hanson,Akash Rusia,Andres Palomo, et al. Treatment of Acute Myocardial Infarction and Cardiogenic Shock: Outcomes of the RECOVER III Postapproval Study by Society of Cardiovascular Angiography and Interventions Shock Stage. Journal of the American Heart Association. 2024;0 (0):0-0. doi:10.1161/jaha.123.031803 Viola Vaccarino. Myocardial Infarction in Young Women. Circulation. 2019;139 (8):1057-1059. doi:10.1161/circulationaha.118.039298 Marcin Ambroziak,Jennifer Franke,Anna Wójcicka, et al. SYNE1 gene novel variant is associated with myocardial infarction in young people with a family history of premature atherosclerosis. European Heart Journal. 2023;44 (Supplement_2):0-0. doi:10.1093/eurheartj/ehad655.3263 Yasmina Benchabi. Is hyperhomocysteinemia a risk factor for myocardial infarction in young people?. Archives of Cardiovascular Diseases Supplements. 2022;14 (1):26-26. doi:10.1016/j.acvdsp.2021.09.060 Zhang P,Wu L,Zou TT, et al. Machine Learning for Early Prediction of Major Adverse Cardiovascular Events After First Percutaneous Coronary Intervention in Patients With Acute Myocardial Infarction: Retrospective Cohort Study. JMIR Form Res. 2024;8:e48487. doi:10.2196/48487 Khera R,Haimovich J,Hurley NC, et al. Use of Machine Learning Models to Predict Death After Acute Myocardial Infarction. JAMA Cardiol. 2021;6 (6):633-641. doi:10.1001/jamacardio.2021.0122 Wang S,Li J,Sun L, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction. BMC Med Inform Decis Mak. 2021;21 (1):301. doi:10.1186/s12911-021-01667-8 Improving the diagnosis of myocardial infarction with machine learning. Nat Med. 2023;29 (5):1070-1071. doi:10.1038/s41591-023-02331-6 Suhuai Wang,Jingjie Li,Li Sun, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction. BMC Medical Informatics and Decision Making. 2021;21 (1):0-0. doi:10.1186/s12911-021-01667-8 Sang Hoon Lee,Sulyun Lee,W. Nick Street, et al. Machine learning approaches to predict the 1-year-after-initial-AMI survival of elderly patients. BMC Medical Informatics and Decision Making. 2022;22 (1):0-0. doi:10.1186/s12911-022-01854-1 Than MP,Pickering JW,Sandoval Y, et al. Machine Learning to Predict the Likelihood of Acute Myocardial Infarction. Circulation. 2019;140 (11):899-909. doi:10.1161/CIRCULATIONAHA.119.041980 McCord J,Gibbs J,Hudson M, et al. Machine Learning to Assess for Acute Myocardial Infarction Within 30 Minutes. Crit Pathw Cardiol. 2022;21 (2):67-72. doi:10.1097/HPC.0000000000000281 Jahmunah V,Ng EYK,San TR, et al. Automated detection of coronary artery disease, myocardial infarction and congestive heart failure using GaborCNN model with ECG signals. Comput Biol Med. 2021;134:104457. doi:10.1016/j.compbiomed.2021.104457 Mandair D,Tiwari P,Simon S, et al. Prediction of incident myocardial infarction using machine learning applied to harmonized electronic health record data. BMC Med Inform Decis Mak. 2020;20 (1):252. doi:10.1186/s12911-020-01268-x Li X,Shang C,Xu C, et al. Development and comparison of machine learning-based models for predicting heart failure after acute myocardial infarction. BMC Med Inform Decis Mak. 2023;23 (1):165. doi:10.1186/s12911-023-02240-1 Austin DE,Lee DS,Wang CX, et al. Comparison of machine learning and the regression-based EHMRG model for predicting early mortality in acute heart failure. Int J Cardiol. 2022;365:78-84. doi:10.1016/j.ijcard.2022.07.035 Li Y,Hu Y,Jiang F, et al. Combining WGCNA and machine learning to identify mechanisms and biomarkers of ischemic heart failure development after acute myocardial infarction. Heliyon. 2024;10 (5):e27165. doi:10.1016/j.heliyon.2024.e27165 Zhang Q,Burrage MK,Shanmuganathan M, et al. Artificial Intelligence for Contrast-Free MRI: Scar Assessment in Myocardial Infarction Using Deep Learning-Based Virtual Native Enhancement. Circulation. 2022;146 (20):1492-1503. doi:10.1161/CIRCULATIONAHA.122.060137 Chandrasekhar J,Gill A,Mehran R. Acute myocardial infarction in young women: current perspectives. Int J Womens Health. 2018;10:267-284. doi:10.2147/IJWH.S107371 Åberg ND,Adiels M,Lindgren M, et al. Diverging trends for onset of acute myocardial infarction, heart failure, stroke and mortality in young males: role of changes in obesity and fitness. J Intern Med. 2021;290 (2):373-385. doi:10.1111/joim.13285 Enas EA,Varkey B,Dharmarajan TS, et al. Lipoprotein(a): An underrecognized genetic risk factor for malignant coronary artery disease in young Indians. Indian Heart J. 2019;71 (3):184-198. doi:10.1016/j.ihj.2019.04.007 Chen Y,Tao Y,Zhang L, et al. Diagnostic and prognostic value of biomarkers in acute myocardial infarction. Postgrad Med J. 2019;95 (1122):210-216. doi:10.1136/postgradmedj-2019-136409 Wang Y. Immune-related biomarkers in myocardial infarction; diagnostic/prognostic value and therapeutic potential. J Biochem Mol Toxicol. 2023;37 (12):e23489. doi:10.1002/jbt.23489 Puelacher C,Wagener M,Honegger U, et al. Combining high-sensitivity cardiac troponin and B-type natriuretic peptide in the detection of inducible myocardial ischemia. Clin Biochem. 2018;52:33-40. doi:10.1016/j.clinbiochem.2017.10.014 Chaulin AM. Cardiac Troponins Metabolism: From Biochemical Mechanisms to Clinical Practice (Literature Review). Int J Mol Sci. 2021 Oct 10;22(20):10928. doi: 10.3390/ijms222010928. PMID: 34681585; PMCID: PMC8535601. Additional Declarations No competing interests reported. Cite Share Download PDF Status: Under Revision Version 1 posted Editorial decision: Revision requested 19 Dec, 2024 Editor assigned by journal 19 Dec, 2024 Submission checks completed at journal 18 Dec, 2024 First submitted to journal 10 Dec, 2024 You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-5614054","acceptedTermsAndConditions":true,"allowDirectSubmit":false,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":392290091,"identity":"ff3aa381-12a3-4e7b-bfb6-732163adf133","order_by":0,"name":"QingQing Ruan","email":"","orcid":"","institution":"Anhui University of Science \u0026 Technology","correspondingAuthor":false,"prefix":"","firstName":"QingQing","middleName":"","lastName":"Ruan","suffix":""},{"id":392290092,"identity":"84061723-ff62-472a-8a5c-8e7bcfff7981","order_by":1,"name":"Shuzhi Su","email":"","orcid":"","institution":"Anhui University of Science \u0026 Technology","correspondingAuthor":false,"prefix":"","firstName":"Shuzhi","middleName":"","lastName":"Su","suffix":""},{"id":392290093,"identity":"37102f75-e1ad-477e-bb37-2cbbfbb40289","order_by":2,"name":"Xian Wang","email":"","orcid":"","institution":"Anhui University of Science \u0026 Technology","correspondingAuthor":false,"prefix":"","firstName":"Xian","middleName":"","lastName":"Wang","suffix":""},{"id":392290094,"identity":"f07a65c1-edb4-44c3-af7e-48e1b36a9726","order_by":3,"name":"Xiumei Li","email":"","orcid":"","institution":"the Sixth People’s Hospital South Campus, Shanghai Jiao Tong University School of Medicine","correspondingAuthor":false,"prefix":"","firstName":"Xiumei","middleName":"","lastName":"Li","suffix":""},{"id":392290096,"identity":"eb7025d4-6bdd-4cc7-acfd-a0a09f560fb3","order_by":4,"name":"Zengyong Qiao","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAA2ElEQVRIiWNgGAWjYBACfmYwZSPHz97AcIAoLZLNYCrNWLLnAJFaDCDKDiUazEgg0mEGx5kfPvzZdiDBQPL5w8MFNQzy/GIELDM4zGZsINl2J89cOsfg8IxjDIYzZxOwzuAwg5mEYduzYsvZOQyHedgYEgxuE9Bidpj9m0Ri2+HEDTePPzjM848ILcaHecwkDoK03ADayNtGhBbJZp5iw4ZzoEAG+oW3T4KwX/j9j298+KMMFJXHH3/m+WYjzy9NQAsYMLLBmRJEKAeDP8QqHAWjYBSMghEJAGfDRxkjG3PtAAAAAElFTkSuQmCC","orcid":"","institution":"the Sixth People’s Hospital South Campus, Shanghai Jiao Tong University School of Medicine","correspondingAuthor":true,"prefix":"","firstName":"Zengyong","middleName":"","lastName":"Qiao","suffix":""},{"id":392290098,"identity":"b6f57de3-521e-4038-896e-162c3e6f5831","order_by":5,"name":"Yong Dai","email":"","orcid":"","institution":"Anhui University of Science \u0026 Technology","correspondingAuthor":false,"prefix":"","firstName":"Yong","middleName":"","lastName":"Dai","suffix":""}],"badges":[],"createdAt":"2024-12-10 07:23:42","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-5614054/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-5614054/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":72289105,"identity":"d175f0cd-0423-44ed-9e69-aed6ca71b685","added_by":"auto","created_at":"2024-12-24 17:18:19","extension":"png","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":260313,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFlow chart of the study design.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSMOTE: Synthetic Minority Over-sampling Technique; ML:machine learning; XGBoost: eXtreme Gradient Boosting; SHAP: SHapley Additive explanation.\u003c/p\u003e","description":"","filename":"1.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/189752800b6d98d08e54dda6.png"},{"id":72292103,"identity":"cd0d6581-9d2c-4b71-a5da-679dc45e7ca7","added_by":"auto","created_at":"2024-12-24 17:34:19","extension":"png","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":65585,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eDistribution of categories within the training set for young and middle-aged groups.\u003c/strong\u003e Figure 2A displays the original categories before the utilization of SMOTE, while Figure 2B displays the categories after the utilization of SMOTE. In this case, '1' stands for acute myocardial infarction and '0' for non-acute myocardial infarction.\u003c/p\u003e","description":"","filename":"2.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/5ca962de1dc3e9439cb26653.png"},{"id":72289094,"identity":"e15350ef-630a-4c9d-9dc1-41b8637eeb9d","added_by":"auto","created_at":"2024-12-24 17:18:19","extension":"png","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":94874,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eA comparison of ROC curves for various ML algorithms applied to the test set. \u003c/strong\u003eFigure 3 compares ROC curves of various ML algorithms in the test set. The horizontal axis shows the False Positive Rate, and the vertical axis shows the True Positive Rate. A comparison of all the above algorithms reveals the best diagnostic models with respect to the prediction of chest pain, represented by XGBoost with 0.973 AUC and GB with 0.968 AUC. Whereas RF, DT, and SVC can be referred to as the strong predictors with 0.965, 0.949, and 0.946 AUC, respectively. Poor results were received for KNN with the AUC of only 0.700. The larger the AUC, the stronger will be the predictive ability of every algorithm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（rf：Random Forest,knn:K-Nearest Neighbors,svc:Support Vector Classification,mlp:Multi-Layer Perceptron,gb:Gradient Boosting,dt:Decision Tree,xgb:eXtreme Gradient Boosting,lr：Logistic Regression）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"3.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/fbec2847f6cab4e75efe5eb8.png"},{"id":72289074,"identity":"cc380514-8352-4488-b18b-e095125532c8","added_by":"auto","created_at":"2024-12-24 17:18:18","extension":"png","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":95111,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eConfusion matrix heatmap of various models. \u003c/strong\u003eThis figure depicts the respective performances of each model on a test set that has exclusively been used for the classification of chest pain data. The columns in the confusion matrix represent the true labels (True Label) of the data, whereas the rows give information about the labels as predicted by the algorithm, Predicted Label. In these confusion matrices, the color intensity represents the frequency of the prediction result. The darker it is, the higher the count. These confusion matrices can visually check the accuracy and error rate of each algorithm in classifying both positive and negative samples.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(1: Acute myocardial infarction, 0: Non-acute myocardial infarction).\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"4.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/728a15df40dd610c83127f84.png"},{"id":72291203,"identity":"c3c7ce78-7a83-45cb-bebc-549a38828af9","added_by":"auto","created_at":"2024-12-24 17:26:19","extension":"png","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":145687,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP value analysis of feature importance rank for Top 20 Features in the XGBoost model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 5A shows the bar plot of the mean SHAP values of the top 20 features. The bars in blue color represent the average SHAP value of each feature, ranking their relative importance to the model prediction output. The three most critical features, according to the chart, are \"Max cTnI\"-maximum cardiac troponin I concentration, \"Max BNP\"-maximum brain natriuretic peptide concentration, and \"SO-to-FMC\"-the time from symptom onset to first medical contact. Other important ones include CK-MB creatine kinase-MB at admission, Smk Hx-smoking history, body mass index, and Adm SCr-creatinine level at admission.\u003c/p\u003e\n\u003cp\u003eFigure 5B: SHAP values of the top 20 features are presented, where on the right side in this honeycomb diagram presents a specific value of each feature in impacting the prediction output for a given sample.The blue-to-red color gradient shows the range of variation in the value of the features. A high SHAP value indicates that this feature is positively contributing to the model's prediction of a high risk. Low values mean otherwise. For instance, high values of \"Max cTnI\" and \"Max BNP\" are usually associated with a high risk from myocardial infarction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"5.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/ffa6306e4bf075c54d39154b.png"},{"id":72289093,"identity":"a60457a4-24f8-410f-957a-4f1b96c87c2f","added_by":"auto","created_at":"2024-12-24 17:18:19","extension":"png","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":144259,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap of Features for Early Screening Models in Young and Middle-aged Myocardial Infarction Patients. \u003c/strong\u003eFigure 6 illustrates the SHAP value heat map for the predicted model features, providing a detailed explanation for each sample by quantifying the specific contribution of individual features to the model's prediction results. The horizontal axis represents the test set samples, while the left vertical axis indicates the importance ranking of features, sorted from highest to lowest influence. The right vertical axis visualizes feature influence, with color depth reflecting the magnitude of SHAP values; darker colors correspond to larger absolute SHAP values, indicating more significant feature impact on model predictions. The top area visualizes the model's prediction results based on these feature values. This analysis corroborates that Max cTnI, Max BNP, and SO-to-FMC are the three most influential features in the model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"6.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/62a346d9f32d20e64e49e05b.png"},{"id":72289079,"identity":"5a26087e-9281-41e8-836c-099fc7741953","added_by":"auto","created_at":"2024-12-24 17:18:18","extension":"png","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":287722,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eHeatmap of Feature Interaction Values and Major Feature Interactions.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 7A presents a heat map illustrating the interaction values among all features, where the color of each grid indicates the absolute average value of the interaction between corresponding feature pairs, with values ranging from 0 to 0.15. \u003c/strong\u003eThe depth of color in the figure corresponds to the strength of the interaction between features; darker colors signify more substantial interaction effects. This heat map reveals significant interactions among certain feature combinations in predicting AMI, notably between Max cTnI and Adm BNP.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 7B illustrates the interactions among the first seven features. \u003c/strong\u003eThese characteristics include: Max cTnI (maximum troponin I level), Max BNP (maximum B-type natriuretic peptide level), SO-to-FMC (time between the onset of symptoms and first medical contact), Adm CKMB (admission muscle acid kinase isoenzyme level), Smk Hx (smoking history), BMI (body mass index), and Adm SCr (serum creatinine level upon admission). Each site in the figure represents the influence of a single feature on the model output, while the color coding reflects the value of a specific feature.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"7.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/925656c7184543136b1bc3dc.png"},{"id":72291204,"identity":"13b23f47-d10d-4d22-a7ce-bba5efa611f2","added_by":"auto","created_at":"2024-12-24 17:26:19","extension":"png","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":76743,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP Dependence Plot Demonstrating Interaction of Max cTnI and Adm BNP.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 8A provides an overview of the general trend in the SHAP value for Max cTnI in relation to its concentration, in an attempt to develop a sense about the impact that this variable can have independently on the model prediction result. From the figure, one can clearly see that as the values of Max cTnI go up, the SHAP value continues increasing and shows its critical role regarding myocardial infarction prediction.\u003c/p\u003e\n\u003cp\u003eThe impact of Adm BNP is represented through the color gradient of Figure 8B, whose value regulates the Max cTnI SHAP. The proximity to red indicates higher values for Adm BNP, while the cooler colors reflect the low values. High Adm BNP values are associated with high SHAP values for Max cTnI; this would mean that Adm BNP increased the explanatory strength of Max cTnI in the model.\u003c/p\u003e\n\u003cp\u003eFigure 8C: SHAP main effect value of Max cTnI. The plot here strengthens the position of Max cTnI as an important predictor, since from the results, it can be observed that above the threshold value of Max cTnI, the SHAP main effect value is high, showing its importance to predict a patient prognosis at an elevated level.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide)\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"8.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/b1e6c55de48945e4dfd73847.png"},{"id":72292101,"identity":"d6625ade-c94f-4318-aad9-4765b4073e75","added_by":"auto","created_at":"2024-12-24 17:34:18","extension":"png","order_by":9,"title":"Figure 9","display":"","copyAsset":false,"role":"figure","size":122596,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eSHAP force plot displaying the feature contributions to model output.\u003c/strong\u003e Each feature contribution can be seen using a red color, which shows the positive contribution, and blue represents the negative contribution. The length of each bar is proportional to the magnitude of the contribution of the respective feature to the final predicted value. Figure 9A shows the average strength of a feature's influence across the test set as a whole and serves well in depicting, on average, the relative importance of various features to the model output. The prominent features are Max BNP and Adm CKMB among others. Figure 9B emphasizes that features such as Max BNP and Adm CKMB have a great impact, whereas Figure 9C shows the contributions of other features like BMI and Adm Myo, which have negative values in the model. Decomposition further reinforces our presented analysis of feature importance for showing the different role played by the features in various sample categories.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Adm Myo：Admission Myoglobin）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"9.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/5fa88bbc62eacbbc5863849c.png"},{"id":72289086,"identity":"b7cd3324-5979-470e-af06-59bb2673d79f","added_by":"auto","created_at":"2024-12-24 17:18:18","extension":"png","order_by":10,"title":"Figure 10","display":"","copyAsset":false,"role":"figure","size":72567,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eAnalysis of feature parsimony versus model performance.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Line graph of feature parsimony illustrating AUC values of the XGBoost model with variation in a number of selected features. The vertical axis is for AUC value while the horizontal axis is for the number of selected features. This graph shows how feature selection affects the performance of the model.\u003c/p\u003e\n\u003cp\u003e(B) ROC Curve of XGBoost Model Based on Three Most Important Features.Showing Classification Performance of Model on Test Set. The shape of this curve and the corresponding AUC values below provide a critical basis for evaluating its accuracy.\u003c/p\u003e","description":"","filename":"10.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/08131f2a9edae863460667cc.png"},{"id":72289073,"identity":"863ffce4-034d-4fce-bb8b-d808ace095a8","added_by":"auto","created_at":"2024-12-24 17:18:18","extension":"png","order_by":11,"title":"Figure 11","display":"","copyAsset":false,"role":"figure","size":54589,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eROC Curve Performance of the XGBoost Model Across Different Age Groups.\u003c/strong\u003eFigure 11 illustrates the ROC curve obtained from the XGBoost model with 30 features in classifying the young group (Figure A) and the middle-aged group (Figure B). The ROC curves of the two models are also almost perfect, reaching an AUC close to 1; thus, the AUCs of the young and middle-aged groups are 0.993 and 0.996, respectively. Besides, these results indicate that the XGBoost model with 30 features can make pretty good predictions about myocardial infarction for different age groups and estimate feature importance corresponding to individuals from different age brackets quite correctly. So, this model should be a good starting point for developing a SHAP Interpreter in order to check differences in feature importance arising among diverse age groups.\u003c/p\u003e","description":"","filename":"11.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/dfd40431ce53798fc9419fbc.png"},{"id":72291207,"identity":"0fd151d6-6d48-4a36-86b4-220f1fd83594","added_by":"auto","created_at":"2024-12-24 17:26:19","extension":"png","order_by":12,"title":"Figure 12","display":"","copyAsset":false,"role":"figure","size":243902,"visible":true,"origin":"","legend":"\u003cp\u003e\u003cstrong\u003eFeature importance distribution across age groups based on SHAP values. \u003c/strong\u003eFigure 12 presents the ranking of feature importance for youth and middle-aged groups derived from an SHAP value evaluation and its contribution toward model output. A comparison of average SHAP values and distribution of various features in both age groups brings out the relative contributions of various features in predicting myocardial infarction. Parts A and C show the average value of each feature's contribution to explain the output of the model, while B and D depict the distribution of the SHAP values, where color is used to emphasize the direction and magnitude of feature contributions with regard to a prediction. Max cTnI, Max BNP, admission CKMB, and SO-to-FMC were significant contributors in both age cohorts. However, other predictors displayed differential effects with age, especially for BMI, whose impact on the prediction outcomes was stronger in the middle-aged group compared to youth.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate）\u003c/strong\u003e\u003c/p\u003e","description":"","filename":"12.png","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/b71fe4ffddc59512486856cf.png"},{"id":72292471,"identity":"93b0983c-cd3a-4018-988b-e0f115ca5e82","added_by":"auto","created_at":"2024-12-24 17:42:19","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3686029,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-5614054/v1/0c8c4294-2df1-47df-bb36-728fd5a92b37.pdf"}],"financialInterests":"No competing interests reported.","formattedTitle":"Development and Validation of an Interpretable Prediction Model for Early Screening of Acute Myocardial Infarction in Young and Middle-Aged Patients","fulltext":[{"header":"1. INTRODUCTION","content":"\u003cp\u003eStudies report that, in the past decade, there has been a near 15% rise in the incidence of AMI in the population aged between 30 to 50 years, with a tendency towards an increasingly younger age of onset \u003csup\u003e[\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]\u003c/sup\u003e. This is attributed to a change in the style of living among youngsters, increased psychosocial stress, and genetic propensities \u003csup\u003e[\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]\u003c/sup\u003e. The increase in cases among younger populations results in a tremendous escalation of medical expenditure and socio-economic burden due to cardiovascular disease. These patients suffer from chronic health problems, which, apart from personal suffering, further enhance the medical resources needed. Though much attention has been directed to acute myocardial infarction in young and middle-aged individuals in recent years, research in the past on myocardial infarction has focused on older patients, and relatively little attention has been paid so far to determining the characteristics and risk factors associated with AMI in younger patients. Consequently, studies on its early screening and diagnosis are still less well developed today. Due to the increase of the incidence in relatively younger populations, the need for an accurate early-screening tool is urgent in AMI. Moreover, an early-screening tool for AMI should be easy to operate for users, accessible, and provide identification of not just high-risk individuals but assistance as early as possible. Conventional screening modalities include ECG and biochemical marker detection. Conventional means have shown suboptimal sensitivity and specificity in diagnosing early AMI\u003csup\u003e[\u003cspan additionalcitationids=\"CR9\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR10\" class=\"CitationRef\"\u003e10\u003c/span\u003e]\u003c/sup\u003e. It is therefore likely that interest in newer methodologies may be necessary to better delineate this disease.\u003c/p\u003e \u003cp\u003eA new approach is needed to transcend the shortcomings of the traditional screening technique to raise the early detection rate of AMI among young and middle-aged populations. The application of machine learning has become widely regarded, with its strong data analytics and predictive powers, as one of the key contributors to improving the effectiveness of early screening in relation to the prediction of AMI. Various studies have shown that the ML algorithms may combine the patient age and sex along with cardiac troponin, hs-cTnI, levels in a diagnostic tool-the Myocardial Ischemic Injury Index, MI3-so as to furnish a rather more personalized estimate of acute myocardial infarction risk than that obtainable by traditional means \u003csup\u003e[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]\u003c/sup\u003e. Some researchers also employed deep learning algorithms to develop a semi-automated system capable of classifying the ECG signals into normal, coronary heart disease, and myocardial infarction categories for early diagnosis and alerts to clinicians \u003csup\u003e[\u003cspan citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e]\u003c/sup\u003e. The construction of machine learning-based predictive models allows the researchers to proactively identify the high-risk patients\u003csup\u003e[\u003cspan citationid=\"CR17\" class=\"CitationRef\"\u003e17\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e]\u003c/sup\u003e.\u003c/p\u003e \u003cp\u003eML enhances early detection of myocardial infarction by increasing diagnostic precision, reducing response time, managing patients personally, and offering enhanced decision support \u003csup\u003e[\u003cspan additionalcitationids=\"CR20\" citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]\u003c/sup\u003e. This study evaluates the application of ML in early screening and diagnosis of AMI in young and middle-aged patients. An interpretable model was developed and tested for predicting AMI in this population by integrating multidimensional variables with age-stratified analyses. The novelty of this study is that the feature selection method and optimization model allow new diagnosis perspectives for different age groups and thus increase the accuracy of early screening for AMI in young and middle-aged populations. Besides, it may improve clinical outcomes for such patients and so lay a good foundation for future clinical application.\u003c/p\u003e"},{"header":"2. METHODS","content":"\u003cp\u003e\u003cstrong\u003e2.1 Study Population and Study Design\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe present study was a retrospective one conducted from the Chest Pain Center of the South Campus of Shanghai Sixth People\u0026apos;s Hospital. The inclusion criteria of study subjects were in line with the diagnostic definition of AMI according to current guidelines, including patients who met the criteria of STEMI and NSTEMI. Controls were patients attending the chest pain center for chest discomfort in the same period but without a diagnosis of AMI. All subjects had to be aged between 18 and 59 years. For the final sample size, which included an analysis, a total of 772 individuals were studied: 640 patients who comprised the AMI group and 132 who formed the non-AMI control group.The subjects were additionally categorized into a young group (\u0026le;44 years old, n=213) and a middle-aged group (45-59 years old, n=559) based on age. This study adhered strictly to the ethical guidelines of the Declaration of Helsinki and received approval from the hospital ethics committee (approval number: 2024-KY-26-02). Given that this study solely involved the collection of relevant data without interfering with routine diagnostic and treatment activities, informed consent was not required. Figure 1 illustrates the design concept of this study, which encompasses three steps: data preprocessing, model development, and subgroup analysis.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 1: Flow chart of the study design.\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eSMOTE: Synthetic Minority Over-sampling Technique; ML:machine learning; XGBoost: eXtreme Gradient Boosting; SHAP: SHapley Additive explanation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e\u0026nbsp;2.2 Data Collection\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn this study, we collected a total of 30 variables from patients\u0026apos; electronic medical records in the emergency department. These variables encompassed patient demographic information, medical history, electrocardiographic characteristics, echocardiographic features, and initial laboratory parameters. The variables included: gender, age at onset, time from symptom onset to first medical consultation (SO-to-FMC), weight (kg), height (cm), body mass index (BMI), systolic blood pressure on admission, diastolic blood pressure on admission, heart rate upon admission, cardiac function class, history of hypertension, hyperlipidemia, diabetes, smoking history, troponin levels upon admission, highest troponin I (cTnI) recorded in the emergency room, serum creatinine on admission, D-dimer on admission, B-type natriuretic peptide (BNP) on admission, myoglobin on admission, creatine kinase MB (CKMB) on admission, highest BNP value in the emergency department, and left ventricular ejection fraction (LVEF%) recorded as the lowest value in the emergency department. Additionally, we noted the history of coronary heart disease, revascularization, cerebrovascular disease, peripheral vascular disease, chronic kidney disease, chronic heart failure, and atrial fibrillation.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.3 Data Processing\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe data integrity of this study is high, with missing data variables not exceeding 15% of the sample size, thus eliminating the need to delete any variables. To address missing data, various strategies\u0026mdash;including random filling, mean filling, and mode filling\u0026mdash;were employed to preserve the shape of the variable distribution and ensure prediction accuracy. Outliers were addressed through random, mean, and mode replacement methods, in accordance with clinical recommendations. Following the conversion of textual non-numeric data into numerical variables, the dataset was divided into a training set and a test set in an 80:20 ratio. The training set utilized SMOTE(Synthetic Minority Over-sampling Technique) oversampling to mitigate the class imbalance problem, while the test set was reserved for evaluating model performance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.4 Data Imbalance Processing\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe dataset was divided into a training set and a test set in an 80:20 ratio. The training set for this study comprised 617 data samples ( Figure 2A), including 513 cases of AMI and 104 cases of non-AMI, which illustrates a significant data imbalance. This imbalance may lead the model to exhibit a stronger learning effect on the majority class samples while inadequately learning from the minority class samples, ultimately impacting the overall performance of the model. To address this issue, we employed SMOTE for oversampling. SMOTE generates synthetic samples by selecting the nearest neighbors of minority class samples, thereby augmenting the number of minority class samples, enhancing the model\u0026apos;s generalization ability towards these classes, and alleviating overfitting issues. As depicted in Figure 2B, following SMOTE processing, the training set was expanded to 1026 samples, comprising 513 cases of acute myocardial infarction and 513 cases of non-acute myocardial infarction, achieving class balance.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 2: Distribution of categories within the training set for young and middle-aged groups.\u003c/strong\u003e Figure 2A displays the original categories before the utilization of SMOTE, while Figure 2B displays the categories after the utilization of SMOTE. In this case, \u0026apos;1\u0026apos; stands for acute myocardial infarction and \u0026apos;0\u0026apos; for non-acute myocardial infarction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.5 Data Set Standardization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eIn order to reduce the effect of different dimensions and numerical ranges on data analysis, model training, and interpretation of results, this study uses Z-score normalization for features of the dataset so that all features contribute equally during model training for stability and generalization ability. Transformed features in the distribution with mean 0 and standard deviation 1 achieve comparability on different scales. This method not only will make the learning relationships between features more effective but also enhance the prediction accuracy of the model, especially in clinical applications, to classify acute myocardial infarction from non-acute myocardial infarction with greater precision. This very same process of standardization was then performed for the test set features to ensure that measures were consistent and hence the validity and reliability of the model could be warranted when applied in the real world.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.6 Feature Selection\u003c/strong\u003e\u0026nbsp;\u003c/p\u003e\n\u003cp\u003eWe screened potential candidate variables through preliminary univariate analysis, establishing a significance level (p-value) of less than 0.05 as the criterion for inclusion. Variables meeting this criterion were included in the candidate variable set. XGBoost was identified as the optimal model through training and testing, and the SHAP interpreter was employed to conduct feature importance analysis, ranking all features from high to low importance and examining the degree of interaction between them. Features were selected based on clinical considerations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.7 Model Development and Verification\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter applying the SMOTE technology, the number of samples in the training set was extended to 1,026. In this paper, eight popular machine learning models were trained: Random Forest, K-Nearest Neighbors, Support Vector Machine, Multi-Layer Perceptron, Gradient Boosting, Decision Tree, Extreme Gradient Boosting, and Logistic Regression. The optimal parameters for each model were searched by using the grid search method along with 5-fold cross-validation. After defining the best parameters, each model\u0026apos;s performance was measured on the test set with accuracy, precision, recall, F1 score, AUC, and confusion matrix. The best performance gave the chosen model for the basis of further research and utilized 30 features for forecasting acute myocardial infarction. This systematic approach will ensure that the proposed model is optimum and clinically applicable.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e2.8 Feature Importance Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAfter the best model was determined, we conducted a feature importance analysis so that more insight into the contribution of each feature on the model\u0026apos;s prediction result could be gathered. We provided both global and local explanations of each model prediction using the SHAP method. Computing the Shapley values of the different input features gave us a ranking of feature importances relative to the entire model predictions. By using global explanations, we could identify which features had the greatest impacts on the predictions of the model, and local explanations showed why the model made a certain decision. This systematic approach enhances model performance, bolstering interpretability and credibility for clinical applications.\u003c/p\u003e"},{"header":"3. Results","content":"\u003cp\u003e\u003cstrong\u003e3.1 Population Characteristics\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe analyzed data from 772 young and middle-aged patients who visited the Chest Pain Center at the South Campus of Shanghai Sixth People\u0026apos;s Hospital between January 2018 and April 2024. This cohort included 640 patients diagnosed with AMI and 132 without AMI, with 213 classified as young (age \u0026le; 44 years) and 559 as middle-aged (ages 45 to 59 years). The study aimed to establish an AMI prediction model. Through preliminary univariate analysis, we identified variables with a significance level (p-value) of less than 0.05, which were subsequently included in the candidate variable set for model construction.The following characteristics demonstrated statistical significance (P \u0026lt; 0.05): SO-to-FMC time (P = 0.004), weight (P = 0.021), height (P = 0.003), troPonin on admission (P \u0026lt; 0.001), maximum troPonin I value (P \u0026lt; 0.001), serum creatinine on admission (P = 0.004), myoglobin on admission (P \u0026lt; 0.001), creatine kinase MB on admission (P \u0026lt; 0.001), maximum Brain natriuretic PePtide (P \u0026lt; 0.001), gender (P \u0026lt; 0.001), smoking history (P \u0026lt; 0.001), history of coronary heart disease (P \u0026lt; 0.001), history of coronary revascularization (P \u0026lt; 0.001), history of cerebrovascular disease (P \u0026lt; 0.001), history of chronic kidney disease (P \u0026lt; 0.001), history of heart failure (P \u0026lt; 0.001), and history of atrial fibrillation (P \u0026lt; 0.001) (Table 1). These characteristics will be given considerable attention during the subsequent model building and optimization processes to enhance the performance and accuracy of the predictive model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 1 Baseline characteristics of patients\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"0\" cellspacing=\"0\" cellpadding=\"0\" width=\"624\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eVariable\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eOverall, N = 772\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0, N = 132\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1, N = 640\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ep-value\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAge at onset, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e48.97 (7.56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e50.01 (7.42)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e48.76 (7.58)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.081\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSO-to-FMC, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e952.92 (2,157.24)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1,710.12 (3,542.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e796.75 (1,702.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBody Weight(kg), mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e75.43 (13.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e73.02 (12.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e75.93 (13.56)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.021\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHeight(cm), mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e169.73 (5.90)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e168.24 (6.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e170.03 (5.80)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.003\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eBMI, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e26.15 (4.35)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e25.78 (4.44)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e26.22 (4.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.298\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm SBP, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e147.57 (28.98)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e150.17 (27.22)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e147.04 (29.33)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.236\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm DBP, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e90.82 (18.97)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e89.71 (17.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e91.05 (19.36)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.423\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm HR, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e80.40 (18.72)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e78.63 (14.75)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e80.77 (19.43)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.154\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm Tn, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e104.76 (341.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e9.64 (26.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e124.37 (371.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMax cTnI, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e55.84 (60.27)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.73 (2.62)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e67.21 (60.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm SCr, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e82.36 (69.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e73.47 (25.49)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e84.19 (75.59)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.004\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm D-dimer, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.96 (2.88)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.76 (0.65)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.01 (3.15)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.069\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm BNP, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e88.22 (285.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e93.38 (464.02)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e87.15 (232.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.881\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm Myo, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e215.89 (301.79)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e63.90 (121.25)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e247.24 (317.99)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAdm CKMB, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e23.53 (49.08)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e3.01 (6.37)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e27.76 (52.85)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMax BNP, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e418.38 (1,652.53)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e99.95 (311.23)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e484.06 (1,802.73)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMin LVEF%, mean (sd)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e59.61 (9.00)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e60.76 (7.86)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e59.37 (9.20)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.075\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSex, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e702.00 (90.93%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e109.00 (82.58%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e593.00 (92.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e70.00 (9.07%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e23.00 (17.42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e47.00 (7.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eNYHA, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.381\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e733.00 (94.95%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e129.00 (97.73%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e604.00 (94.38%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e2\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e16.00 (2.07%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e2.00 (1.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e14.00 (2.19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e3\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.00 (0.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.00 (0.00%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e4.00 (0.63%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e4\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e19.00 (2.46%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e1.00 (0.76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e18.00 (2.81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of HTN, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.986\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e365.00 (47.28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e63.00 (47.73%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e302.00 (47.19%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e407.00 (52.72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e69.00 (52.27%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e338.00 (52.81%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of HLD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.446\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e436.00 (56.48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e79.00 (59.85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e357.00 (55.78%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e336.00 (43.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e53.00 (40.15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e283.00 (44.22%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of DM, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.373\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e582.00 (75.39%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e95.00 (71.97%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e487.00 (76.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e190.00 (24.61%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e37.00 (28.03%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e153.00 (23.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSmk Hx, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e184.00 (23.83%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e67.00 (50.76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e117.00 (18.28%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e588.00 (76.17%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e65.00 (49.24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e523.00 (81.72%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of CAD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e707.00 (91.58%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e100.00 (75.76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e607.00 (94.84%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e65.00 (8.42%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e32.00 (24.24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e33.00 (5.16%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of Revasc for CAD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e729.00 (94.43%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e108.00 (81.82%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e621.00 (97.03%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e43.00 (5.57%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e24.00 (18.18%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e19.00 (2.97%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of CVD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e734.00 (95.08%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e111.00 (84.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e623.00 (97.34%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e38.00 (4.92%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e21.00 (15.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e17.00 (2.66%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of PVD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e0.840\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e500.00 (64.77%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e87.00 (65.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e413.00 (64.53%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e272.00 (35.23%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e45.00 (34.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e227.00 (35.47%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of CKD, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e722.00 (93.52%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e112.00 (84.85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e610.00 (95.31%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e50.00 (6.48%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e20.00 (15.15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e30.00 (4.69%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of CHF, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e750.00 (97.15%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e117.00 (88.64%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e633.00 (98.91%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e22.00 (2.85%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e15.00 (11.36%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e7.00 (1.09%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eHx of AF, n (p%)\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e772\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026lt;0.001\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e747.00 (96.76%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e115.00 (87.12%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e632.00 (98.75%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 134px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e1\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 74px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e25.00 (3.24%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e17.00 (12.88%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e8.00 (1.25%)\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 104px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e(1: Indicates the presence of a relevant medical history or condition.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e0: Signifies the absence of such a history or condition.)\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.2 Model Construction and Evaluation\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis article tries to model and compare the performances of eight machine learning algorithms on a dataset of young and middle-aged patients presenting with chest pain to determine the best model. Regarding this, the eight machine learning algorithms employed in this research are RF, KNN, SVM, MLP, GB, DT, XGBoost, and LR. We used a combined approach of grid search and 5-fold cross-validation within the training set as a means of optimizing the parameters of all the algorithms by systematically changing model parameters such that optimum performance may be yielded. Having finished training, we tested the performance of our optimized models on the test set to see their accuracy and reliability in processing chest pain data.Curves AUC-ROC were constructed, calculating the AUC as quantification of the predictive power of each of the models. This is shown in Figure 3.\u003c/p\u003e\n\u003cp\u003eWe then further presented, with the use of heat maps, the results of test set classification by each model and drew a confusion matrix for each model (Figure 4). From the confusion matrix we calculated for each model the accuracy, precision, recall, and F1 score in order to enable a detailed evaluation of their performance (Table 2). The result of this wide review thus suggests that the best performance is from the XGBoost algorithm, and as such, this is the model which forms the basis for further research. More detailed performance measures in each category of the XGBoost model are shown on the test set in Table 3.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 3 .A comparison of ROC curves for various ML algorithms applied to the test set.\u0026nbsp;\u003c/strong\u003eFigure 3 compares ROC curves of various ML algorithms in the test set. The horizontal axis shows the False Positive Rate, and the vertical axis shows the True Positive Rate. A comparison of all the above algorithms reveals the best diagnostic models with respect to the prediction of chest pain, represented by XGBoost with 0.973 AUC and GB with 0.968 AUC. Whereas RF, DT, and SVC can be referred to as the strong predictors with 0.965, 0.949, and 0.946 AUC, respectively. Poor results were received for KNN with the AUC of only 0.700. The larger the AUC, the stronger will be the predictive ability of every algorithm.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e（\u003c/strong\u003e\u003cstrong\u003erf\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003cstrong\u003eRandom Forest,knn:K-Nearest Neighbors,svc:Support Vector Classification,mlp:Multi-Layer Perceptron,gb:Gradient Boosting,dt:Decision Tree,xgb:eXtreme Gradient Boosting,lr\u003c/strong\u003e\u003cstrong\u003e：\u003c/strong\u003e\u003cstrong\u003eLogistic Regression\u003c/strong\u003e\u003cstrong\u003e）\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 4: Confusion matrix heatmap of various models.\u0026nbsp;\u003c/strong\u003eThis figure depicts the respective performances of each model on a test set that has exclusively been used for the classification of chest pain data. The columns in the confusion matrix represent the true labels (True Label) of the data, whereas the rows give information about the labels as predicted by the algorithm, Predicted Label. In these confusion matrices, the color intensity represents the frequency of the prediction result. The darker it is, the higher the count. These confusion matrices can visually check the accuracy and error rate of each algorithm in classifying both positive and negative samples.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e(1: Acute myocardial infarction, 0: Non-acute myocardial infarction).\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTable 2.Performance Evaluation of Various Machine Learning Models on the Dataset of Young and Middle-aged Myocardial Infarction Patients.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd rowspan=\"2\" style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAlgorithm Model\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd colspan=\"5\" style=\"width: 461px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eEvaluation Metrics\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAccuracy\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003ePrecision\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRecall\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eF1-Score\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eAUC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eRF\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.942\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.965\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eKNN\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.690\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.70\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.62\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.700\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eSVC\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.929\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.946\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eMLP\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.890\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.83\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.938\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eGB\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.942\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.968\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eDT\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.935\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.87\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.93\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.90\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.949\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eXGBoost\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.948\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.91\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.91\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003e0.973\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e\u003cstrong\u003eLR\u003c/strong\u003e\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.916\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.88\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd style=\"width: 92px;\"\u003e\n \u003cp\u003e0.942\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003eTable 3: Specific Performance Metrics of the XGBoost Model for Each Class on the Test Set.\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eF1_score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eSupport\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.86\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.97\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e127\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eMacro avg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eWeighted avg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u0026nbsp;(1: acute myocardial infarction, 0: non-acute myocardial infarction).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.3 Feature Importance Analysis\u0026nbsp;\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eUsing the optimized XGBoost model, we built a SHAP interpreter for carrying out feature importance analysis, as shown in Figure 5. SHAP values quantify how much each feature contributes to the model\u0026apos;s predictions and thus allow us to identify the top 20 features which may have the most influence on the prediction outcomes for the MI patients. This ranking not only indicates the important factors in the decision of the model but also lays the foundation for identifying significant biomarkers in clinical practice. To further study the specific contribution of each feature to the prediction result of each sample, a feature heatmap was plotted (Figure 6).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 5. SHAP value analysis of feature importance rank for Top 20 Features in the XGBoost model\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 5A shows the bar plot of the mean SHAP values of the top 20 features. The bars in blue color represent the average SHAP value of each feature, ranking their relative importance to the model prediction output. The three most critical features, according to the chart, are \u0026quot;Max cTnI\u0026quot;-maximum cardiac troponin I concentration, \u0026quot;Max BNP\u0026quot;-maximum brain natriuretic peptide concentration, and \u0026quot;SO-to-FMC\u0026quot;-the time from symptom onset to first medical contact. Other important ones include CK-MB creatine kinase-MB at admission, Smk Hx-smoking history, body mass index, and Adm SCr-creatinine level at admission.\u003c/p\u003e\n\u003cp\u003eFigure 5B: SHAP values of the top 20 features are presented, where on the right side in this honeycomb diagram presents a specific value of each feature in impacting the prediction output for a given sample.The blue-to-red color gradient shows the range of variation in the value of the features. A high SHAP value indicates that this feature is positively contributing to the model\u0026apos;s prediction of a high risk. Low values mean otherwise. For instance, high values of \u0026quot;Max cTnI\u0026quot; and \u0026quot;Max BNP\u0026quot; are usually associated with a high risk from myocardial infarction.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 6. Heatmap of Features for Early Screening Models in Young and Middle-aged Myocardial Infarction Patients.\u0026nbsp;\u003c/strong\u003eFigure 6 illustrates the SHAP value heat map for the predicted model features, providing a detailed explanation for each sample by quantifying the specific contribution of individual features to the model\u0026apos;s prediction results. The horizontal axis represents the test set samples, while the left vertical axis indicates the importance ranking of features, sorted from highest to lowest influence. The right vertical axis visualizes feature influence, with color depth reflecting the magnitude of SHAP values; darker colors correspond to larger absolute SHAP values, indicating more significant feature impact on model predictions. The top area visualizes the model\u0026apos;s prediction results based on these feature values. This analysis corroborates that Max cTnI, Max BNP, and SO-to-FMC are the three most influential features in the model.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.4 Interaction Between Features and Key Feature Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo further explore feature-to-feature interactions, we computed the interaction values of SHAP. Computing the mean absolute values for every paired value of feature-to-feature interactions and plotting these results in a heat map (Figure 7A) can help us identify both the magnitude and the patterns of feature interactions very well. First, to give an intuition of the interactive relationships among the primary features, we drew an interaction summary diagram for the first seven features. Figure 7B: There are significant interactions among some combinations of features for predicting AMI, such as interaction between Max cTnI and Adm BNP. A dependence plot was constructed to further investigate the interaction of features for Max cTnI and Adm BNP (Figure 8). Finally, to understand the impact of the contribution of each feature on model output, a SHAP force diagram was used (Figure 9).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 7. Heatmap of Feature Interaction Values and Major Feature Interactions.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 7A presents a heat map illustrating the interaction values among all features, where the color of each grid indicates the absolute average value of the interaction between corresponding feature pairs, with values ranging from 0 to 0.15.\u0026nbsp;\u003c/strong\u003eThe depth of color in the figure corresponds to the strength of the interaction between features; darker colors signify more substantial interaction effects. This heat map reveals significant interactions among certain feature combinations in predicting AMI, notably between Max cTnI and Adm BNP.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 7B illustrates the interactions among the first seven features.\u0026nbsp;\u003c/strong\u003eThese characteristics include: Max cTnI (maximum troponin I level), Max BNP (maximum B-type natriuretic peptide level), SO-to-FMC (time between the onset of symptoms and first medical contact), Adm CKMB (admission muscle acid kinase isoenzyme level), Smk Hx (smoking history), BMI (body mass index), and Adm SCr (serum creatinine level upon admission). Each site in the figure represents the influence of a single feature on the model output, while the color coding reflects the value of a specific feature.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 8. SHAP Dependence Plot Demonstrating Interaction of Max cTnI and Adm BNP.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eFigure 8A provides an overview of the general trend in the SHAP value for Max cTnI in relation to its concentration, in an attempt to develop a sense about the impact that this variable can have independently on the model prediction result. From the figure, one can clearly see that as the values of Max cTnI go up, the SHAP value continues increasing and shows its critical role regarding myocardial infarction prediction.\u003c/p\u003e\n\u003cp\u003eThe impact of Adm BNP is represented through the color gradient of Figure 8B, whose value regulates the Max cTnI SHAP. The proximity to red indicates higher values for Adm BNP, while the cooler colors reflect the low values. High Adm BNP values are associated with high SHAP values for Max cTnI; this would mean that Adm BNP increased the explanatory strength of Max cTnI in the model.\u003c/p\u003e\n\u003cp\u003eFigure 8C: SHAP main effect value of Max cTnI. The plot here strengthens the position of Max cTnI as an important predictor, since from the results, it can be observed that above the threshold value of Max cTnI, the SHAP main effect value is high, showing its importance to predict a patient prognosis at an elevated level.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide)\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 9: SHAP force plot displaying the feature contributions to model output.\u003c/strong\u003e Each feature contribution can be seen using a red color, which shows the positive contribution, and blue represents the negative contribution. The length of each bar is proportional to the magnitude of the contribution of the respective feature to the final predicted value. Figure 9A shows the average strength of a feature\u0026apos;s influence across the test set as a whole and serves well in depicting, on average, the relative importance of various features to the model output. The prominent features are Max BNP and Adm CKMB among others. Figure 9B emphasizes that features such as Max BNP and Adm CKMB have a great impact, whereas Figure 9C shows the contributions of other features like BMI and Adm Myo, which have negative values in the model. Decomposition further reinforces our presented analysis of feature importance for showing the different role played by the features in various sample categories.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Adm Myo：Admission Myoglobin）\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003e3.5 Feature Reduction and Model Optimization\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eTo reduce the bad effect of redundant features on the classification results and further apply the model easier in clinical practice in the future, feature reduction is conducted. Based on the SHAP analysis explained above, we got an importance ranking of 30 input features, thus carrying out the feature selection. We recorded and visualized the AUC values of the XGBoost model with optimized parameters running on varying numbers of features. This is shown in Figure 10(A). It can be observed from the analysis that the optimal classification performance of the XGBoost classifier can be achieved using the three most important features: emergency maximum troponin I [Max cTnI], emergency maximum BNP [Max BNP], and the time from symptom onset to first medical care [SO-to-FMC]. Based on these three features, we decided on this XGBoost model as our final model after consultation with clinicians. We plot the ROC on the test set for this final model, whose AUC is 0.979, shown in Figure 10(B). In addition, detailed performance metrics of the final model on the test set are given in Table 4, which further confirm the excellent predictive performance of this model.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 10: Analysis of feature parsimony versus model performance.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e(A) Line graph of feature parsimony illustrating AUC values of the XGBoost model with variation in a number of selected features. The vertical axis is for AUC value while the horizontal axis is for the number of selected features. This graph shows how feature selection affects the performance of the model.\u003c/p\u003e\n\u003cp\u003e(B) ROC Curve of XGBoost Model Based on Three Most Important Features.Showing Classification Performance of Model on Test Set. The shape of this curve and the corresponding AUC values below provide a critical basis for evaluating its accuracy.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eTABLE 4: Classification Performance Metrics of the Final XGBoost Model\u003c/strong\u003e\u003c/p\u003e\n\u003ctable border=\"1\" cellspacing=\"0\" cellpadding=\"0\"\u003e\n \u003ctbody\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003ePrecision\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eRecall\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eF1_score\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eSupport\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.81\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.85\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e28\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e1\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.98\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.96\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e127\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eAccuracy\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e\u0026nbsp;\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eMacro avg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.89\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.92\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.91\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003ctr\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003eWeighted avg\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.95\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e0.94\u003c/p\u003e\n \u003c/td\u003e\n \u003ctd valign=\"top\" style=\"width: 111px;\"\u003e\n \u003cp\u003e155\u003c/p\u003e\n \u003c/td\u003e\n \u003c/tr\u003e\n \u003c/tbody\u003e\n\u003c/table\u003e\n\u003cp\u003e\u003cstrong\u003e3.6 Subgroup Analysis\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eWe conducted a subgroup analysis in order to review and study the effects that age has on different groups as it comes to perceived feature importance. The dataset includes both young and middle-aged patients, divided into two groups. The young group consists of \u0026le; 44-year-olds, which accounts for 213 cases (185 myocardial infarction, 28 no myocardial infarction), while the middle-aged group consists of people aged between 45 to 59 years old, totaling 559 cases (455 myocardial infarction, 104 no myocardial infarction). We hypothesize that the importance of features in early screening for AMI may vary with age. To confirm that, we tested the XGBoost model with 30 features of young and middle-aged groups. The ROC curve performance for this model in various age groups is given in Figure 11. Figure 12 presents feature importance ranks of young and middle-aged groups based on SHAP value evaluations and their contribution to model output. Afterwards, we compared the average SHAP values along with the distribution of each feature across the two age groups to understand the relative contributions of different features in predicting myocardial infarction.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 11.ROC Curve Performance of the XGBoost Model Across Different Age Groups.\u003c/strong\u003eFigure 11 illustrates the ROC curve obtained from the XGBoost model with 30 features in classifying the young group (Figure A) and the middle-aged group (Figure B). The ROC curves of the two models are also almost perfect, reaching an AUC close to 1; thus, the AUCs of the young and middle-aged groups are 0.993 and 0.996, respectively. Besides, these results indicate that the XGBoost model with 30 features can make pretty good predictions about myocardial infarction for different age groups and estimate feature importance corresponding to individuals from different age brackets quite correctly. So, this model should be a good starting point for developing a SHAP Interpreter in order to check differences in feature importance arising among diverse age groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFigure 12: Feature importance distribution across age groups based on SHAP values.\u0026nbsp;\u003c/strong\u003eFigure 12 presents the ranking of feature importance for youth and middle-aged groups derived from an SHAP value evaluation and its contribution toward model output. A comparison of average SHAP values and distribution of various features in both age groups brings out the relative contributions of various features in predicting myocardial infarction. Parts A and C show the average value of each feature\u0026apos;s contribution to explain the output of the model, while B and D depict the distribution of the SHAP values, where color is used to emphasize the direction and magnitude of feature contributions with regard to a prediction. Max cTnI, Max BNP, admission CKMB, and SO-to-FMC were significant contributors in both age cohorts. However, other predictors displayed differential effects with age, especially for BMI, whose impact on the prediction outcomes was stronger in the middle-aged group compared to youth.\u003c/p\u003e\n\u003cp\u003e（Max cTnI：Maximum Cardiac Troponin I,Max BNP：Maximum B-type Natriuretic Peptide,SO-to-FMC：Symptom Onset to First Medical Contact,Adm CKMB：Admission Creatine Kinase-MB,Smk Hx：Smoking History,BMI：Body Mass Index,Adm SCr：Admission Serum Creatinine,Age at onset：Age at Onset,Min LVEF%：Minimum Left Ventricular Ejection Fraction Percentage,Adm Myo：Admission Myoglobin,Adm BNP：Admission B-type Natriuretic Peptide,Adm Tn：Admission Troponin,Hx of HTN：History of Hypertension,Adm D-dimer：Admission D-dimer,Hx of HLD：History of Hyperlipidemia,Adm SBP：Admission Systolic Blood Pressure,Adm HR：Admission Heart Rate）\u003c/p\u003e"},{"header":"DISCUSSION","content":"\u003cp\u003eIn recent decades, the incidence of AMI has significantly risen among young people. It is now an important concern in terms of worldwide public health compared to earlier decades \u003csup\u003e[22-24]\u003c/sup\u003e. While typically perceived to occur in older adults, cases among younger patients are strikingly increasing. Over 30,000 young women under the age of 55 years are now admitted every year in the United States, which reveals how serious and urgent this condition needs to be taken as\u003csup\u003e\u0026nbsp;[22]\u003c/sup\u003e. The increase in the incidence of AMI in younger populations is related mainly to the changes in the lifestyle and increase in the traditional risk factors, such as metabolic syndrome and diabetes, while alternative risk factors \u0026nbsp;are gaining increasing recognition as major determinants of cardiovascular health in young people. This trend puts not only the health of younger subjects at risk but also increases the medical and economic burden on society. In view of an increasing incidence of AMI in patients of young age group, most of the studies relating to AMI have been conducted in elderly populations, and this has resulted in a lack of comprehensive understanding regarding characteristics, causes, and needs of the young patients with AMI \u003csup\u003e[3]\u003c/sup\u003e. Given the growing trends of AMI in younger age groups and limitations identified among previous studies, this represents a pressing priority to improve research into younger age groups in order to bridge the knowledge gap. The early development of AMI screening tools will be of importance in relation to young and middle-aged individuals, as this will help the healthcare provider understand who is at high risk and thus develop preventive measures that will reduce incidences of AMI and its complications.\u003c/p\u003e\n\u003cp\u003eWe modeled the dataset using the following and compared the performance of eight machine learning algorithms. Then, for each of the chosen algorithms, we used a grid search and performed 5-fold cross-validation to optimize the parameters, after which we measured their performance on the test set. Among these algorithms, we have identified the XGBoost algorithm for the best performance for chest pain data processing. This model has an AUC of 0.92, accuracy of 0.88, precision rate of 0.85, recall rate of 0.87, and F1 score of 0.86. For understanding and further optimization of the XGBoost model, we used the SHAP method to rank the feature importances with calculation of the Shapley value. Feature selection: In this regard, we examined the SHAP value analysis and selected three critical features in the prediction of young and middle-aged AMI patients. These features included emergency maximum troponin I, emergency maximum BNP, and the time from symptom onset to first medical care. Based on this, we developed the three features-based XGBoost model and gave its performance, stating that this model performed with an AUC value of 0.90 on the test set with an accuracy of 0.87, a precision of 0.84, a recall of 0.86, and an F1 score of 0.85, thus proving the effectiveness and reliability in clinical applications.\u003c/p\u003e\n\u003cp\u003eSHAP value analysis for feature importance assessment identified three biomarkers, which are the most important in early screening of AMI among young and middle-aged people, and all three biomarkers are of important clinical significance in early diagnosis. As a myocardial-specific marker, the elevation of cTnI reflects the degree of damage to the myocardial cells and has been regarded as the gold standard for diagnosing AMI\u003csup\u003e\u0026nbsp;[25-26]\u003c/sup\u003e. BNP is a sensitive indicator of heart failure. When BNP is elevated, increased ventricular wall tension and impaired cardiac function are indicated. It acts as an independent predictor of prognosis in AMI \u003csup\u003e[27]\u003c/sup\u003e. SO-to-FMC reflects the timeliness of treatment and is closely related to the prognosis of AMI \u003csup\u003e[28]\u003c/sup\u003e. This study originally integrates the three indicators into an early screening model for AMI among young and middle-aged individuals, using machine learning methods to calculate the relative importance. The approach may provide valued information in the early identification of AMI and risk stratification during clinical practice, hence bringing new ideas and tools to clinicians.\u003c/p\u003e\n\u003cp\u003eWe then compute the SHAP interaction values for all interactions of features and observe huge variation in both magnitude and pattern of interaction effects between each pair of features in predicting AMI. The heat map shows that the highest interaction effect is between Max cTnI and Adm BNP, underlining that these biomarkers are clinically relevant when interacting with each other. In particular, the dependence plot reflects that the modulatory effect of Adm BNP levels on myocardial infarction prediction is strongly increased at higher concentrations of Max cTnI, which also indicates that the combination of both predictors allows more exact risk predictions concerning myocardial infarction in patients at hospital admission. SHAP visualizes a given individual characteristic contribution to model prediction results. Force plot of features: Adm BNP and Adm CKMB contribute positively to predict the AMI, while BMI and Adm Myo make negative contributions in this particular case. These findings have important implications for clinical practice: clinicians should be particularly interested in combined assessment of Max cTnI and Adm BNP as a means for rapid assessment of patients for myocardial infarction risk in the emergency setting, thus helping optimize treatment decisions and resource distribution. Moreover, the negative contribution of BMI and Adm Myo, especially in specific patient populations, needs to be recognized because these can decrease overall predictive accuracy.\u003c/p\u003e\n\u003cp\u003eTo further analyze the dependence of feature importance on age, we performed subgroup analysis. We divided the patients into a young group ≤44 years old (n = 213) and a middle-aged group aged 45-59 years old (n = 559). XGBoost models were developed for both groups, followed by SHAP value analysis. Both models showed very excellent predictive capability: the AUCs were 0.993 for the young group and 0.996 for the middle-aged group. Therefore, key predictors that could be noted in both the cohorts were Max cTnI, Max BNP, and SO-to-FMC by comparing the ranking of feature importance between the two groups. However, in the case of other variables such as BMI, the impact was relatively more important in the middle-aged group compared with that in the young group. According to this analysis, an age-stratified approach should be performed in the strategy for early AMI screening, developing differentiated assessment standards and intervention measures for different age groups. The history of smoking is much more important in the younger group; therefore, the most attention must be paid to lifestyle management and enforcement of anti-smoking interventions and other harmful behaviors. Dominant age-related factors in the middle-aged group mean that smoking cessation, weight control, and active treatment of comorbidities such as hypertension and diabetes should be a point of concern. This differential approach to management may allow for better prognosis in the patients of all age groups.\u003c/p\u003e\n\u003cp\u003eThis study provides a quite accurate prediction model for the young and middle-aged to conduct early screening for AMI. Yet, some limitations do exist, and we do acknowledge these. This was a single-center investigation, and while the sample size is sufficient to support our conclusions, increasing it can provide stronger robustness of the results. Although the model showed very good performance when internally validated, an independent validation dataset is not available, thus our testing of generalization performance across diverse populations is limited. Despite the best efforts, unidentified selection biases exist. The model, incorporating many clinically significant features, is the result of work presented herein; future studies may also consider adding new biomarkers or comprehensive imaging features to enhance the performance. These limitations thus provide some avenues into which future research may focus: multi-center validation, expansion of sample size, and inclusion of extended clinical characteristics.\u003c/p\u003e"},{"header":"Conclusion","content":"\u003cp\u003eThis study successfully developed an early screening model of high accuracy for AMI in young and middle-aged people using machine learning techniques. The result manifested that Max cTnI, Max BNP, and SO-to-FMC were the three most significant factors among the list of predictors of AMI in this age group, while the model performed well on the test set with an AUC of 0.979. Inter-features interaction analysis showed the complicated interaction between different biomarkers. Subgroup analysis also demonstrated distinct predictive patterns of AMI among different age groups, which means risk assessment should be necessary for individualized study. Although these results are promising, further validation through a multicenter external study and from prospective research is required to establish the generalizability and clinical applicability of the model.\u003c/p\u003e"},{"header":"Declarations","content":"\u003cp\u003e\u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe 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. This study was conducted in accordance with the Declaration of Helsinki (as revised in 2013) and was approved by the hospital ethics committee (approval number: 2024-KY-26-02). The requirement for informed consent was waived due to the retrospective nature of the study and the use of anonymized data from routine clinical practice.\u003cstrong\u003eClinical trial number: not applicable.\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConsent for publication\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis manuscript does not contain any potentially identifiable images or data. Due to the retrospective nature of the study and the use of anonymized data from routine clinical practice, the requirement for informed consent for publication was waived.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAvailability of data and materials\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAdditional data and materials are available from the corresponding author on reasonable request.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eCompeting interests\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eAll authors have completed the ICMJE uniform disclosure form. The authors have no conflicts of interest to declare.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eFunding\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThis research was supported by the Research Funds of Joint Research Center for Occupational Medicine and Health of IHM (Nos. OMH-2023-05 and OMH-2023-24), the Medical Special Cultivation Project of Anhui University of Science and Technology (No. YZ2023H2A007), the National Natural Science Foundation of China (Nos. 52374155 and 61806006), and the Natural Science Research Project of Colleges and Universities in Anhui Province (No. 2022AH040113).\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAuthors' contributions\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eQingqing Ruan and Shuzhi Su contributed equally to this work. Qingqing Ruan, Shuzhi Su, and Zengyong Qiao conceived and designed the experiments. Xian Wang and Xiumei Li performed the experiments. Yong Dai and Zengyong Qiao analyzed the data. Qingqing Ruan and Shuzhi Su wrote the paper. All authors contributed to editorial changes in the paper.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eAcknowledgements\u003c/strong\u003e\u003c/p\u003e\n\u003cp\u003eThe authors would like to acknowledge all individuals and institutions that contributed to this work. Special thanks to the Joint Research Center for Occupational Medicine and Health of IHM for their support in providing resources and technical assistance. We also appreciate the contributions of the team at the Department of Cardiovascular Medicine at the Sixth People’s Hospital South Campus, Shanghai Jiao Tong University School of Medicine, for their general support.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\n \u003cli\u003eR. D. Barnett. Acute myocardial infarction. The Lancet. 2019;393 (10191):2580-2580. doi:10.1016/s0140-6736(19)31419-9\u003c/li\u003e\n \u003cli\u003eShukla AN,Jayaram AA,Doshi D, et al. The Young Myocardial Infarction Study of the Western Indians: YOUTH Registry. Glob Heart. 2019;14 (1):27-33. doi:10.1016/j.gheart.2018.12.001\u003c/li\u003e\n \u003cli\u003eKrittanawong C,Khawaja M,Tamis-Holland JE, et al. Acute Myocardial Infarction: Etiologies and Mimickers in Young Patients. J Am Heart Assoc. 2023;12 (18):e029971. doi:10.1161/JAHA.123.029971\u003c/li\u003e\n \u003cli\u003eIvan Hanson,Akash Rusia,Andres Palomo, et al. Treatment of Acute Myocardial Infarction and Cardiogenic Shock: Outcomes of the RECOVER III Postapproval Study by Society of Cardiovascular Angiography and Interventions Shock Stage. Journal of the American Heart Association. 2024;0 (0):0-0. doi:10.1161/jaha.123.031803\u003c/li\u003e\n \u003cli\u003eViola Vaccarino. Myocardial Infarction in Young Women. Circulation. 2019;139 (8):1057-1059. doi:10.1161/circulationaha.118.039298\u003c/li\u003e\n \u003cli\u003eMarcin Ambroziak,Jennifer Franke,Anna W\u0026oacute;jcicka, et al. SYNE1 gene novel variant is associated with myocardial infarction in young people with a family history of premature atherosclerosis. European Heart Journal. 2023;44 (Supplement_2):0-0. doi:10.1093/eurheartj/ehad655.3263\u003c/li\u003e\n \u003cli\u003eYasmina Benchabi. Is hyperhomocysteinemia a risk factor for myocardial infarction in young people?. Archives of Cardiovascular Diseases Supplements. 2022;14 (1):26-26. doi:10.1016/j.acvdsp.2021.09.060\u003c/li\u003e\n \u003cli\u003eZhang P,Wu L,Zou TT, et al. Machine Learning for Early Prediction of Major Adverse Cardiovascular Events After First Percutaneous Coronary Intervention in Patients With Acute Myocardial Infarction: Retrospective Cohort Study. JMIR Form Res. 2024;8:e48487. doi:10.2196/48487\u003c/li\u003e\n \u003cli\u003eKhera R,Haimovich J,Hurley NC, et al. Use of Machine Learning Models to Predict Death After Acute Myocardial Infarction. JAMA Cardiol. 2021;6 (6):633-641. doi:10.1001/jamacardio.2021.0122\u003c/li\u003e\n \u003cli\u003eWang S,Li J,Sun L, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction. BMC Med Inform Decis Mak. 2021;21 (1):301. doi:10.1186/s12911-021-01667-8\u003c/li\u003e\n \u003cli\u003eImproving the diagnosis of myocardial infarction with machine learning. Nat Med. 2023;29 (5):1070-1071. doi:10.1038/s41591-023-02331-6\u003c/li\u003e\n \u003cli\u003eSuhuai Wang,Jingjie Li,Li Sun, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction. BMC Medical Informatics and Decision Making. 2021;21 (1):0-0. doi:10.1186/s12911-021-01667-8\u003c/li\u003e\n \u003cli\u003eSang Hoon Lee,Sulyun Lee,W. Nick Street, et al. Machine learning approaches to predict the 1-year-after-initial-AMI survival of elderly patients. BMC Medical Informatics and Decision Making. 2022;22 (1):0-0. doi:10.1186/s12911-022-01854-1\u003c/li\u003e\n \u003cli\u003eThan MP,Pickering JW,Sandoval Y, et al. Machine Learning to Predict the Likelihood of Acute Myocardial Infarction. Circulation. 2019;140 (11):899-909. doi:10.1161/CIRCULATIONAHA.119.041980\u003c/li\u003e\n \u003cli\u003eMcCord J,Gibbs J,Hudson M, et al. Machine Learning to Assess for Acute Myocardial Infarction Within 30 Minutes. Crit Pathw Cardiol. 2022;21 (2):67-72. doi:10.1097/HPC.0000000000000281\u003c/li\u003e\n \u003cli\u003eJahmunah V,Ng EYK,San TR, et al. Automated detection of coronary artery disease, myocardial infarction and congestive heart failure using GaborCNN model with ECG signals. Comput Biol Med. 2021;134:104457. doi:10.1016/j.compbiomed.2021.104457\u003c/li\u003e\n \u003cli\u003eMandair D,Tiwari P,Simon S, et al. Prediction of incident myocardial infarction using machine learning applied to harmonized electronic health record data. BMC Med Inform Decis Mak. 2020;20 (1):252. doi:10.1186/s12911-020-01268-x\u003c/li\u003e\n \u003cli\u003eLi X,Shang C,Xu C, et al. Development and comparison of machine learning-based models for predicting heart failure after acute myocardial infarction. BMC Med Inform Decis Mak. 2023;23 (1):165. doi:10.1186/s12911-023-02240-1\u003c/li\u003e\n \u003cli\u003eAustin DE,Lee DS,Wang CX, et al. Comparison of machine learning and the regression-based EHMRG model for predicting early mortality in acute heart failure. Int J Cardiol. 2022;365:78-84. doi:10.1016/j.ijcard.2022.07.035\u003c/li\u003e\n \u003cli\u003eLi Y,Hu Y,Jiang F, et al. Combining WGCNA and machine learning to identify mechanisms and biomarkers of ischemic heart failure development after acute myocardial infarction. Heliyon. 2024;10 (5):e27165. doi:10.1016/j.heliyon.2024.e27165\u003c/li\u003e\n \u003cli\u003eZhang Q,Burrage MK,Shanmuganathan M, et al. Artificial Intelligence for Contrast-Free MRI: Scar Assessment in Myocardial Infarction Using Deep Learning-Based Virtual Native Enhancement. Circulation. 2022;146 (20):1492-1503. doi:10.1161/CIRCULATIONAHA.122.060137\u003c/li\u003e\n \u003cli\u003eChandrasekhar J,Gill A,Mehran R. Acute myocardial infarction in young women: current perspectives. Int J Womens Health. 2018;10:267-284. doi:10.2147/IJWH.S107371\u003c/li\u003e\n \u003cli\u003e\u0026Aring;berg ND,Adiels M,Lindgren M, et al. Diverging trends for onset of acute myocardial infarction, heart failure, stroke and mortality in young males: role of changes in obesity and fitness. J Intern Med. 2021;290 (2):373-385. doi:10.1111/joim.13285\u003c/li\u003e\n \u003cli\u003eEnas EA,Varkey B,Dharmarajan TS, et al. Lipoprotein(a): An underrecognized genetic risk factor for malignant coronary artery disease in young Indians. Indian Heart J. 2019;71 (3):184-198. doi:10.1016/j.ihj.2019.04.007\u003c/li\u003e\n \u003cli\u003eChen Y,Tao Y,Zhang L, et al. Diagnostic and prognostic value of biomarkers in acute myocardial infarction. Postgrad Med J. 2019;95 (1122):210-216. doi:10.1136/postgradmedj-2019-136409\u003c/li\u003e\n \u003cli\u003eWang Y. Immune-related biomarkers in myocardial infarction; diagnostic/prognostic value and therapeutic potential. J Biochem Mol Toxicol. 2023;37 (12):e23489. doi:10.1002/jbt.23489\u003c/li\u003e\n \u003cli\u003ePuelacher C,Wagener M,Honegger U, et al. Combining high-sensitivity cardiac troponin and B-type natriuretic peptide in the detection of inducible myocardial ischemia. Clin Biochem. 2018;52:33-40. doi:10.1016/j.clinbiochem.2017.10.014\u003c/li\u003e\n \u003cli\u003eChaulin AM. Cardiac Troponins Metabolism: From Biochemical Mechanisms to Clinical Practice (Literature Review). Int J Mol Sci. 2021 Oct 10;22(20):10928. doi: 10.3390/ijms222010928. PMID: 34681585; PMCID: PMC8535601.\u003c/li\u003e\n\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":false,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"bmc-cardiovascular-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcar","sideBox":"Learn more about [BMC Cardiovascular Disorders](http://bmccardiovascdisord.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcar/default.aspx","title":"BMC Cardiovascular Disorders","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true},"keywords":"Acute Myocardial Infarction, Interpretable Prediction Model, Machine Learning, XGBoost, Early Screening","lastPublishedDoi":"10.21203/rs.3.rs-5614054/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-5614054/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003cp\u003e\u003cstrong\u003eBackground: \u003c/strong\u003eIn recent years, the incidence of acute myocardial infarction (AMI) has been rising among young individuals. However, existing research predominantly concentrates on AMI patients who are elderly. This study employs machine learning models to analyze multidimensional clinical features, with the objective of developing an accurate early screening tool for AMI in young and middle-aged populations.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eMethods: \u003c/strong\u003eWe analyzed data from 772 young and middle-aged patients who visited the Chest Pain Center at the South Campus of Shanghai Sixth People's Hospital between January 2018 and April 2024. This cohort included 640 patients diagnosed with AMI and 132 patients with non-AMI conditions. We optimized model parameters and evaluated the performance of eight machine learning algorithms. The SHAP (SHapley Additive exPlanations) method was employed to analyze feature importance and conduct feature screening to identify the optimal model. Additionally, we performed age-stratified SHAP analysis to investigate variations in feature importance across different age groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eResults:\u003c/strong\u003e Among the eight machine learning models evaluated, the eXtreme Gradient Boosting (XGBoost) model exhibited the highest performance, achieving an AUC of 0.973. Utilizing the ranking of SHAP feature importance, a refined three-feature XGBoost model was developed, which demonstrated an improved AUC of 0.979. The final selected features included: the maximum emergency troponin value (Max cTnI), the maximum emergency BNP (Max BNP), and the duration from symptom onset to first medical treatment (SO-to-FMC). Subgroup analysis revealed variations in feature importance across different age groups.\u003c/p\u003e\n\u003cp\u003e\u003cstrong\u003eConclusion:\u003c/strong\u003eThis study developed and validated a machine learning model using XGBoost for the early screening of AMI in young and middle-aged individuals, demonstrating high predictive accuracy and excellent interpretability, thereby making it suitable for diverse age cohorts within these populations.\u003c/p\u003e","manuscriptTitle":"Development and Validation of an Interpretable Prediction Model for Early Screening of Acute Myocardial Infarction in Young and Middle-Aged Patients","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2024-12-24 17:18:13","doi":"10.21203/rs.3.rs-5614054/v1","editorialEvents":[{"type":"communityComments","content":0},{"type":"decision","content":"Revision requested","date":"2024-12-19T09:28:46+00:00","index":"","fulltext":""},{"type":"editorAssigned","content":"","date":"2024-12-19T08:39:09+00:00","index":"","fulltext":""},{"type":"checksComplete","content":"","date":"2024-12-18T09:50:16+00:00","index":"","fulltext":""},{"type":"submitted","content":"BMC Cardiovascular Disorders","date":"2024-12-10T07:11:56+00:00","index":"","fulltext":""}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"bmc-cardiovascular-disorders","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":false,"externalIdentity":"bcar","sideBox":"Learn more about [BMC Cardiovascular Disorders](http://bmccardiovascdisord.biomedcentral.com/)","snPcode":"","submissionUrl":"https://www.editorialmanager.com/bcar/default.aspx","title":"BMC Cardiovascular Disorders","twitterHandle":"BMC_series","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"em","reportingPortfolio":"BMC Series","inReviewEnabled":true,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"f13fe82c-07a4-4669-af64-9ac04e9cb04a","owner":[],"postedDate":"December 24th, 2024","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"in-revision","subjectAreas":[],"tags":[],"updatedAt":"2026-03-05T10:25:14+00:00","versionOfRecord":[],"versionCreatedAt":"2024-12-24 17:18:13","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-5614054","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-5614054","identity":"rs-5614054","version":["v1"]},"buildId":"qtupq5eGEP_6zYnWcrvyt","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}