Predicting the risk of mechanical complications in acute myocardial infarction using an interpretable machine learning model

Predicting the risk of mechanical complications in acute myocardial infarction using an interpretable machine learning model | 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 Predicting the risk of mechanical complications in acute myocardial infarction using an interpretable machine learning model Yajie Chang, Jiahui Yu, Yang Liu, Guangyuan Chen, Xiaoxu Duan, and 4 more This is a preprint; it has not been peer reviewed by a journal. https://doi.org/ 10.21203/rs.3.rs-8461400/v1 This work is licensed under a CC BY 4.0 License Status: Posted Version 1 posted You are reading this latest preprint version Abstract Background Mechanical complications in acute myocardial infarction (AMI) progress rapidly and carry very high mortality, underscoring the need for early risk prediction. Existing studies use a narrow range of models with poorly characterized decision-making. This study aimed to develop and validate an interpretable prediction model for post-AMI mechanical complications to guide individualized treatment and optimize resource allocation. Methods A total of 850 patients with and without mechanical complications enrolled in this study. 58 features were selected for model training and validation. Eight machine learning algorithms were used to build prediction models, whose performance was assessed AUC, accuracy, F1 score and other indicators. The SHAP method was applied to rank feature importance and interpret the final model. Results Among the eight machine learning models, LightGBM showed the best discriminative performance. After feature reduction, a 13-variable interpretable LightGBM model was established, achieving excellent discrimination for mechanical complications in the validation set (AUC = 0.9587). SHAP analysis identified age, hsCRP (High-Sensitivity C-Reactive Protein), drinking history, D-dimer, sex, NEUT (Neutrophils), PCI (Percutaneous Coronary Intervention), MPV (Mean Platelet Volume), history of chronic lung disease, SGLT2i (Sodium-Glucose Cotransporter 2 Inhibitors) use, LDH (Lactate Dehydrogenase), AMI category, and MA (Malignant Arrhythmia) as the most influential predictors. Conclusions The interpretable ML model provides both global and patient-level explanations, and the simplified key-feature model is suitable for deployment as a clinical decision-support tool for rapid screening and risk assessment in emergency settings. Figures Figure 1 Figure 2 Figure 3 Figure 4 Figure 5 Figure 6 Figure 7 Figure 8 Introduction Mechanical complications in AMI typically refer to spontaneous myocardial rupture occurring after acute myocardial infarction. Based on the site of rupture, mechanical complications in AMI can be classified into free wall rupture (FWR), ventricular septal rupture (VSR), papillary muscle rupture (PMR), and other types such as pseudoaneurysm and true aneurysm [ 1 ]. The widespread application of reperfusion therapy has significantly reduced the incidence of mechanical complications [ 1 – 2 ]. Unfortunately, the associated mortality rate shown no significant decline over the past two decades; the in-hospital mortality risk for patients with mechanical complications in AMI is more than fourfold higher than that of non-mechanical complications counterparts [ 1 , 3 ]. Hence, although mechanical complications in AMI are relatively uncommon, they remain a critical determinant of clinical outcomes. Transthoracic echocardiography is the primary tool for diagnosing mechanical complications in AMI. However, surgical exploration or autopsy remains the gold standard for confirmation [ 4 ]. Imaging examinations typically detect mechanical complications only after the event has occurred, making it difficult to identify pre-rupture signs and thereby limiting the intervention window. The establishment of predictive models based on electronic medical records (EMR) have attracted increasing attention in recent years. Previous studies have identified common clinical characteristics of patients with mechanical complications in AMI, including advanced age, female sex, a history of heart failure or chronic kidney disease, and delayed reperfusion [ 1 , 5 ]. However, some studies rely heavily on traditional statistical models [ 6 ], which only characterize linear relationships between features and outcomes. Given the complex pathophysiology of mechanical complications, relying on a single biomarker is insufficient to comprehensively capture the multifaceted risk profile [ 7 ]. Machine learning (ML), a core branch of artificial intelligence, can efficiently integrate multi-source data in clinical diagnosis, precision therapy, health management, and monitoring scenarios[ 8 – 11 ]. To the best of the author’s knowledge, studies utilizing ML models to predict the risk of mechanical complications in AMI remain limited[ 12 , 13 ]. The few available investigations have generally relied on a single machine learning algorithm[ 14 ], and due to the inherent nature of these models as "black boxes", they offer insufficient interpretability. This study aims to develop and validate an interpretable ML model for the early and accurate prediction of mechanical complications risk in AMI. Beyond predicting the risk of mechanical complications in AMI, we employ the Shapley Additive Explanations (SHAP) to achieve interpretability at both the global and individual level. By enabling early identification and dynamic monitoring of high-risk patients, this model may facilitate timely interventions, including surgical procedures, and ultimately reduce mortality related to mechanical complications. Methods Study population This retrospective, single-center cohort study was conducted at the First Affiliated Hospital of Harbin Medical University to derive and validate a predictive model. The study included AMI patients admitted to the CCU from September 2015 to August 2025, among whom 239 patients developing mechanical complications in AMI. To ensure accurate predictive performance, we randomly selected 611 patients from the same period who did not develop mechanical complications, resulting in a total of 850 patients. All patients received standardized AMI treatment in accordance with CCU protocols and clinical guidelines. The study adhered to the Declaration of Helsinki and was approved by the hospital's Ethics Committee. Due to its retrospective design and the de-identified nature of the data, the Ethics Committee waived the requirement for written informed consent. Inclusion criteria for patients with mechanical complications in AMI: (1) Age ≥ 18 years; (2) The inclusion criteria for AMI patients followed the latest guidelines for the diagnosis of acute myocardial infarction issued by the European Society of Cardiology[ 15 ]; (3) Patients admitted for AMI who had emergent cardiac catheterization or hemodynamic compromise with echocardiographic evidence of mechanical complications were included. Diagnostic criteria for mechanical complications: FWR are typically characterized by the abrupt onset of consciousness disturbance, cardiogenic shock, electrical-mechanical dissociation, and acute cardiac tamponade. For diagnosis, at least one of the following criteria must be met: (1) Transthoracic echocardiography revealing pericardial effusion > 1 cm with abnormal echogenicity and signs of pericardial tamponade; (2) Pericardiocentesis confirming hemopericardium; (3) Anatomical confirmation through surgery or autopsy. VSR presents as sudden onset heart failure or cardiogenic shock, accompanied by a new, coarse systolic murmur at the left sternal border (3rd–4th intercostal spaces), with systolic thrill in some patients. Echocardiography may show interventricular septal echo discontinuity with left-to-right shunting. PMR manifests as sudden acute left heart failure, with new systolic murmurs or exacerbation of pre-existing murmurs in the mitral valve area. Echocardiography typically shows severe mitral regurgitation or direct visualization of papillary muscle rupture[ 16 – 18 ]. Exclusion criteria: (1) Severe deficiency in clinical or laboratory data; (2) Comorbid severe infections, severe autoimmune diseases, or severe hematologic disorders; (3) mechanical complications secondary to other conditions, including infective endocarditis, cardiac tumors, myocarditis, cardiac amyloidosis, or iatrogenic injuries. Data collection and processing The model input features comprised a comprehensive set of variables, including patient demographics, initial medical contact details, cardiovascular risk factors, biochemical markers, metabolomic biomarkers, and treatment options, all collected within 48 hours of CCU admission. The data were obtained from the EMR system. To minimize potential bias, features with more than 25% missing values were excluded from subsequent analyses. For the remaining variables, continuous data were imputed using the median, while categorical data were imputed using the mode [ 19 ], to mitigate the impact of missing values on model performance. To address potential multicollinearity that could compromise predictive accuracy, we evaluated the correlation structure between features using Spearman’s rank correlation coefficient. For any pair of features with a correlation coefficient > 0.6, we retained the feature more strongly associated with the outcome and excluded the other. In addition, we conducted statistical tests for all variables and removed those that were not statistically significant. The detailed implementation procedure is described in the Statistical Analysis section. Model development and performance evaluation We employed eight ML models to predict the likelihood of mechanical complications in AMI: Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LightGBM), Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost). The original dataset was randomly split into training and validation sets in a 7:3 ratio, with stratified sampling used to address class imbalance. Five-fold cross-validation was applied during training to optimize hyperparameters for each model, and the validation set was reserved for independent performance evaluation. It is important to note that predicting mechanical complications in AMI is a binary classification task, in which the non-complication group is substantially larger than the complication group, resulting in a typical class imbalance scenario[ 20 ]. In such settings, accuracy alone is insufficient to evaluate a model’s ability to identify the minority class. Therefore, the F1 score was adopted as a comprehensive performance metric[ 21 ]. Moreover, because the primary objective of this study was to maximize the detection of positive cases (i.e., patients with mechanical complications), recall was prioritized, and precision–recall (P–R) curves were plotted for each model. This metric framework has been shown to be more sensitive to the minority class in imbalanced datasets[ 22 ]. Model performance was comprehensively evaluated and compared using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Model calibration was assessed with calibration curves and the Brier score. Clinical utility was evaluated using decision curve analysis (DCA). In addition, to examine the impact of data distribution consistency on model stability, five-fold and ten-fold cross-validation were performed within the derivation cohort. Feature selection and model interpretation Identifying the most informative variables for disease risk from a large pool of candidate features remains a major challenge in both clinical practice and data science. The SHAP framework addresses this by providing a unified approach to explain ML predictions: it quantifies each input feature’s contribution to the model output in a consistent, ordered manner, thereby mitigating the “black box” nature of ML models[ 23 ]. Using SHAP values for feature selection enables guided dimensionality reduction, allowing key features to be retained while preserving strong discriminative performance and enhancing the model’s applicability in real-world clinical settings. Differences in AUC among models were compared using the DeLong nonparametric method[ 24 ]. During feature reduction, features selected by the ML model were iteratively removed until a significant decline in AUC was observed. In this study, SHAP was applied at both global and individual levels. At the global level, the absolute magnitude of SHAP values reflects the significance of each feature in the model, thereby clarifying the overall relationship between predictors and mechanical complications risk. At the individual level, SHAP values further pinpoint the specific driving factors influencing each patient’s prediction. This dual-level interpretability strengthens transparency and explainability across both population-level trends and individual patient predictions [ 23 , 25 ]. Statistical analysis Statistical analyses were performed using Python version 3.9. Continuous variables with non-normal distributions were expressed as medians with interquartile ranges, and between-group comparisons were performed using the Mann–Whitney U test or Kruskal–Wallis H test, as appropriate. Categorical variables were presented as frequencies and percentages, with comparisons made using the chi-square test or Fisher’s exact test. The discriminatory ability of the models was evaluated by the AUC, and the optimal threshold was determined by maximizing the Youden index (sensitivity + specificity − 1). A two-sided P value < 0.05 was considered statistically significant. Results Patient characteristics This retrospective study established a derivation cohort of 850 patients to identify key predictors and develop a predictive model. Patients were grouped according to the occurrence of mechanical complications in AMI. The complete methodological workflow is illustrated in Fig. 1 . The results of the correlation analysis are presented in Fig. 2 . Strong correlations were observed between height, weight, and BMI (see Fig. 2 A). After excluding collinearity, a total of 73 variables were retained for following analysis, and a heatmap (Fig. 2 B) further confirmed the low correlations among the remaining variables. The clinical characteristics of patients with and without mechanical complications in AMI are summarized in Supplementary Table S1 . Among the 850 AMI patients included in the derivation cohort, 239 (28%) developed mechanical complications during CCU hospitalization. At the time of initial medical contact, patients in the mechanical complications group were older, with a median age of 73 years (interquartile range [IQR]: 66.5–79), compared with 64 years (IQR: 53.5–71) in the non-mechanical complications group (P < 0.001). The proportion of female patients was also higher in the mechanical complications group (60.67% vs. 30.61%). Hematological parameters also differed significantly between groups, with NEUT, LYMPH, HGB, and MPV all showing marked between-group differences (all P < 0.001), supporting a role for inflammatory and immune dysregulation in the pathogenic process. Conventional cardiovascular biomarkers were substantially elevated in the mechanical complications group, which exhibited higher cTnI, LDH, NT-proBNP, and hsCRP levels than the non-mechanical complications group (all P < 0.001). Taken together, these findings highlight the importance of early and accurate risk stratification for mechanical complications in patients with AMI. Model development and performance comparison Eight ML models were developed using clinical variables obtained within the first 48 hours after admission to predict the occurrence of mechanical complications in AMI patients during CCU hospitalization. The areas under the receiver operating characteristic curves (AUCs) for these models, along with pairwise DeLong tests assessing the statistical significance of performance differences, are shown in Fig. 3 A. LightGBM achieved the highest discriminative performance (AUC = 0.9622), followed by AdaBoost (AUC = 0.9603) and XGBoost (AUC = 0.9595). The precision–recall curves in Fig. 3 B further indicate that LightGBM maintained the most favorable performance under the imbalanced class distribution. Accuracy, sensitivity, specificity, PPV, NPV, and F1 score for each model are presented in Table 2 . Notably the optimal hyperparameters for all models were determined via cross-validation. The results indicate that the LightGBM model outperformed other models. In contrast, the MLP and RF models demonstrated relatively poorer performance. Table 2 Predictive performance metrics of the different models Model Accuracy Sensitivity Specificity PPV NPV F1 Brier XGBoost 0.9098 0.8056 0.9508 0.8657 0.9255 0.8345 0.0763 LightGBM 0.9137 0.8194 0.9508 0.8676 0.9305 0.8429 0.0702 GBM 0.8941 0.7222 0.9617 0.8814 0.8980 0.7939 0.0854 AdaBoost 0.9020 0.7639 0.9563 0.8730 0.9115 0.8148 0.1592 MLP 0.8392 0.6389 0.9180 0.7541 0.8660 0.6917 0.1306 LR 0.8667 0.8750 0.8634 0.7159 0.9416 0.7875 0.1094 SVM 0.8588 0.8194 0.8743 0.7195 0.9249 0.7662 0.1043 RF 0.8510 0.5833 0.9563 0.8400 0.8537 0.6885 0.1178 During feature reduction based on feature-importance ranking, the five best-performing ML models were selected for progressive feature elimination. As shown in Fig. 4 , the AUC trajectories indicate that the LightGBM model consistently maintained the highest predictive performance among these models. The final model was selected during the feature reduction process of the LightGBM model. As shown in Fig. 5 and Supplementary Table S2, The model with 13 features represents a clear inflection point in performance, and adding more than 13 features does not yield a significant performance gain. Compared with the 58-feature model, the 13-feature model provided a favorable net benefit across a wider range of threshold probabilities; in the decision curve, the net benefit of the LightGBM model in the test set remained consistently above the reference line, indicating good clinical utility (Supplementary Fig. 1). Moreover, the area under the precision–recall curve of the 13-feature model was only marginally lower than that of the 58-feature model, suggesting similarly high applicability in clinical practice (Supplementary Fig. 2). The calibration curve of the 13-feature model demonstrated good agreement between predicted and observed probabilities, and the relatively low Brier score further indicated the absence of substantial overfitting (Supplementary Fig. 3and 4). Evaluation of the optimal model Finally, we focused on the 13-feature LightGBM model, which included the following predictors: age, hsCRP, history of alcohol use, D-dimer, sex, NEUT, PCI, MPV, history of chronic pulmonary disease, SGLT2i use, LDH, AMI type, and MA. This 13-feature model was retained as the final model for subsequent analyses. The LightGBM model achieved an AUC of 0.9587 for predicting mechanical complications in AMI, demonstrating excellent discrimination between cases with and without mechanical complications. Overall classification performance was robust, with an accuracy of 0.898 and an F1 score of 0.8116, indicating a good balance between precision and recall. The sensitivity was 0.7779, reflecting a strong ability to identify patients with mechanical complications, and the specificity was 0.9454, indicating effective exclusion of patients without mechanical complications. These results support the reliability of the model’s predictions and its potential utility for early screening and risk stratification in patients with AMI. Accordingly, the 13-feature LightGBM model was selected as the optimal model for this study. After hyperparameter tuning, the final LightGBM model was configured as follows: random_state = 42, learning_rate = 0.173, max_depth = 2, min_child_weight = 5, n_estimators = 30, subsample = 0.8, feature_fraction = 0.8, and eval_metric = "logloss". To further assess the model’s robustness to inter-center variability, additional cross-validation was performed. As shown in Supplementary Fig. 5A and 5B, the mean AUCs of the final model in five-fold and ten-fold cross-validation were 0.9548 ± 0.0106 and 0.9485 ± 0.0115, respectively, confirming the stability and reliability of the model. Model explanation Because data-driven predictive models are often difficult to interpret directly, their broader clinical adoption can be limited. SHAP provides two complementary forms of interpretability: global, feature-level explanations and local, individual-level explanations. A radar chart based on mean absolute SHAP values, scaled to the maximum value, is shown in Fig. 6 A. This plot quantifies and visualizes each feature’s contribution to the model output, providing an intuitive overview of how individual predictors influence predictions across different value ranges. As shown in Fig. 6 B, age has the greatest impact on SHAP values. Further analysis showed that blood-related features such as NEUT, MPV, and D-dimer also substantially influence prediction outcomes. This suggests that even in the absence of disease-specific biomarkers, routinely available clinical variables can still provide strong predictive capability. SHAP dependence plots illustrate how individual features influence model predictions. As shown in Fig. 7 , increasing age, hsCRP, NEUT, D-dimer, and LDH has a marked positive effect on the model’s risk estimates, indicating that higher values of these variables are associated with an increased likelihood of mechanical complications. These findings highlight the multifactorial and nonlinear effects of routine clinical indicators on model outputs in critically ill patients with AMI. The case of a patient who did not develop mechanical complications during CCU hospitalization is illustrated in Figs. 8 A–C. In this patient, age, hsCRP, sex, D-dimer, PCI, MPV, history of chronic pulmonary disease, LDH, and MA predominantly acted as risk-attenuating factors. The final model output was − 0.97, corresponding to a low estimated probability of mechanical complications. The waterfall plot further displays the actual observed values for each feature, demonstrating that most variables (e.g., age, hsCRP) fell within normal or lower-risk ranges, thereby shifting the prediction toward the non-mechanical complications category. Notably, even in the presence of otherwise favorable feature values, a history of alcohol use alone was sufficient to increase the estimated risk, although not to a degree that altered the final classification, which remained “non- mechanical complications”. The waterfall plot for a patient who developed mechanical complications during CCU hospitalization is shown in Figs. 8 D–F. In this case, hsCRP and a history of chronic pulmonary disease were the dominant factors driving the prediction toward higher risk. Additional contributors—including age, D-dimer, history of alcohol use, MA, NEUT, SGLT2i use, LDH, and PCI—further enhanced the model output, collectively indicating a significant likelihood of mechanical complications. The final model output was 4.721, corresponding to a high predicted risk of mechanical complications for this patient. Discussion This retrospective, single-center study aimed to develop and compare eight ML models for predicting mechanical complications in patients with AMI. Predictors readily available from the EMR were extracted, and ML-based feature selection was used to identify the most informative variables. The overarching goal was to provide a more accurate and clinically applicable prediction tool to support early detection of mechanical complications and to guide individualized management and risk stratification [ 26 – 28 ]. LightGBM is a gradient boosting decision tree framework that performs particularly well on large, structured datasets, offering fast training and addressing key limitations of conventional decision tree algorithms[ 29 ]. It maintains high predictive accuracy even in large but incomplete datasets[ 30 ], making it well suited to predicting mechanical complications in AMI, a rapidly evolving and complex condition. Although ML models often function as “black boxes,” limiting clinicians’ willingness to rely on their predictions, this limitation is mitigated in our study by integrating SHAP[ 31 ]. SHAP analysis identified age as the most influential predictor of early mechanical complications in AMI, with substantially higher average SHAP values than other features, consistent with the clinical understanding that age-related myocardial and vascular sclerosis and functional decline reduce cardiovascular reserve and increase susceptibility to mechanical complications[ 32 , 33 ]. Additionally, SHAP values highlighted hsCRP as a key predictor, consistent with its role as an inflammatory marker associated with myocardial necrosis [ 34 – 36 ]. Excessive and irregular alcohol consumption has also been linked to increased risks of ischemic heart disease and hypertension, further aggravating cardiovascular damag[ 37 , 38 ]. In line with the GRACE registry, which reports a higher incidence of mechanical complications in STEMI than in NSTEMI patients[ 39 – 41 ], our study likewise found STEMI, female sex, and delayed PCI to be important risk factors [ 42 – 44 ]. Moreover, we identified several less widely recognized predictors—D-dimer, NEUT, MPV, LDH, history of chronic pulmonary disease, SGLT2i use, and MA, which may provide additional value for early identification and risk stratification of patients at risk for mechanical complications in AMI. An increase in NEUT and a decrease in MPV may reflect intense inflammatory responses. Patients with a history of chronic pulmonary disease often experience chronic hypoxia, inflammation, and oxidative stress, significantly increasing cardiac fragility and increasing cardiovascular disease risk[ 45 ]. LDH, which originates from myocardial, skeletal muscle, liver, red blood cells, and intestines[ 46 , 47 ], is a marker for systemic organ hypoperfusion and exacerbated myocardial necrosis[ 48 ]. Data from the international myocardial infarction and diabetes registry and the EMPULSE trial confirm that patients who start SGLT2i treatment during hospitalization after MI experience significantly reduced cardiovascular death, arrhythmia burden, and acute kidney injury [ 49 , 50 ]. MA, particularly ventricular fibrillation and sustained ventricular tachycardia, often signify extensive infarct size and severe myocardial ischemia and necrosis. Although current literature does not specifically identify these factors as independent predictors of mechanical complications in AMI, our study confirms that when combined with other factors, they are associated with an increased risk of mechanical complications in acute myocardial infarction. We acknowledge several limitations of this study. First, the temporal span of the data may introduce heterogeneity between samples, potentially affecting the stability and generalizability of the model. Future studies should incorporate prospective data to further validate the model’s predictability and extend its applicability to diverse populations and disease states. Although ML techniques require large datasets for model development, no established standard exists for calculating the sample size for machine learning-based predictive models. Nevertheless, robust cross-validation indicates that the sample size is appropriate. In conclusion, we developed an interpretable ML model based on routinely available EMR data to predict mechanical complications in patients with AMI, with the final LightGBM model demonstrating excellent discriminatory performance. Using an optimal probability threshold, critically ill patients identified as high risk can be prioritized for closer monitoring and timely intervention. Prospective randomized trials are needed to determine whether model-guided, individualized management can improve outcomes in AMI-related mechanical complications. Declarations Ethics approval and consent to participate This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study was approved by the Ethics Committee of the First Affiliated Hospital of Harbin Medical University, ethical number: 2025369, and the right to exempt informed consent was obtained. Consent for publication Not applicable. Competing interests The authors declare no competing interests. Funding This study was supported by the Natural Science Foundation of Heilongjiang Province (PL2024H032) Author Contribution Y.C. and J.Y. designed and experimented and participated in the data collection, and was a major contributor in writing the manuscript. Y.L. and G.C. collected data on patients with AMI, while X.D. and J.Z. analyzed data on patients with mechanical complications in patients with AMI. J.W. is responsible for organizing the data. J.L. and L.S., as corresponding authors, guided the progress of the study throughout the process to ensure the authenticity of the data. All authors reviewed the manuscript. Acknowledgements Not applicable. Data Availability The datasets used and/or analysed during the current study available from the corresponding author on reasonable request. References Puerto E, Viana-Tejedor A, Martínez-Sellés M, et al. Temporal trends in mechanical complications of acute myocardial infarction in the elderly[J]. J Am Coll Cardiol. 2018;72(9):959–66. https://doi.org/10.1016/j.jcin.2019.04.039 . Birnbaum Y, Fishbein MC, Blanche C, et al. Ventricular septal rupture after acute myocardial infarction[J]. N Engl J Med. 2002;347(18):1426–32. https://doi.org/10.1056/NEJMra020228 . Elbadawi A, Elgendy IY, Mahmoud K, et al. Temporal trends and outcomes of mechanical complications in patients with acute myocardial infarction[J]. JACC: Cardiovasc Interventions. 2019;12(18):1825–36. https://doi.org/10.1016/j.jcin.2019.04.039 . Gong FF, Vaitenas I, Malaisrie SC, et al. Mechanical complications of acute myocardial infarction: a review[J]. JAMA Cardiol. 2021;6(3):341–9. https://doi.org/10.1001/jamacardio.2020.3690 . Damluji AA, Forman DE, Van Diepen S, et al. Older adults in the cardiac intensive care unit: factoring geriatric syndromes in the management, prognosis, and process of care: a scientific statement from the American Heart Association[J]. Circulation. 2020;141(2):e6–32. https://doi.org/10.1161/CIR.0000000000000741 . Fu Y, Li KB, Yang XC. A risk score model for predicting cardiac rupture after acute myocardial infarction[J]. Chin Med J. 2019;132(09):1037–44. https://doi.org/10.1097/CM9.0000000000000175 . Saito Y, Shiko Y, Tateishi K, et al. Combined Risk Stratification With Patient Characteristics and Biomarkers in Patients Treated With the Impella for Cardiogenic Shock[J]. J Am Heart Association. 2025;14(9):e040487. https://doi.org/10.1161/JAHA.124.040487 . Cheng X, Manandhar I, Aryal S, et al. Application of artificial intelligence in cardiovascular medicine[J]. Compr Physiol. 2021;11(4):2455–66. https://doi.org/10.1002/j.2040-4603.2021.tb00186.x . Goecks J, Jalili V, Heiser LM et al. How machine learning will transform biomedicine[J].Cell,2020,181(1):92101. https://doi.org/10.1016/j.cell.2020.03.022 Gould MK, Huang BZ, Tammemagi MC, et al. Machine learning for early lung cancer identification using routine clinical and laboratory data[J]. Am J Respir Crit Care Med. 2021;204(4):445–53. https://doi.org/10.1164/rccm.202007-2791OC . Gao J, Lu Y, Ashrafi N, et al. Prediction of sepsis mortality in ICU patients using machine learning methods[J]. BMC Med Inf Decis Mak. 2024;24(1):228. https://doi.org/10.1186/s12911-024-02630-z . Zhou H, Li F, Liu X. Early prediction of septic shock in ICU patients using machine learning: development, external validation, and explainability with SHAP[J]. Int J Med Informatics. 2025;106169. https://doi.org/10.1016/j.ijmedinf.2025.106169 . Wang S, Li J, Sun L, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction[J]. BMC Med Inf Decis Mak. 2021;21(1):301. Al Azwari S. Predicting myocardial rupture after acute myocardial infarction in hospitalized patients using machine learning[C]//2021 National Computing Colleges Conference (NCCC). IEEE, 2021: 1–6. Byrne RA, Rossello X, Coughlan JJ, ESC Scientific Document Group, et al. Linee guida ESC 2023 per il trattamento delle sindromi coronariche acute [2023 ESC Guidelines for the management of acute coronary syndromes]. G Ital Cardiol (Rome). 2024;25(2 Suppl 2):e1–112. https://doi.org/10.1714/4191.41785 . Italian. Thygesen K, Alpert JS, Jaffe AS, et al. Fourth universal definition of myocardial infarction (2018)[J]. J Am Coll Cardiol. 2018;72(18):2231–64. https://doi.org/10.1016/j.jacc.2018.08.1038 . Chinese Society of Cardiology, Chinese Journal of Cardiovascular Diseases Editorial Committee. Guidelines for the Diagnosis and Treatment of Acute ST-segment Elevation Myocardial Infarction. (2019) [J]. Chinese Journal of Cardiovascular Diseases, 2019, 47(10): 766–783. https://doi.org/10.3760/cma.j.issn.0253-3758.2019.10.003 Damluji AA, Van Diepen S, Katz JN, et al. Mechanical complications of acute myocardial infarction: a scientific statement from the American Heart Association[J]. Circulation. 2021;144(2):e16–35. https://doi.org/10.1161/CIR.0000000000000985 . Berkelmans GFN, Read SH, Gudbjörnsdottir S, et al. Population median imputation was noninferior to complex approaches for imputing missing values in cardiovascular prediction models in clinical practice[J]. J Clin Epidemiol. 2022;145:70–80. https://doi.org/10.1016/j.jclinepi.2022.01.011 . He H, Garcia EA. Learning from imbalanced data[J]. IEEE Trans Knowl Data Eng. 2009;21(9):1263–84. https://doi.org/10.1109/TKDE.2008.239 . Hoens TR, Chawla NV. Imbalanced datasets: from sampling to classifiers[J]. Imbalanced learning: Foundations, algorithms, and applications, 2013: 43–59. https://doi.org/10.1002/9781118646106.ch3 Davis J, Goadrich M. The relationship between Precision-Recall and ROC curves[C]//Proceedings of the 23rd international conference on Machine learning. 2006: 233–240. https://doi.org/10.1145/1143844.1143874 Lundberg SM, Lee SI. A unified approach to interpreting model predictions[J]. Adv Neural Inf Process Syst, 2017, 30. DeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach[J]. Biometrics, 1988: 837–45. Azodi CB, Tang J, Shiu SH. Opening the black box: interpretable machine learning for geneticists[J]. Trends Genet. 2020;36(6):442–55. https://doi.org/10.1016/j.tig.2020.03.005 . Rajkomar A, Dean J, Kohane I. Machine Learning in Medicine.N. Engl. J Med. 2019;380:1347–58. Zhang X, Yan C, Gao C, et al. Predicting missing values in medical data via XGBoost regression[J]. J Healthc Inf Res. 2020;4(4):383–94. https://doi.org/10.1007/s41666-020-00077-1 . Li Y, Sperrin M, Ashcroft DM, et al. Consistency of variety of machine learning and statistical models in predicting clinical risks of individual patients: longitudinal cohort study using cardiovascular disease as exemplar[J]. BMJ. 2020;371. https://doi.org/10.1136/bmj.m3919 . Ke G, Meng Q, Finley T et al. Lightgbm: A highly efficient gradient boosting decision tree[J]. Adv Neural Inf Process Syst, 2017, 30. Yan J, Xu Y, Cheng Q et al. LightGBM: accelerated genomically designed crop breeding through ensemble learning[J]. https://doi.org/10.1186/s13059-021-02492-y Marcílio WE, Eler DM. From explanations to feature selection: assessing SHAP values as feature selection mechanism[C]//2020 33rd SIBGRAPI conference on Graphics, Patterns and Images (SIBGRAPI). Ieee, 2020: 340–347. https://doi.org/10.1109/SIBGRAPI51738.2020.00053 Valley TS, Sjoding MW, Ryan AM, et al. Intensive care unit admission and survival among older patients with chronic obstructive pulmonary disease, heart failure, or myocardial infarction[J]. Annals Am Thorac Soc. 2017;14(6):943–51. https://doi.org/10.1513/AnnalsATS.201611-847OC . Saaby L, Poulsen TS, Hosbond S, et al. Classification of myocardial infarction: frequency and features of type 2 myocardial infarction[J]. Am J Med. 2013;126(9):789–97. https://doi.org/10.1016/j.amjmed.2013.02.029 . Ueda S, Ikeda U, Yamamoto K, et al. C-reactive protein as a predictor of cardiac rupture after acute myocardial infarction[J]. Am Heart J. 1996;131(5):857–60. Miller M, Zhan M, Havas S. High attributable risk of elevated C-reactive protein level to conventional coronary heart disease risk factors: the Third National Health and Nutrition Examination Survey[J]. Arch Intern Med. 2005;165(18):2063–8. Maier W, Altwegg LA, Corti R, et al. Inflammatory markers at the site of ruptured plaque in acute myocardial infarction: locally increased interleukin-6 and serum amyloid A but decreased C-reactive protein[J]. Circulation. 2005;111(11):1355–61. https://doi.org/10.1161/01.CIR.0000158479.58589.0A . Roerecke M, Rehm J. Chronic heavy drinking and ischaemic heart disease: a systematic review and meta-analysis[J]. Open heart. 2014;1(1):e000135. https://doi.org/10.1136/openhrt-2014-000135 . Vaideeswar P, Chaudhari C, Rane S, et al. Cardiac pathology in chronic alcoholics: a preliminary study[J]. J Postgrad Med. 2014;60(4):372–6. https://doi.org/10.4103/0022-3859.143958 . López-Sendón J, Gurfinkel EP, Lopez de Sa E, et al. Factors related to heart rupture in acute coronary syndromes in the Global Registry of Acute Coronary Events[J]. Eur Heart J. 2010;31(12):1449–56. https://doi.org/10.1093/eurheartj/ehq089 . Crenshaw BS, Granger CB, Birnbaum Y, et al. Risk factors, angiographic patterns, and outcomes in patients with ventricular septal defect complicating acute myocardial infarction[J]. Circulation. 2000;101(1):27–32. https://doi.org/10.1161/01.CIR.101.1.27 . Goldberg RJ, Spencer FA, Gore JM, et al. Thirty-year trends (1975 to 2005) in the magnitude of, management of, and hospital death rates associated with cardiogenic shock in patients with acute myocardial infarction: a population-based perspective[J]. Circulation. 2009;119(9):1211–9. https://doi.org/10.1161/CIRCULATIONAHA.108.814947 . Menon V, Webb JG, Hillis LD, et al. Outcome and profile of ventricular septal rupture with cardiogenic shock after myocardial infarction: a report from the SHOCK Trial Registry[J]. J Am Coll Cardiol. 2000;36(3S1):1110–6. https://doi.org/10.1016/S0735-1097(00)00878-0 . Crenshaw BS, Granger CB, Birnbaum Y, et al. Risk factors, angiographic patterns, and outcomes in patients with ventricular septal defect complicating acute myocardial infarction[J]. Circulation. 2000;101(1):27–32. https://doi.org/10.1161/01.CIR. 101.1.27 . Yang Y, Tang J, Ma L, et al. A systematic comparison of short-term and long-term mortality prediction in acute myocardial infarction using machine learning models[J]. BMC Med Inf Decis Mak. 2025;25(1):208. https://doi.org/10.1186/s12911-025-03052-1 . Dunlay SM, Chamberlain AM. Multimorbidity in older patients with cardiovascular disease[J]. Curr Cardiovasc risk Rep. 2016;10(1):3. https://doi.org/10.1007/s12170-016-0491-8 . Storey KB. Comparative enzymology—new insights from studies of an old enzyme, lactate dehydrogenase[J]. Comp Biochem Physiol B: Biochem Mol Biol. 2016;199:13–20. https://doi.org/10.1016/j.cbpb.2015.12.004 . Oliveira M, Seringa J, Pinto FJ, et al. Machine learning prediction of mortality in acute myocardial infarction[J]. BMC Med Inf Decis Mak. 2023;23(1):70. https://doi.org/10.1186/s12911-023-02168-6 . Zweck E, Hassager C, Beske RP, et al. Microaxial flow pump use and renal outcomes in infarct-related cardiogenic shock: a secondary analysis of the DanGer Shock Trial[J]. Circulation. 2024;150(25):1990–2003. https://doi.org/10.1161/CIRCULATIONAHA.124.072370 . Paolisso P, Bergamaschi L, Gragnano F, et al. Outcomes in diabetic patients treated with SGLT2-Inhibitors with acute myocardial infarction undergoing PCI: The SGLT2-I AMI PROTECT Registry[J]. Pharmacol Res. 2023;187:106597. von Lewinski D, Kolesnik E, Aziz F, et al. Timing of SGLT2i initiation after acute myocardial infarction[J]. Cardiovasc Diabetol. 2023;22(1):269. https://doi.org/10.1186/s12933-023-02000-5 . Additional Declarations No competing interests reported. Supplementary Files SupplementaryMaterial.docx Cite Share Download PDF Status: Posted Version 1 posted You are reading this latest preprint version Research Square lets you share your work early, gain feedback from the community, and start making changes to your manuscript prior to peer review in a journal. As a division of Research Square Company, we’re committed to making research communication faster, fairer, and more useful. We do this by developing innovative software and high quality services for the global research community. Our growing team is made up of researchers and industry professionals working together to solve the most critical problems facing scientific publishing. Also discoverable on Platform About Our Team In Review Editorial Policies Advisory Board Help Center Resources Author Services Accessibility API Access RSS feed Manage Cookie Preferences © Research Square 2026 | ISSN 2693-5015 (online) Privacy Policy Terms of Service Do Not Sell My Personal Information {"props":{"pageProps":{"initialData":{"identity":"rs-8461400","acceptedTermsAndConditions":true,"allowDirectSubmit":true,"archivedVersions":[],"articleType":"Research Article","associatedPublications":[],"authors":[{"id":578344017,"identity":"ced0d975-0216-4715-be4c-b4dc062c07b3","order_by":0,"name":"Yajie Chang","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yajie","middleName":"","lastName":"Chang","suffix":""},{"id":578344018,"identity":"f46f14c3-86ad-42ac-aa23-1c89c2b0ca3b","order_by":1,"name":"Jiahui Yu","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jiahui","middleName":"","lastName":"Yu","suffix":""},{"id":578344019,"identity":"7303916a-d57d-4e93-9796-3418dbeebfc4","order_by":2,"name":"Yang Liu","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Yang","middleName":"","lastName":"Liu","suffix":""},{"id":578344020,"identity":"034c9cbf-8649-4331-af7d-326ed73cced7","order_by":3,"name":"Guangyuan Chen","email":"","orcid":"","institution":"First Hospital of Harbin City","correspondingAuthor":false,"prefix":"","firstName":"Guangyuan","middleName":"","lastName":"Chen","suffix":""},{"id":578344023,"identity":"7ae3853d-dffd-4748-a149-51f2c5366f5f","order_by":4,"name":"Xiaoxu Duan","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Xiaoxu","middleName":"","lastName":"Duan","suffix":""},{"id":578344025,"identity":"283ddce1-4ffb-4f57-a3d6-4cf6a29a5555","order_by":5,"name":"Jintong Zhang","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jintong","middleName":"","lastName":"Zhang","suffix":""},{"id":578344026,"identity":"8fc354c5-c7bc-4d2e-8479-6e5c106c9121","order_by":6,"name":"Jie Wang","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Jie","middleName":"","lastName":"Wang","suffix":""},{"id":578344029,"identity":"15770517-fea1-4c03-a8dc-1a5cff245de6","order_by":7,"name":"Haiyu Zhang","email":"","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":false,"prefix":"","firstName":"Haiyu","middleName":"","lastName":"Zhang","suffix":""},{"id":578344032,"identity":"3cc4ceb2-b556-47ff-9736-72e3ff503250","order_by":8,"name":"Jia Li","email":"data:image/png;base64,iVBORw0KGgoAAAANSUhEUgAAAZAAAAAyAQMAAABI0h/eAAAABlBMVEX///8AAABVwtN+AAAACXBIWXMAAA7EAAAOxAGVKw4bAAAAvklEQVRIiWNgGAWjYBACPmaGhAMMFQcYDMBcNiK0sIG1nCFJC4hgbCNJCzvDwwM/593JNxc7Y8DwoewwA//sBsIOO9i77Znlztk5Bowzzh1mkLhzgLCWA7zbDhsY3M4xYOZtO8xgIJFAhC1/50C1/CVWy2HeBqgWRqK1yBw7bGA5O63gYM+5dB6JGwS08POfSf74puawgbl08sYHP8qs5fhnENDCwMCDUHEAxCWkHgjYDxChaBSMglEwCkY0AAC/9ULm0e48cgAAAABJRU5ErkJggg==","orcid":"","institution":"First Affiliated Hospital of Harbin Medical University","correspondingAuthor":true,"prefix":"","firstName":"Jia","middleName":"","lastName":"Li","suffix":""}],"badges":[],"createdAt":"2025-12-27 13:38:31","currentVersionCode":1,"declarations":"","doi":"10.21203/rs.3.rs-8461400/v1","doiUrl":"https://doi.org/10.21203/rs.3.rs-8461400/v1","draftVersion":[],"editorialEvents":[],"editorialNote":"","failedWorkflow":false,"files":[{"id":100898005,"identity":"d3069578-87ad-4570-8515-aabfd51e752e","added_by":"auto","created_at":"2026-01-22 14:26:12","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10398497,"visible":true,"origin":"","legend":"","description":"","filename":"Manuscript.docx","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/79f4229ccb3e717deb689d7c.docx"},{"id":100898022,"identity":"daac62cf-1749-4e4a-b99a-e2a2c4a53b64","added_by":"auto","created_at":"2026-01-22 14:26:15","extension":"json","order_by":1,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":10000,"visible":true,"origin":"","legend":"","description":"","filename":"1edeac5ab0934209aaf50f42adf4efbb.json","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/58f865bfd4312b4c952b37cc.json"},{"id":100898019,"identity":"3d70c208-b33c-4f67-ba91-18be22f7f7f9","added_by":"auto","created_at":"2026-01-22 14:26:15","extension":"docx","order_by":2,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":12064811,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/f7cb2f460800db06dbaeb068.docx"},{"id":100898016,"identity":"c2473451-8232-426a-8e74-68263bc01d5f","added_by":"auto","created_at":"2026-01-22 14:26:14","extension":"xml","order_by":3,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":119205,"visible":true,"origin":"","legend":"","description":"","filename":"1edeac5ab0934209aaf50f42adf4efbb1enriched.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/3c9c9e97d9271b90b77ebb27.xml"},{"id":100898006,"identity":"53fc4377-92d0-4dbe-859c-d244c8ab6492","added_by":"auto","created_at":"2026-01-22 14:26:13","extension":"png","order_by":12,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":77198,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage1.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/90d5f36c481af129578e6d56.png"},{"id":100898020,"identity":"a36447e2-e61d-41be-86f9-380bc9bfb635","added_by":"auto","created_at":"2026-01-22 14:26:15","extension":"png","order_by":13,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":153256,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage2.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/9dcfc9e209bf78d8b12915f9.png"},{"id":100898038,"identity":"e5a01801-f979-4655-ada0-079f387e2d7e","added_by":"auto","created_at":"2026-01-22 14:26:21","extension":"png","order_by":14,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":69995,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage3.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/6c40442b59bc4bcca9cbd911.png"},{"id":100898001,"identity":"09cc1d7a-32ea-4a96-a715-342a3aae760e","added_by":"auto","created_at":"2026-01-22 14:26:12","extension":"png","order_by":15,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":42286,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage4.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/28bff6eb2433e9c64b7a8b43.png"},{"id":100898033,"identity":"3f378744-5ef9-45d6-86ae-2bfc3633dcfa","added_by":"auto","created_at":"2026-01-22 14:26:19","extension":"png","order_by":16,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":62123,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage5.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/ec989ac8ac8b9f8b02758e27.png"},{"id":100897929,"identity":"e469a44b-97b0-4d8d-b09e-0101643817fe","added_by":"auto","created_at":"2026-01-22 14:26:04","extension":"png","order_by":17,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":57974,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage6.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/d30ea3bff89acf65491f8453.png"},{"id":100898009,"identity":"afa8f30f-9227-45fe-914e-92059ebde400","added_by":"auto","created_at":"2026-01-22 14:26:13","extension":"png","order_by":18,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":94773,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage7.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/41449d1a21b9b9ae4e469c47.png"},{"id":100897923,"identity":"efeacf77-1b9e-4fed-a79b-6622f6d7fc40","added_by":"auto","created_at":"2026-01-22 14:26:03","extension":"png","order_by":19,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":84558,"visible":true,"origin":"","legend":"","description":"","filename":"Onlinefloatimage8.png","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/4c73fdb10a19dac2799207f9.png"},{"id":100898044,"identity":"844e1db2-38e0-4e8f-a6a1-38e37236d753","added_by":"auto","created_at":"2026-01-22 14:26:22","extension":"xml","order_by":20,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":117910,"visible":true,"origin":"","legend":"","description":"","filename":"1edeac5ab0934209aaf50f42adf4efbb1structuring.xml","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/43100dc6ce4798cd98a25222.xml"},{"id":100898040,"identity":"fda0aba3-3666-461d-9669-c418b964c350","added_by":"auto","created_at":"2026-01-22 14:26:21","extension":"html","order_by":21,"title":"","display":"","copyAsset":false,"role":"acdc-reference","size":131494,"visible":true,"origin":"","legend":"","description":"","filename":"earlyproof.html","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/6101a0d7ec9ea110a81ba2f4.html"},{"id":100898030,"identity":"caf64818-464a-40cc-983e-f4b7b392c115","added_by":"auto","created_at":"2026-01-22 14:26:18","extension":"jpeg","order_by":1,"title":"Figure 1","display":"","copyAsset":false,"role":"figure","size":398579,"visible":true,"origin":"","legend":"\u003cp\u003eOverall flow chart of the study\u003c/p\u003e","description":"","filename":"floatimage1.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/6436151bd32479e2233d8eba.jpeg"},{"id":100950980,"identity":"ef5792af-f192-4dbc-abef-f93e4db0d2a9","added_by":"auto","created_at":"2026-01-23 07:09:45","extension":"jpeg","order_by":2,"title":"Figure 2","display":"","copyAsset":false,"role":"figure","size":495563,"visible":true,"origin":"","legend":"\u003cp\u003eCorrelation heatmap between variables: (A) Correlation heatmap of all variables. (B) Correlation heatmap after removing collinear variables\u003c/p\u003e","description":"","filename":"floatimage2.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/a93d9d0c49cf622aaf789900.jpeg"},{"id":100897996,"identity":"b0ec81bc-2cb9-46e1-ac9f-4b5936d14044","added_by":"auto","created_at":"2026-01-22 14:26:11","extension":"jpeg","order_by":3,"title":"Figure 3","display":"","copyAsset":false,"role":"figure","size":346598,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance of the ML model to predict mechanical complications. (A) ROC curves of the eight ML models. (B) P-R curves of the eight ML models\u003c/p\u003e","description":"","filename":"floatimage3.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/baf13d463f1decf2b5a939bf.jpeg"},{"id":100897995,"identity":"2c5a253c-2cc0-499c-864a-bca1e9ad34ed","added_by":"auto","created_at":"2026-01-22 14:26:11","extension":"jpeg","order_by":4,"title":"Figure 4","display":"","copyAsset":false,"role":"figure","size":186749,"visible":true,"origin":"","legend":"\u003cp\u003eAUCs of the top five best-performing ML models with varied number of features\u003c/p\u003e","description":"","filename":"floatimage4.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/f3a1e71b43fccdd1cd9a0f8a.jpeg"},{"id":100897985,"identity":"5f4699b7-6200-437b-8b9a-7bfc6ef1e8ad","added_by":"auto","created_at":"2026-01-22 14:26:08","extension":"jpeg","order_by":5,"title":"Figure 5","display":"","copyAsset":false,"role":"figure","size":321957,"visible":true,"origin":"","legend":"\u003cp\u003ePerformance indicators of LightGBM model with varied number of features\u003c/p\u003e","description":"","filename":"floatimage5.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/e9c812a789bdb24465f27fbc.jpeg"},{"id":100898042,"identity":"2b4817fb-80d8-4f2a-860e-d8a5432a5759","added_by":"auto","created_at":"2026-01-22 14:26:22","extension":"jpeg","order_by":6,"title":"Figure 6","display":"","copyAsset":false,"role":"figure","size":266334,"visible":true,"origin":"","legend":"\u003cp\u003eGlobal interpretability analysis results (A) SHAP aggregated radar chart. (B) SHAP summary dot plot.\u003c/p\u003e","description":"","filename":"floatimage6.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/f35407ae947d99fdf49b4f6b.jpeg"},{"id":100897991,"identity":"4701947f-a1ee-4028-98da-3765be82e9c3","added_by":"auto","created_at":"2026-01-22 14:26:10","extension":"jpeg","order_by":7,"title":"Figure 7","display":"","copyAsset":false,"role":"figure","size":518931,"visible":true,"origin":"","legend":"\u003cp\u003eThe SHAP correlation results of each feature\u003c/p\u003e","description":"","filename":"floatimage7.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/8ab928767bed013f3b0f99fb.jpeg"},{"id":100898027,"identity":"43738ddd-f859-4209-a01b-b30aeeabfab7","added_by":"auto","created_at":"2026-01-22 14:26:17","extension":"jpeg","order_by":8,"title":"Figure 8","display":"","copyAsset":false,"role":"figure","size":527570,"visible":true,"origin":"","legend":"\u003cp\u003eIndividual interpretability analysis results. (A)-(C) Non-mechanical complication; (D)-(F) mechanical complication\u003c/p\u003e","description":"","filename":"floatimage8.jpeg","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/d532fa1a0196e21e831d1936.jpeg"},{"id":109429416,"identity":"6a05a199-dd56-4dd7-bd28-50aee35b675b","added_by":"auto","created_at":"2026-05-18 03:55:49","extension":"pdf","order_by":0,"title":"","display":"","copyAsset":false,"role":"manuscript-pdf","size":3293600,"visible":true,"origin":"","legend":"","description":"","filename":"manuscript.pdf","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/cfae7016-41de-4712-a120-3d739fbc1516.pdf"},{"id":100898010,"identity":"c3ed7f50-ba6f-46d9-b38d-379eb2630d91","added_by":"auto","created_at":"2026-01-22 14:26:13","extension":"docx","order_by":0,"title":"","display":"","copyAsset":false,"role":"supplement","size":12064811,"visible":true,"origin":"","legend":"","description":"","filename":"SupplementaryMaterial.docx","url":"https://assets-eu.researchsquare.com/files/rs-8461400/v1/e7142797cdf7ad84f50e89cf.docx"}],"financialInterests":"No competing interests reported.","formattedTitle":"Predicting the risk of mechanical complications in acute myocardial infarction using an interpretable machine learning model","fulltext":[{"header":"Introduction","content":"\u003cp\u003eMechanical complications in AMI typically refer to spontaneous myocardial rupture occurring after acute myocardial infarction. Based on the site of rupture, mechanical complications in AMI can be classified into free wall rupture (FWR), ventricular septal rupture (VSR), papillary muscle rupture (PMR), and other types such as pseudoaneurysm and true aneurysm [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e]. The widespread application of reperfusion therapy has significantly reduced the incidence of mechanical complications [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR2\" class=\"CitationRef\"\u003e2\u003c/span\u003e]. Unfortunately, the associated mortality rate shown no significant decline over the past two decades; the in-hospital mortality risk for patients with mechanical complications in AMI is more than fourfold higher than that of non-mechanical complications counterparts [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR3\" class=\"CitationRef\"\u003e3\u003c/span\u003e]. Hence, although mechanical complications in AMI are relatively uncommon, they remain a critical determinant of clinical outcomes.\u003c/p\u003e \u003cp\u003eTransthoracic echocardiography is the primary tool for diagnosing mechanical complications in AMI. However, surgical exploration or autopsy remains the gold standard for confirmation [\u003cspan citationid=\"CR4\" class=\"CitationRef\"\u003e4\u003c/span\u003e]. Imaging examinations typically detect mechanical complications only after the event has occurred, making it difficult to identify pre-rupture signs and thereby limiting the intervention window.\u003c/p\u003e \u003cp\u003eThe establishment of predictive models based on electronic medical records (EMR) have attracted increasing attention in recent years. Previous studies have identified common clinical characteristics of patients with mechanical complications in AMI, including advanced age, female sex, a history of heart failure or chronic kidney disease, and delayed reperfusion [\u003cspan citationid=\"CR1\" class=\"CitationRef\"\u003e1\u003c/span\u003e, \u003cspan citationid=\"CR5\" class=\"CitationRef\"\u003e5\u003c/span\u003e]. However, some studies rely heavily on traditional statistical models [\u003cspan citationid=\"CR6\" class=\"CitationRef\"\u003e6\u003c/span\u003e], which only characterize linear relationships between features and outcomes. Given the complex pathophysiology of mechanical complications, relying on a single biomarker is insufficient to comprehensively capture the multifaceted risk profile [\u003cspan citationid=\"CR7\" class=\"CitationRef\"\u003e7\u003c/span\u003e]. Machine learning (ML), a core branch of artificial intelligence, can efficiently integrate multi-source data in clinical diagnosis, precision therapy, health management, and monitoring scenarios[\u003cspan additionalcitationids=\"CR9 CR10\" citationid=\"CR8\" class=\"CitationRef\"\u003e8\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR11\" class=\"CitationRef\"\u003e11\u003c/span\u003e]. To the best of the author\u0026rsquo;s knowledge, studies utilizing ML models to predict the risk of mechanical complications in AMI remain limited[\u003cspan citationid=\"CR12\" class=\"CitationRef\"\u003e12\u003c/span\u003e, \u003cspan citationid=\"CR13\" class=\"CitationRef\"\u003e13\u003c/span\u003e]. The few available investigations have generally relied on a single machine learning algorithm[\u003cspan citationid=\"CR14\" class=\"CitationRef\"\u003e14\u003c/span\u003e], and due to the inherent nature of these models as \"black boxes\", they offer insufficient interpretability.\u003c/p\u003e \u003cp\u003eThis study aims to develop and validate an interpretable ML model for the early and accurate prediction of mechanical complications risk in AMI. Beyond predicting the risk of mechanical complications in AMI, we employ the Shapley Additive Explanations (SHAP) to achieve interpretability at both the global and individual level. By enabling early identification and dynamic monitoring of high-risk patients, this model may facilitate timely interventions, including surgical procedures, and ultimately reduce mortality related to mechanical complications.\u003c/p\u003e"},{"header":"Methods","content":"\u003cdiv id=\"Sec3\" class=\"Section2\"\u003e \u003ch2\u003eStudy population\u003c/h2\u003e \u003cp\u003eThis retrospective, single-center cohort study was conducted at the First Affiliated Hospital of Harbin Medical University to derive and validate a predictive model. The study included AMI patients admitted to the CCU from September 2015 to August 2025, among whom 239 patients developing mechanical complications in AMI. To ensure accurate predictive performance, we randomly selected 611 patients from the same period who did not develop mechanical complications, resulting in a total of 850 patients. All patients received standardized AMI treatment in accordance with CCU protocols and clinical guidelines. The study adhered to the Declaration of Helsinki and was approved by the hospital's Ethics Committee. Due to its retrospective design and the de-identified nature of the data, the Ethics Committee waived the requirement for written informed consent.\u003c/p\u003e \u003cp\u003eInclusion criteria for patients with mechanical complications in AMI: (1) Age\u0026thinsp;\u0026ge;\u0026thinsp;18 years; (2) The inclusion criteria for AMI patients followed the latest guidelines for the diagnosis of acute myocardial infarction issued by the European Society of Cardiology[\u003cspan citationid=\"CR15\" class=\"CitationRef\"\u003e15\u003c/span\u003e]; (3) Patients admitted for AMI who had emergent cardiac catheterization or hemodynamic compromise with echocardiographic evidence of mechanical complications were included.\u003c/p\u003e \u003cp\u003eDiagnostic criteria for mechanical complications: FWR are typically characterized by the abrupt onset of consciousness disturbance, cardiogenic shock, electrical-mechanical dissociation, and acute cardiac tamponade. For diagnosis, at least one of the following criteria must be met: (1) Transthoracic echocardiography revealing pericardial effusion\u0026thinsp;\u0026gt;\u0026thinsp;1 cm with abnormal echogenicity and signs of pericardial tamponade; (2) Pericardiocentesis confirming hemopericardium; (3) Anatomical confirmation through surgery or autopsy. VSR presents as sudden onset heart failure or cardiogenic shock, accompanied by a new, coarse systolic murmur at the left sternal border (3rd\u0026ndash;4th intercostal spaces), with systolic thrill in some patients. Echocardiography may show interventricular septal echo discontinuity with left-to-right shunting. PMR manifests as sudden acute left heart failure, with new systolic murmurs or exacerbation of pre-existing murmurs in the mitral valve area. Echocardiography typically shows severe mitral regurgitation or direct visualization of papillary muscle rupture[\u003cspan additionalcitationids=\"CR17\" citationid=\"CR16\" class=\"CitationRef\"\u003e16\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR18\" class=\"CitationRef\"\u003e18\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eExclusion criteria: (1) Severe deficiency in clinical or laboratory data; (2) Comorbid severe infections, severe autoimmune diseases, or severe hematologic disorders; (3) mechanical complications secondary to other conditions, including infective endocarditis, cardiac tumors, myocarditis, cardiac amyloidosis, or iatrogenic injuries.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eData collection and processing\u003c/h3\u003e\n\u003cp\u003eThe model input features comprised a comprehensive set of variables, including patient demographics, initial medical contact details, cardiovascular risk factors, biochemical markers, metabolomic biomarkers, and treatment options, all collected within 48 hours of CCU admission. The data were obtained from the EMR system. To minimize potential bias, features with more than 25% missing values were excluded from subsequent analyses. For the remaining variables, continuous data were imputed using the median, while categorical data were imputed using the mode [\u003cspan citationid=\"CR19\" class=\"CitationRef\"\u003e19\u003c/span\u003e], to mitigate the impact of missing values on model performance.\u003c/p\u003e \u003cp\u003eTo address potential multicollinearity that could compromise predictive accuracy, we evaluated the correlation structure between features using Spearman\u0026rsquo;s rank correlation coefficient. For any pair of features with a correlation coefficient\u0026thinsp;\u0026gt;\u0026thinsp;0.6, we retained the feature more strongly associated with the outcome and excluded the other. In addition, we conducted statistical tests for all variables and removed those that were not statistically significant. The detailed implementation procedure is described in the Statistical Analysis section.\u003c/p\u003e\n\u003ch3\u003eModel development and performance evaluation\u003c/h3\u003e\n\u003cp\u003eWe employed eight ML models to predict the likelihood of mechanical complications in AMI: Adaptive Boosting (AdaBoost), Gradient Boosting Machine (GBM), Light Gradient Boosting Machine (LightGBM), Logistic Regression (LR), Random Forest (RF), Support Vector Machine (SVM), Multi-Layer Perceptron (MLP), and Extreme Gradient Boosting (XGBoost). The original dataset was randomly split into training and validation sets in a 7:3 ratio, with stratified sampling used to address class imbalance. Five-fold cross-validation was applied during training to optimize hyperparameters for each model, and the validation set was reserved for independent performance evaluation.\u003c/p\u003e \u003cp\u003eIt is important to note that predicting mechanical complications in AMI is a binary classification task, in which the non-complication group is substantially larger than the complication group, resulting in a typical class imbalance scenario[\u003cspan citationid=\"CR20\" class=\"CitationRef\"\u003e20\u003c/span\u003e]. In such settings, accuracy alone is insufficient to evaluate a model\u0026rsquo;s ability to identify the minority class. Therefore, the F1 score was adopted as a comprehensive performance metric[\u003cspan citationid=\"CR21\" class=\"CitationRef\"\u003e21\u003c/span\u003e]. Moreover, because the primary objective of this study was to maximize the detection of positive cases (i.e., patients with mechanical complications), recall was prioritized, and precision\u0026ndash;recall (P\u0026ndash;R) curves were plotted for each model. This metric framework has been shown to be more sensitive to the minority class in imbalanced datasets[\u003cspan citationid=\"CR22\" class=\"CitationRef\"\u003e22\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eModel performance was comprehensively evaluated and compared using the area under the receiver operating characteristic curve (AUC), accuracy, sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). Model calibration was assessed with calibration curves and the Brier score. Clinical utility was evaluated using decision curve analysis (DCA). In addition, to examine the impact of data distribution consistency on model stability, five-fold and ten-fold cross-validation were performed within the derivation cohort.\u003c/p\u003e\n\u003ch3\u003eFeature selection and model interpretation\u003c/h3\u003e\n\u003cp\u003eIdentifying the most informative variables for disease risk from a large pool of candidate features remains a major challenge in both clinical practice and data science. The SHAP framework addresses this by providing a unified approach to explain ML predictions: it quantifies each input feature\u0026rsquo;s contribution to the model output in a consistent, ordered manner, thereby mitigating the \u0026ldquo;black box\u0026rdquo; nature of ML models[\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e]. Using SHAP values for feature selection enables guided dimensionality reduction, allowing key features to be retained while preserving strong discriminative performance and enhancing the model\u0026rsquo;s applicability in real-world clinical settings. Differences in AUC among models were compared using the DeLong nonparametric method[\u003cspan citationid=\"CR24\" class=\"CitationRef\"\u003e24\u003c/span\u003e]. During feature reduction, features selected by the ML model were iteratively removed until a significant decline in AUC was observed.\u003c/p\u003e \u003cp\u003eIn this study, SHAP was applied at both global and individual levels. At the global level, the absolute magnitude of SHAP values reflects the significance of each feature in the model, thereby clarifying the overall relationship between predictors and mechanical complications risk. At the individual level, SHAP values further pinpoint the specific driving factors influencing each patient\u0026rsquo;s prediction. This dual-level interpretability strengthens transparency and explainability across both population-level trends and individual patient predictions [\u003cspan citationid=\"CR23\" class=\"CitationRef\"\u003e23\u003c/span\u003e, \u003cspan citationid=\"CR25\" class=\"CitationRef\"\u003e25\u003c/span\u003e].\u003c/p\u003e \u003cdiv id=\"Sec7\" class=\"Section2\"\u003e \u003ch2\u003eStatistical analysis\u003c/h2\u003e \u003cp\u003eStatistical analyses were performed using Python version 3.9. Continuous variables with non-normal distributions were expressed as medians with interquartile ranges, and between-group comparisons were performed using the Mann\u0026ndash;Whitney U test or Kruskal\u0026ndash;Wallis H test, as appropriate. Categorical variables were presented as frequencies and percentages, with comparisons made using the chi-square test or Fisher\u0026rsquo;s exact test. The discriminatory ability of the models was evaluated by the AUC, and the optimal threshold was determined by maximizing the Youden index (sensitivity\u0026thinsp;+\u0026thinsp;specificity\u0026thinsp;\u0026minus;\u0026thinsp;1). A two-sided P value\u0026thinsp;\u0026lt;\u0026thinsp;0.05 was considered statistically significant.\u003c/p\u003e \u003c/div\u003e"},{"header":"Results","content":"\u003cdiv id=\"Sec9\" class=\"Section2\"\u003e \u003ch2\u003ePatient characteristics\u003c/h2\u003e \u003cp\u003eThis retrospective study established a derivation cohort of 850 patients to identify key predictors and develop a predictive model. Patients were grouped according to the occurrence of mechanical complications in AMI. The complete methodological workflow is illustrated in Fig.\u0026nbsp;\u003cspan refid=\"Fig1\" class=\"InternalRef\"\u003e1\u003c/span\u003e.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe results of the correlation analysis are presented in Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Strong correlations were observed between height, weight, and BMI (see Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eA). After excluding collinearity, a total of 73 variables were retained for following analysis, and a heatmap (Fig.\u0026nbsp;\u003cspan refid=\"Fig2\" class=\"InternalRef\"\u003e2\u003c/span\u003eB) further confirmed the low correlations among the remaining variables.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe clinical characteristics of patients with and without mechanical complications in AMI are summarized in Supplementary Table \u003cspan refid=\"MOESM1\" class=\"InternalRef\"\u003eS1\u003c/span\u003e. Among the 850 AMI patients included in the derivation cohort, 239 (28%) developed mechanical complications during CCU hospitalization. At the time of initial medical contact, patients in the mechanical complications group were older, with a median age of 73 years (interquartile range [IQR]: 66.5\u0026ndash;79), compared with 64 years (IQR: 53.5\u0026ndash;71) in the non-mechanical complications group (P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). The proportion of female patients was also higher in the mechanical complications group (60.67% vs. 30.61%).\u003c/p\u003e \u003cp\u003eHematological parameters also differed significantly between groups, with NEUT, LYMPH, HGB, and MPV all showing marked between-group differences (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.001), supporting a role for inflammatory and immune dysregulation in the pathogenic process. Conventional cardiovascular biomarkers were substantially elevated in the mechanical complications group, which exhibited higher cTnI, LDH, NT-proBNP, and hsCRP levels than the non-mechanical complications group (all P\u0026thinsp;\u0026lt;\u0026thinsp;0.001). Taken together, these findings highlight the importance of early and accurate risk stratification for mechanical complications in patients with AMI.\u003c/p\u003e \u003c/div\u003e\n\u003ch3\u003eModel development and performance comparison\u003c/h3\u003e\n\u003cp\u003eEight ML models were developed using clinical variables obtained within the first 48 hours after admission to predict the occurrence of mechanical complications in AMI patients during CCU hospitalization. The areas under the receiver operating characteristic curves (AUCs) for these models, along with pairwise DeLong tests assessing the statistical significance of performance differences, are shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eA. LightGBM achieved the highest discriminative performance (AUC\u0026thinsp;=\u0026thinsp;0.9622), followed by AdaBoost (AUC\u0026thinsp;=\u0026thinsp;0.9603) and XGBoost (AUC\u0026thinsp;=\u0026thinsp;0.9595). The precision\u0026ndash;recall curves in Fig.\u0026nbsp;\u003cspan refid=\"Fig3\" class=\"InternalRef\"\u003e3\u003c/span\u003eB further indicate that LightGBM maintained the most favorable performance under the imbalanced class distribution.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eAccuracy, sensitivity, specificity, PPV, NPV, and F1 score for each model are presented in Table\u0026nbsp;\u003cspan refid=\"Tab1\" class=\"InternalRef\"\u003e2\u003c/span\u003e. Notably the optimal hyperparameters for all models were determined via cross-validation. The results indicate that the LightGBM model outperformed other models. In contrast, the MLP and RF models demonstrated relatively poorer performance.\u003c/p\u003e \u003cp\u003e \u003cdiv class=\"gridtable\"\u003e\u003ctable float=\"Yes\" id=\"Tab1\" border=\"1\"\u003e \u003ccaption language=\"En\"\u003e \u003cdiv class=\"CaptionNumber\"\u003eTable 2\u003c/div\u003e \u003cdiv class=\"CaptionContent\"\u003e \u003cp\u003ePredictive performance metrics of the different models\u003c/p\u003e \u003c/div\u003e \u003c/caption\u003e \u003ccolgroup cols=\"8\"\u003e \u003cdiv align=\"left\" class=\"colspec\" colname=\"c1\" colnum=\"1\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c2\" colnum=\"2\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c3\" colnum=\"3\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c4\" colnum=\"4\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c5\" colnum=\"5\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c6\" colnum=\"6\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c7\" colnum=\"7\"\u003e\u003c/div\u003e \u003cdiv align=\"char\" char=\".\" class=\"colspec\" colname=\"c8\" colnum=\"8\"\u003e\u003c/div\u003e \u003cthead\u003e \u003ctr\u003e \u003cth align=\"left\" colname=\"c1\"\u003e \u003cp\u003eModel\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c2\"\u003e \u003cp\u003eAccuracy\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c3\"\u003e \u003cp\u003eSensitivity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c4\"\u003e \u003cp\u003eSpecificity\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c5\"\u003e \u003cp\u003ePPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c6\"\u003e \u003cp\u003eNPV\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c7\"\u003e \u003cp\u003eF1\u003c/p\u003e \u003c/th\u003e \u003cth align=\"left\" colname=\"c8\"\u003e \u003cp\u003eBrier\u003c/p\u003e \u003c/th\u003e \u003c/tr\u003e \u003c/thead\u003e \u003ctbody\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eXGBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9098\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8056\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9508\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8657\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9255\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8345\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0763\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLightGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9137\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8194\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9508\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8676\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9305\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8429\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0702\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eGBM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8941\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7222\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9617\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8814\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8980\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7939\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.0854\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eAdaBoost\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.9020\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.7639\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8730\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9115\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.8148\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.1592\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eMLP\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8392\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.6389\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9180\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7541\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8660\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.6917\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.1306\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eLR\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8667\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8750\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8634\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7159\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9416\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7875\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.1094\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eSVM\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8588\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.8194\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.8743\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.7195\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.9249\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.7662\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.1043\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003ctr\u003e \u003ctd align=\"left\" colname=\"c1\"\u003e \u003cp\u003eRF\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c2\"\u003e \u003cp\u003e0.8510\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c3\"\u003e \u003cp\u003e0.5833\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c4\"\u003e \u003cp\u003e0.9563\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c5\"\u003e \u003cp\u003e0.8400\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c6\"\u003e \u003cp\u003e0.8537\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c7\"\u003e \u003cp\u003e0.6885\u003c/p\u003e \u003c/td\u003e \u003ctd align=\"char\" char=\".\" colname=\"c8\"\u003e \u003cp\u003e0.1178\u003c/p\u003e \u003c/td\u003e \u003c/tr\u003e \u003c/tbody\u003e \u003c/colgroup\u003e \u003c/table\u003e\u003c/div\u003e \u003c/p\u003e \u003cp\u003eDuring feature reduction based on feature-importance ranking, the five best-performing ML models were selected for progressive feature elimination. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig4\" class=\"InternalRef\"\u003e4\u003c/span\u003e, the AUC trajectories indicate that the LightGBM model consistently maintained the highest predictive performance among these models.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe final model was selected during the feature reduction process of the LightGBM model. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig5\" class=\"InternalRef\"\u003e5\u003c/span\u003e and Supplementary Table S2, The model with 13 features represents a clear inflection point in performance, and adding more than 13 features does not yield a significant performance gain. Compared with the 58-feature model, the 13-feature model provided a favorable net benefit across a wider range of threshold probabilities; in the decision curve, the net benefit of the LightGBM model in the test set remained consistently above the reference line, indicating good clinical utility (Supplementary Fig.\u0026nbsp;1). Moreover, the area under the precision\u0026ndash;recall curve of the 13-feature model was only marginally lower than that of the 58-feature model, suggesting similarly high applicability in clinical practice (Supplementary Fig.\u0026nbsp;2). The calibration curve of the 13-feature model demonstrated good agreement between predicted and observed probabilities, and the relatively low Brier score further indicated the absence of substantial overfitting (Supplementary Fig.\u0026nbsp;3and 4).\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cdiv id=\"Sec11\" class=\"Section2\"\u003e \u003ch2\u003eEvaluation of the optimal model\u003c/h2\u003e \u003cp\u003eFinally, we focused on the 13-feature LightGBM model, which included the following predictors: age, hsCRP, history of alcohol use, D-dimer, sex, NEUT, PCI, MPV, history of chronic pulmonary disease, SGLT2i use, LDH, AMI type, and MA. This 13-feature model was retained as the final model for subsequent analyses. The LightGBM model achieved an AUC of 0.9587 for predicting mechanical complications in AMI, demonstrating excellent discrimination between cases with and without mechanical complications. Overall classification performance was robust, with an accuracy of 0.898 and an F1 score of 0.8116, indicating a good balance between precision and recall. The sensitivity was 0.7779, reflecting a strong ability to identify patients with mechanical complications, and the specificity was 0.9454, indicating effective exclusion of patients without mechanical complications. These results support the reliability of the model\u0026rsquo;s predictions and its potential utility for early screening and risk stratification in patients with AMI. Accordingly, the 13-feature LightGBM model was selected as the optimal model for this study.\u003c/p\u003e \u003cp\u003eAfter hyperparameter tuning, the final LightGBM model was configured as follows: random_state\u0026thinsp;=\u0026thinsp;42, learning_rate\u0026thinsp;=\u0026thinsp;0.173, max_depth\u0026thinsp;=\u0026thinsp;2, min_child_weight\u0026thinsp;=\u0026thinsp;5, n_estimators\u0026thinsp;=\u0026thinsp;30, subsample\u0026thinsp;=\u0026thinsp;0.8, feature_fraction\u0026thinsp;=\u0026thinsp;0.8, and eval_metric = \"logloss\". To further assess the model\u0026rsquo;s robustness to inter-center variability, additional cross-validation was performed. As shown in Supplementary Fig.\u0026nbsp;5A and 5B, the mean AUCs of the final model in five-fold and ten-fold cross-validation were 0.9548\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0106 and 0.9485\u0026thinsp;\u0026plusmn;\u0026thinsp;0.0115, respectively, confirming the stability and reliability of the model.\u003c/p\u003e \u003c/div\u003e \u003cdiv id=\"Sec12\" class=\"Section2\"\u003e \u003ch2\u003eModel explanation\u003c/h2\u003e \u003cp\u003eBecause data-driven predictive models are often difficult to interpret directly, their broader clinical adoption can be limited. SHAP provides two complementary forms of interpretability: global, feature-level explanations and local, individual-level explanations. A radar chart based on mean absolute SHAP values, scaled to the maximum value, is shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eA. This plot quantifies and visualizes each feature\u0026rsquo;s contribution to the model output, providing an intuitive overview of how individual predictors influence predictions across different value ranges. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig6\" class=\"InternalRef\"\u003e6\u003c/span\u003eB, age has the greatest impact on SHAP values. Further analysis showed that blood-related features such as NEUT, MPV, and D-dimer also substantially influence prediction outcomes. This suggests that even in the absence of disease-specific biomarkers, routinely available clinical variables can still provide strong predictive capability.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eSHAP dependence plots illustrate how individual features influence model predictions. As shown in Fig.\u0026nbsp;\u003cspan refid=\"Fig7\" class=\"InternalRef\"\u003e7\u003c/span\u003e, increasing age, hsCRP, NEUT, D-dimer, and LDH has a marked positive effect on the model\u0026rsquo;s risk estimates, indicating that higher values of these variables are associated with an increased likelihood of mechanical complications. These findings highlight the multifactorial and nonlinear effects of routine clinical indicators on model outputs in critically ill patients with AMI.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003cp\u003eThe case of a patient who did not develop mechanical complications during CCU hospitalization is illustrated in Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eA\u0026ndash;C. In this patient, age, hsCRP, sex, D-dimer, PCI, MPV, history of chronic pulmonary disease, LDH, and MA predominantly acted as risk-attenuating factors. The final model output was \u0026minus;\u0026thinsp;0.97, corresponding to a low estimated probability of mechanical complications. The waterfall plot further displays the actual observed values for each feature, demonstrating that most variables (e.g., age, hsCRP) fell within normal or lower-risk ranges, thereby shifting the prediction toward the non-mechanical complications category. Notably, even in the presence of otherwise favorable feature values, a history of alcohol use alone was sufficient to increase the estimated risk, although not to a degree that altered the final classification, which remained \u0026ldquo;non- mechanical complications\u0026rdquo;.\u003c/p\u003e \u003cp\u003eThe waterfall plot for a patient who developed mechanical complications during CCU hospitalization is shown in Figs.\u0026nbsp;\u003cspan refid=\"Fig8\" class=\"InternalRef\"\u003e8\u003c/span\u003eD\u0026ndash;F. In this case, hsCRP and a history of chronic pulmonary disease were the dominant factors driving the prediction toward higher risk. Additional contributors\u0026mdash;including age, D-dimer, history of alcohol use, MA, NEUT, SGLT2i use, LDH, and PCI\u0026mdash;further enhanced the model output, collectively indicating a significant likelihood of mechanical complications. The final model output was 4.721, corresponding to a high predicted risk of mechanical complications for this patient.\u003c/p\u003e \u003cp\u003e \u003c/p\u003e \u003c/div\u003e"},{"header":"Discussion","content":"\u003cp\u003eThis retrospective, single-center study aimed to develop and compare eight ML models for predicting mechanical complications in patients with AMI. Predictors readily available from the EMR were extracted, and ML-based feature selection was used to identify the most informative variables. The overarching goal was to provide a more accurate and clinically applicable prediction tool to support early detection of mechanical complications and to guide individualized management and risk stratification [\u003cspan additionalcitationids=\"CR27\" citationid=\"CR26\" class=\"CitationRef\"\u003e26\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR28\" class=\"CitationRef\"\u003e28\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eLightGBM is a gradient boosting decision tree framework that performs particularly well on large, structured datasets, offering fast training and addressing key limitations of conventional decision tree algorithms[\u003cspan citationid=\"CR29\" class=\"CitationRef\"\u003e29\u003c/span\u003e]. It maintains high predictive accuracy even in large but incomplete datasets[\u003cspan citationid=\"CR30\" class=\"CitationRef\"\u003e30\u003c/span\u003e], making it well suited to predicting mechanical complications in AMI, a rapidly evolving and complex condition.\u003c/p\u003e \u003cp\u003eAlthough ML models often function as \u0026ldquo;black boxes,\u0026rdquo; limiting clinicians\u0026rsquo; willingness to rely on their predictions, this limitation is mitigated in our study by integrating SHAP[\u003cspan citationid=\"CR31\" class=\"CitationRef\"\u003e31\u003c/span\u003e]. SHAP analysis identified age as the most influential predictor of early mechanical complications in AMI, with substantially higher average SHAP values than other features, consistent with the clinical understanding that age-related myocardial and vascular sclerosis and functional decline reduce cardiovascular reserve and increase susceptibility to mechanical complications[\u003cspan citationid=\"CR32\" class=\"CitationRef\"\u003e32\u003c/span\u003e, \u003cspan citationid=\"CR33\" class=\"CitationRef\"\u003e33\u003c/span\u003e].\u003c/p\u003e \u003cp\u003eAdditionally, SHAP values highlighted hsCRP as a key predictor, consistent with its role as an inflammatory marker associated with myocardial necrosis [\u003cspan additionalcitationids=\"CR35\" citationid=\"CR34\" class=\"CitationRef\"\u003e34\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR36\" class=\"CitationRef\"\u003e36\u003c/span\u003e]. Excessive and irregular alcohol consumption has also been linked to increased risks of ischemic heart disease and hypertension, further aggravating cardiovascular damag[\u003cspan citationid=\"CR37\" class=\"CitationRef\"\u003e37\u003c/span\u003e, \u003cspan citationid=\"CR38\" class=\"CitationRef\"\u003e38\u003c/span\u003e]. In line with the GRACE registry, which reports a higher incidence of mechanical complications in STEMI than in NSTEMI patients[\u003cspan additionalcitationids=\"CR40\" citationid=\"CR39\" class=\"CitationRef\"\u003e39\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR41\" class=\"CitationRef\"\u003e41\u003c/span\u003e], our study likewise found STEMI, female sex, and delayed PCI to be important risk factors [\u003cspan additionalcitationids=\"CR43\" citationid=\"CR42\" class=\"CitationRef\"\u003e42\u003c/span\u003e\u0026ndash;\u003cspan citationid=\"CR44\" class=\"CitationRef\"\u003e44\u003c/span\u003e]. Moreover, we identified several less widely recognized predictors\u0026mdash;D-dimer, NEUT, MPV, LDH, history of chronic pulmonary disease, SGLT2i use, and MA, which may provide additional value for early identification and risk stratification of patients at risk for mechanical complications in AMI.\u003c/p\u003e \u003cp\u003eAn increase in NEUT and a decrease in MPV may reflect intense inflammatory responses. Patients with a history of chronic pulmonary disease often experience chronic hypoxia, inflammation, and oxidative stress, significantly increasing cardiac fragility and increasing cardiovascular disease risk[\u003cspan citationid=\"CR45\" class=\"CitationRef\"\u003e45\u003c/span\u003e]. LDH, which originates from myocardial, skeletal muscle, liver, red blood cells, and intestines[\u003cspan citationid=\"CR46\" class=\"CitationRef\"\u003e46\u003c/span\u003e, \u003cspan citationid=\"CR47\" class=\"CitationRef\"\u003e47\u003c/span\u003e], is a marker for systemic organ hypoperfusion and exacerbated myocardial necrosis[\u003cspan citationid=\"CR48\" class=\"CitationRef\"\u003e48\u003c/span\u003e]. Data from the international myocardial infarction and diabetes registry and the EMPULSE trial confirm that patients who start SGLT2i treatment during hospitalization after MI experience significantly reduced cardiovascular death, arrhythmia burden, and acute kidney injury [\u003cspan citationid=\"CR49\" class=\"CitationRef\"\u003e49\u003c/span\u003e, \u003cspan citationid=\"CR50\" class=\"CitationRef\"\u003e50\u003c/span\u003e]. MA, particularly ventricular fibrillation and sustained ventricular tachycardia, often signify extensive infarct size and severe myocardial ischemia and necrosis. Although current literature does not specifically identify these factors as independent predictors of mechanical complications in AMI, our study confirms that when combined with other factors, they are associated with an increased risk of mechanical complications in acute myocardial infarction.\u003c/p\u003e \u003cp\u003eWe acknowledge several limitations of this study. First, the temporal span of the data may introduce heterogeneity between samples, potentially affecting the stability and generalizability of the model. Future studies should incorporate prospective data to further validate the model\u0026rsquo;s predictability and extend its applicability to diverse populations and disease states. Although ML techniques require large datasets for model development, no established standard exists for calculating the sample size for machine learning-based predictive models. Nevertheless, robust cross-validation indicates that the sample size is appropriate.\u003c/p\u003e \u003cp\u003eIn conclusion, we developed an interpretable ML model based on routinely available EMR data to predict mechanical complications in patients with AMI, with the final LightGBM model demonstrating excellent discriminatory performance. Using an optimal probability threshold, critically ill patients identified as high risk can be prioritized for closer monitoring and timely intervention. Prospective randomized trials are needed to determine whether model-guided, individualized management can improve outcomes in AMI-related mechanical complications.\u003c/p\u003e"},{"header":"Declarations","content":" \u003cp\u003e \u003cstrong\u003eEthics approval and consent to participate\u003c/strong\u003e \u003cp\u003e This study was conducted in accordance with the ethical principles of the Declaration of Helsinki. The study was approved by the Ethics Committee of the First Affiliated Hospital of Harbin Medical University, ethical number: 2025369, and the right to exempt informed consent was obtained.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eConsent for publication\u003c/strong\u003e \u003cp\u003eNot applicable.\u003c/p\u003e \u003c/p\u003e \u003cp\u003e \u003cstrong\u003eCompeting interests\u003c/strong\u003e \u003cp\u003eThe authors declare no competing interests.\u003c/p\u003e \u003c/p\u003e\u003ch2\u003eFunding\u003c/h2\u003e \u003cp\u003eThis study was supported by the Natural Science Foundation of Heilongjiang Province (PL2024H032)\u003c/p\u003e\u003ch2\u003eAuthor Contribution\u003c/h2\u003e\u003cp\u003eY.C. and J.Y. designed and experimented and participated in the data collection, and was a major contributor in writing the manuscript. Y.L. and G.C. collected data on patients with AMI, while X.D. and J.Z. analyzed data on patients with mechanical complications in patients with AMI. J.W. is responsible for organizing the data. J.L. and L.S., as corresponding authors, guided the progress of the study throughout the process to ensure the authenticity of the data. All authors reviewed the manuscript.\u003c/p\u003e\u003ch2\u003eAcknowledgements\u003c/h2\u003e \u003cp\u003eNot applicable.\u003c/p\u003e\u003ch2\u003eData Availability\u003c/h2\u003e\u003cp\u003eThe datasets used and/or analysed during the current study available from the corresponding author on reasonable request.\u003c/p\u003e"},{"header":"References","content":"\u003col\u003e\u003cli\u003e\u003cspan\u003ePuerto E, Viana-Tejedor A, Mart\u0026iacute;nez-Sell\u0026eacute;s M, et al. Temporal trends in mechanical complications of acute myocardial infarction in the elderly[J]. J Am Coll Cardiol. 2018;72(9):959\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jcin.2019.04.039\u003c/span\u003e\u003cspan address=\"10.1016/j.jcin.2019.04.039\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBirnbaum Y, Fishbein MC, Blanche C, et al. Ventricular septal rupture after acute myocardial infarction[J]. N Engl J Med. 2002;347(18):1426\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1056/NEJMra020228\u003c/span\u003e\u003cspan address=\"10.1056/NEJMra020228\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eElbadawi A, Elgendy IY, Mahmoud K, et al. Temporal trends and outcomes of mechanical complications in patients with acute myocardial infarction[J]. JACC: Cardiovasc Interventions. 2019;12(18):1825\u0026ndash;36. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jcin.2019.04.039\u003c/span\u003e\u003cspan address=\"10.1016/j.jcin.2019.04.039\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGong FF, Vaitenas I, Malaisrie SC, et al. Mechanical complications of acute myocardial infarction: a review[J]. JAMA Cardiol. 2021;6(3):341\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1001/jamacardio.2020.3690\u003c/span\u003e\u003cspan address=\"10.1001/jamacardio.2020.3690\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDamluji AA, Forman DE, Van Diepen S, et al. Older adults in the cardiac intensive care unit: factoring geriatric syndromes in the management, prognosis, and process of care: a scientific statement from the American Heart Association[J]. Circulation. 2020;141(2):e6\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIR.0000000000000741\u003c/span\u003e\u003cspan address=\"10.1161/CIR.0000000000000741\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eFu Y, Li KB, Yang XC. A risk score model for predicting cardiac rupture after acute myocardial infarction[J]. Chin Med J. 2019;132(09):1037\u0026ndash;44. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1097/CM9.0000000000000175\u003c/span\u003e\u003cspan address=\"10.1097/CM9.0000000000000175\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaito Y, Shiko Y, Tateishi K, et al. Combined Risk Stratification With Patient Characteristics and Biomarkers in Patients Treated With the Impella for Cardiogenic Shock[J]. J Am Heart Association. 2025;14(9):e040487. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/JAHA.124.040487\u003c/span\u003e\u003cspan address=\"10.1161/JAHA.124.040487\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCheng X, Manandhar I, Aryal S, et al. Application of artificial intelligence in cardiovascular medicine[J]. Compr Physiol. 2021;11(4):2455\u0026ndash;66. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/j.2040-4603.2021.tb00186.x\u003c/span\u003e\u003cspan address=\"10.1002/j.2040-4603.2021.tb00186.x\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoecks J, Jalili V, Heiser LM et al. How machine learning will transform biomedicine[J].Cell,2020,181(1):92101.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cell.2020.03.022\u003c/span\u003e\u003cspan address=\"10.1016/j.cell.2020.03.022\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGould MK, Huang BZ, Tammemagi MC, et al. Machine learning for early lung cancer identification using routine clinical and laboratory data[J]. Am J Respir Crit Care Med. 2021;204(4):445\u0026ndash;53. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1164/rccm.202007-2791OC\u003c/span\u003e\u003cspan address=\"10.1164/rccm.202007-2791OC\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGao J, Lu Y, Ashrafi N, et al. Prediction of sepsis mortality in ICU patients using machine learning methods[J]. BMC Med Inf Decis Mak. 2024;24(1):228. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-024-02630-z\u003c/span\u003e\u003cspan address=\"10.1186/s12911-024-02630-z\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhou H, Li F, Liu X. Early prediction of septic shock in ICU patients using machine learning: development, external validation, and explainability with SHAP[J]. Int J Med Informatics. 2025;106169. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.ijmedinf.2025.106169\u003c/span\u003e\u003cspan address=\"10.1016/j.ijmedinf.2025.106169\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eWang S, Li J, Sun L, et al. Application of machine learning to predict the occurrence of arrhythmia after acute myocardial infarction[J]. BMC Med Inf Decis Mak. 2021;21(1):301.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAl Azwari S. Predicting myocardial rupture after acute myocardial infarction in hospitalized patients using machine learning[C]//2021 National Computing Colleges Conference (NCCC). IEEE, 2021: 1\u0026ndash;6.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eByrne RA, Rossello X, Coughlan JJ, ESC Scientific Document Group, et al. Linee guida ESC 2023 per il trattamento delle sindromi coronariche acute [2023 ESC Guidelines for the management of acute coronary syndromes]. G Ital Cardiol (Rome). 2024;25(2 Suppl 2):e1\u0026ndash;112. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1714/4191.41785\u003c/span\u003e\u003cspan address=\"10.1714/4191.41785\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e. Italian.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eThygesen K, Alpert JS, Jaffe AS, et al. Fourth universal definition of myocardial infarction (2018)[J]. J Am Coll Cardiol. 2018;72(18):2231\u0026ndash;64. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jacc.2018.08.1038\u003c/span\u003e\u003cspan address=\"10.1016/j.jacc.2018.08.1038\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eChinese Society of Cardiology, Chinese Journal of Cardiovascular Diseases Editorial Committee. Guidelines for the Diagnosis and Treatment of Acute ST-segment Elevation Myocardial Infarction. (2019) [J]. Chinese Journal of Cardiovascular Diseases, 2019, 47(10): 766\u0026ndash;783. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.3760/cma.j.issn.0253-3758.2019.10.003\u003c/span\u003e\u003cspan address=\"10.3760/cma.j.issn.0253-3758.2019.10.003\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDamluji AA, Van Diepen S, Katz JN, et al. Mechanical complications of acute myocardial infarction: a scientific statement from the American Heart Association[J]. Circulation. 2021;144(2):e16\u0026ndash;35. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIR.0000000000000985\u003c/span\u003e\u003cspan address=\"10.1161/CIR.0000000000000985\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eBerkelmans GFN, Read SH, Gudbj\u0026ouml;rnsdottir S, et al. Population median imputation was noninferior to complex approaches for imputing missing values in cardiovascular prediction models in clinical practice[J]. J Clin Epidemiol. 2022;145:70\u0026ndash;80. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.jclinepi.2022.01.011\u003c/span\u003e\u003cspan address=\"10.1016/j.jclinepi.2022.01.011\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHe H, Garcia EA. Learning from imbalanced data[J]. IEEE Trans Knowl Data Eng. 2009;21(9):1263\u0026ndash;84. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/TKDE.2008.239\u003c/span\u003e\u003cspan address=\"10.1109/TKDE.2008.239\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eHoens TR, Chawla NV. Imbalanced datasets: from sampling to classifiers[J]. Imbalanced learning: Foundations, algorithms, and applications, 2013: 43\u0026ndash;59.\u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1002/9781118646106.ch3\u003c/span\u003e\u003cspan address=\"10.1002/9781118646106.ch3\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDavis J, Goadrich M. The relationship between Precision-Recall and ROC curves[C]//Proceedings of the 23rd international conference on Machine learning. 2006: 233\u0026ndash;240. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1145/1143844.1143874\u003c/span\u003e\u003cspan address=\"10.1145/1143844.1143874\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLundberg SM, Lee SI. A unified approach to interpreting model predictions[J]. Adv Neural Inf Process Syst, 2017, 30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDeLong ER, DeLong DM, Clarke-Pearson DL. Comparing the areas under two or more correlated receiver operating characteristic curves: a nonparametric approach[J]. Biometrics, 1988: 837\u0026ndash;45.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eAzodi CB, Tang J, Shiu SH. Opening the black box: interpretable machine learning for geneticists[J]. Trends Genet. 2020;36(6):442\u0026ndash;55. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.tig.2020.03.005\u003c/span\u003e\u003cspan address=\"10.1016/j.tig.2020.03.005\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRajkomar A, Dean J, Kohane I. Machine Learning in Medicine.N. Engl. J Med. 2019;380:1347\u0026ndash;58.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZhang X, Yan C, Gao C, et al. Predicting missing values in medical data via XGBoost regression[J]. J Healthc Inf Res. 2020;4(4):383\u0026ndash;94. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s41666-020-00077-1\u003c/span\u003e\u003cspan address=\"10.1007/s41666-020-00077-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eLi Y, Sperrin M, Ashcroft DM, et al. Consistency of variety of machine learning and statistical models in predicting clinical risks of individual patients: longitudinal cohort study using cardiovascular disease as exemplar[J]. BMJ. 2020;371. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/bmj.m3919\u003c/span\u003e\u003cspan address=\"10.1136/bmj.m3919\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eKe G, Meng Q, Finley T et al. Lightgbm: A highly efficient gradient boosting decision tree[J]. Adv Neural Inf Process Syst, 2017, 30.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYan J, Xu Y, Cheng Q et al. LightGBM: accelerated genomically designed crop breeding through ensemble learning[J]. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s13059-021-02492-y\u003c/span\u003e\u003cspan address=\"10.1186/s13059-021-02492-y\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMarc\u0026iacute;lio WE, Eler DM. From explanations to feature selection: assessing SHAP values as feature selection mechanism[C]//2020 33rd SIBGRAPI conference on Graphics, Patterns and Images (SIBGRAPI). Ieee, 2020: 340\u0026ndash;347. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1109/SIBGRAPI51738.2020.00053\u003c/span\u003e\u003cspan address=\"10.1109/SIBGRAPI51738.2020.00053\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eValley TS, Sjoding MW, Ryan AM, et al. Intensive care unit admission and survival among older patients with chronic obstructive pulmonary disease, heart failure, or myocardial infarction[J]. Annals Am Thorac Soc. 2017;14(6):943\u0026ndash;51. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1513/AnnalsATS.201611-847OC\u003c/span\u003e\u003cspan address=\"10.1513/AnnalsATS.201611-847OC\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eSaaby L, Poulsen TS, Hosbond S, et al. Classification of myocardial infarction: frequency and features of type 2 myocardial infarction[J]. Am J Med. 2013;126(9):789\u0026ndash;97. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.amjmed.2013.02.029\u003c/span\u003e\u003cspan address=\"10.1016/j.amjmed.2013.02.029\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eUeda S, Ikeda U, Yamamoto K, et al. C-reactive protein as a predictor of cardiac rupture after acute myocardial infarction[J]. Am Heart J. 1996;131(5):857\u0026ndash;60.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMiller M, Zhan M, Havas S. High attributable risk of elevated C-reactive protein level to conventional coronary heart disease risk factors: the Third National Health and Nutrition Examination Survey[J]. Arch Intern Med. 2005;165(18):2063\u0026ndash;8.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMaier W, Altwegg LA, Corti R, et al. Inflammatory markers at the site of ruptured plaque in acute myocardial infarction: locally increased interleukin-6 and serum amyloid A but decreased C-reactive protein[J]. Circulation. 2005;111(11):1355\u0026ndash;61. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/01.CIR.0000158479.58589.0A\u003c/span\u003e\u003cspan address=\"10.1161/01.CIR.0000158479.58589.0A\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eRoerecke M, Rehm J. Chronic heavy drinking and ischaemic heart disease: a systematic review and meta-analysis[J]. Open heart. 2014;1(1):e000135. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1136/openhrt-2014-000135\u003c/span\u003e\u003cspan address=\"10.1136/openhrt-2014-000135\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eVaideeswar P, Chaudhari C, Rane S, et al. Cardiac pathology in chronic alcoholics: a preliminary study[J]. J Postgrad Med. 2014;60(4):372\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.4103/0022-3859.143958\u003c/span\u003e\u003cspan address=\"10.4103/0022-3859.143958\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eL\u0026oacute;pez-Send\u0026oacute;n J, Gurfinkel EP, Lopez de Sa E, et al. Factors related to heart rupture in acute coronary syndromes in the Global Registry of Acute Coronary Events[J]. Eur Heart J. 2010;31(12):1449\u0026ndash;56. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1093/eurheartj/ehq089\u003c/span\u003e\u003cspan address=\"10.1093/eurheartj/ehq089\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrenshaw BS, Granger CB, Birnbaum Y, et al. Risk factors, angiographic patterns, and outcomes in patients with ventricular septal defect complicating acute myocardial infarction[J]. Circulation. 2000;101(1):27\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/01.CIR.101.1.27\u003c/span\u003e\u003cspan address=\"10.1161/01.CIR.101.1.27\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eGoldberg RJ, Spencer FA, Gore JM, et al. Thirty-year trends (1975 to 2005) in the magnitude of, management of, and hospital death rates associated with cardiogenic shock in patients with acute myocardial infarction: a population-based perspective[J]. Circulation. 2009;119(9):1211\u0026ndash;9. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIRCULATIONAHA.108.814947\u003c/span\u003e\u003cspan address=\"10.1161/CIRCULATIONAHA.108.814947\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eMenon V, Webb JG, Hillis LD, et al. Outcome and profile of ventricular septal rupture with cardiogenic shock after myocardial infarction: a report from the SHOCK Trial Registry[J]. J Am Coll Cardiol. 2000;36(3S1):1110\u0026ndash;6. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/S0735-1097(00)00878-0\u003c/span\u003e\u003cspan address=\"10.1016/S0735-1097(00)00878-0\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eCrenshaw BS, Granger CB, Birnbaum Y, et al. Risk factors, angiographic patterns, and outcomes in patients with ventricular septal defect complicating acute myocardial infarction[J]. Circulation. 2000;101(1):27\u0026ndash;32. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/01.CIR. 101.1.27\u003c/span\u003e\u003cspan address=\"10.1161/01.CIR. 101.1.27\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eYang Y, Tang J, Ma L, et al. A systematic comparison of short-term and long-term mortality prediction in acute myocardial infarction using machine learning models[J]. BMC Med Inf Decis Mak. 2025;25(1):208. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-025-03052-1\u003c/span\u003e\u003cspan address=\"10.1186/s12911-025-03052-1\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eDunlay SM, Chamberlain AM. Multimorbidity in older patients with cardiovascular disease[J]. Curr Cardiovasc risk Rep. 2016;10(1):3. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1007/s12170-016-0491-8\u003c/span\u003e\u003cspan address=\"10.1007/s12170-016-0491-8\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eStorey KB. Comparative enzymology\u0026mdash;new insights from studies of an old enzyme, lactate dehydrogenase[J]. Comp Biochem Physiol B: Biochem Mol Biol. 2016;199:13\u0026ndash;20. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1016/j.cbpb.2015.12.004\u003c/span\u003e\u003cspan address=\"10.1016/j.cbpb.2015.12.004\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eOliveira M, Seringa J, Pinto FJ, et al. Machine learning prediction of mortality in acute myocardial infarction[J]. BMC Med Inf Decis Mak. 2023;23(1):70. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12911-023-02168-6\u003c/span\u003e\u003cspan address=\"10.1186/s12911-023-02168-6\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003eZweck E, Hassager C, Beske RP, et al. Microaxial flow pump use and renal outcomes in infarct-related cardiogenic shock: a secondary analysis of the DanGer Shock Trial[J]. Circulation. 2024;150(25):1990\u0026ndash;2003. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1161/CIRCULATIONAHA.124.072370\u003c/span\u003e\u003cspan address=\"10.1161/CIRCULATIONAHA.124.072370\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003ePaolisso P, Bergamaschi L, Gragnano F, et al. Outcomes in diabetic patients treated with SGLT2-Inhibitors with acute myocardial infarction undergoing PCI: The SGLT2-I AMI PROTECT Registry[J]. Pharmacol Res. 2023;187:106597.\u003c/span\u003e\u003c/li\u003e \u003cli\u003e\u003cspan\u003evon Lewinski D, Kolesnik E, Aziz F, et al. Timing of SGLT2i initiation after acute myocardial infarction[J]. Cardiovasc Diabetol. 2023;22(1):269. \u003cspan class=\"ExternalRef\"\u003e\u003cspan class=\"RefSource\"\u003ehttps://doi.org/10.1186/s12933-023-02000-5\u003c/span\u003e\u003cspan address=\"10.1186/s12933-023-02000-5\" targettype=\"DOI\" class=\"RefTarget\"\u003e\u003c/span\u003e\u003c/span\u003e.\u003c/span\u003e\u003c/li\u003e\u003c/ol\u003e"}],"fulltextSource":"","fullText":"","funders":[],"hasAdminPriorityOnWorkflow":false,"hasManuscriptDocX":true,"hasOptedInToPreprint":true,"hasPassedJournalQc":"","hasAnyPriority":false,"hideJournal":true,"highlight":"","institution":"","isAcceptedByJournal":false,"isAuthorSuppliedPdf":false,"isDeskRejected":"","isHiddenFromSearch":false,"isInQc":false,"isInWorkflow":false,"isPdf":false,"isPdfUpToDate":true,"isWithdrawnOrRetracted":false,"journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true},"keywords":"","lastPublishedDoi":"10.21203/rs.3.rs-8461400/v1","lastPublishedDoiUrl":"https://doi.org/10.21203/rs.3.rs-8461400/v1","license":{"name":"CC BY 4.0","url":"https://creativecommons.org/licenses/by/4.0/"},"manuscriptAbstract":"\u003ch2\u003eBackground\u003c/h2\u003e \u003cp\u003eMechanical complications in acute myocardial infarction (AMI) progress rapidly and carry very high mortality, underscoring the need for early risk prediction. Existing studies use a narrow range of models with poorly characterized decision-making. This study aimed to develop and validate an interpretable prediction model for post-AMI mechanical complications to guide individualized treatment and optimize resource allocation.\u003c/p\u003e\u003ch2\u003eMethods\u003c/h2\u003e \u003cp\u003eA total of 850 patients with and without mechanical complications enrolled in this study. 58 features were selected for model training and validation. Eight machine learning algorithms were used to build prediction models, whose performance was assessed AUC, accuracy, F1 score and other indicators. The SHAP method was applied to rank feature importance and interpret the final model.\u003c/p\u003e\u003ch2\u003eResults\u003c/h2\u003e \u003cp\u003eAmong the eight machine learning models, LightGBM showed the best discriminative performance. After feature reduction, a 13-variable interpretable LightGBM model was established, achieving excellent discrimination for mechanical complications in the validation set (AUC\u0026thinsp;=\u0026thinsp;0.9587). SHAP analysis identified age, hsCRP (High-Sensitivity C-Reactive Protein), drinking history, D-dimer, sex, NEUT (Neutrophils), PCI (Percutaneous Coronary Intervention), MPV (Mean Platelet Volume), history of chronic lung disease, SGLT2i (Sodium-Glucose Cotransporter 2 Inhibitors) use, LDH (Lactate Dehydrogenase), AMI category, and MA (Malignant Arrhythmia) as the most influential predictors.\u003c/p\u003e\u003ch2\u003eConclusions\u003c/h2\u003e \u003cp\u003eThe interpretable ML model provides both global and patient-level explanations, and the simplified key-feature model is suitable for deployment as a clinical decision-support tool for rapid screening and risk assessment in emergency settings.\u003c/p\u003e","manuscriptTitle":"Predicting the risk of mechanical complications in acute myocardial infarction using an interpretable machine learning model","msid":"","msnumber":"","nonDraftVersions":[{"code":1,"date":"2026-01-22 14:23:45","doi":"10.21203/rs.3.rs-8461400/v1","editorialEvents":[{"type":"communityComments","content":0}],"status":"published","journal":{"display":true,"email":"[email protected]","identity":"researchsquare","isNatureJournal":false,"hasQc":true,"allowDirectSubmit":true,"externalIdentity":"","sideBox":"","snPcode":"","submissionUrl":"/submission","title":"Research Square","twitterHandle":"researchsquare","acdcEnabled":true,"dfaEnabled":false,"editorialSystem":"","reportingPortfolio":"","inReviewEnabled":false,"inReviewRevisionsEnabled":true}}],"origin":"","ownerIdentity":"17b12be5-a219-4021-8ad8-86b8506cd1fe","owner":[],"postedDate":"January 22nd, 2026","published":true,"recentEditorialEvents":[],"rejectedJournal":[],"revision":"","amendment":"","status":"posted","subjectAreas":[],"tags":[],"updatedAt":"2026-05-18T03:55:39+00:00","versionOfRecord":[],"versionCreatedAt":"2026-01-22 14:23:45","video":"","vorDoi":"","vorDoiUrl":"","workflowStages":[]},"version":"v1","identity":"rs-8461400","journalConfig":"researchsquare"},"__N_SSP":true},"page":"/article/[identity]/[[...version]]","query":{"redirect":"/article/rs-8461400","identity":"rs-8461400","version":["v1"]},"buildId":"XKTyCvWXoU3ODBz1xrDgd","isFallback":false,"isExperimentalCompile":false,"dynamicIds":[84888],"gssp":true,"scriptLoader":[]}