In recent years, the prevalence and mortality of acute myocardial infarction (AMI) have tended to be younger and are the leading cause of premature death worldwide [1], with about 4%–10% of AMI patients reported to be aged $\le$40 or 45 years [2, 3]. The increase of metabolic risk factors in young people, such as obesity, diabetes, high uric acid, and hypertension, has increased the incidence of premature myocardial infarction (PMI) and major adverse cardiovascular events (MACEs) [4], which seriously affect the workability and quality of life of patients, causing a certain burden on families and social economy. Obtaining accurate risk prediction of long-term MACEs after PMI, and therefore early intervention to improve patient prognosis as much as possible, is of utmost importance in clinical management [5, 6, 7, 8].

Machine learning (ML) algorithms provide powerful tools for researchers to learn rules in data and make data-driven outcome predictions by capturing high-dimensional, linear, or non-linear relationships between clinical variables [9]. ML has been used in many medical-related fields, such as diagnosis, outcome prediction, treatment, and medical image interpretation, and is superior to proven traditional risk stratification tools [10, 11, 12, 13, 14]. For example, a study using the American College of Cardiology Chest Pain-MI registry that used an ML model to predict death after AMI reported an area under the curve (AUC) value of close to 0.9 for each ML model, with extreme gradient boosting (XGBoost) provide better risk solutions for high-risk individuals [15]. Another ML-based study of adverse event prediction in acute coronary syndrome (apolipoprotein A1/B, ApoA1/B) showed that different machine learning models showed good predictive performance in predicting all-cause death, myocardial infarction, and major bleeding in acute coronary syndrome (ACS) patients at 1 year after discharge, and compared with traditional risk prediction tools, ML algorithm has advantages in predicting MACEs [16].

Compared with elderly MI patients, young myocardial infarction (MI) patients may have a different risk factor spectrum, and PMI patients often have other unique metabolic risk factors [17]. There are still few studies on the related risk factors affecting the occurrence of AMI adverse events in young adults, and the previous studies using machine learning algorithms to establish MACEs prediction models in young MI patients are also limited. Therefore, the development of machine learning predictive models for these patients to guide early clinical intervention has certain research value. In summary, the purpose of this study was to use machine learning algorithms to assess and stratify the risk of long-term MACEs in PMI patients, and to analyze key clinical variables affecting the occurrence of MACEs.

This study developed and validated an interpretable ML risk prediction model for predicting the risk of long-term MACEs in PMI patients and analyzed clinical variables that influence the development of MACEs. The evaluation results of comprehensive discrimination, calibration and clinical utility showed that the RSF model performed best. Using the risk score (24.90) calculated by the RSF model as the critical value, patients were divided into high-risk group and low-risk group, and there was significant difference in Kaplan-Meier survival analysis curve between the two groups (p 0.0001). The ten clinical variables of feature importance ranking are FFA, cardiac shock, HbA1c, ApoB, Diuretics, LVEF, BNP, BMI, Killip $\ge$II, and UA. By calculating the risk of MACEs through a risk calculator and explaining individual risk sources and possible intervention directions through SHAP values, it is hoped that personalized and transparent clinical management can be achieved.

The results of this study highlight the superior performance of the RSF model in predicting the risk of MACEs in AMI patients, as evidenced by its discrimination, reliable calibration, and robust clinical utility. Compared with traditional models such as the COX proportional hazards model, RSF overcomes key limitations by capturing complex nonlinear relationships and high-order interactions without relying on the proportional hazards assumption. This adaptability is particularly valuable in real-world clinical scenarios, where these assumptions are often violated [20, 21]. Although XGBoost is a powerful machine learning algorithm, its application to survival data often requires additional modifications, such as implementing COX loss functions, which may introduce constraints, and it also demands extensive parameter tuning [22]. Similarly, DeepSurv, as a deep learning method, requires large-scale datasets to perform optimally and is prone to overfitting with limited data [23]. In contrast, RSF natively supports survival analysis, offering seamless integration, robust performance even with moderate sample sizes, and higher predictive accuracy without the need for extensive modifications or tuning, making it more practical for real-world clinical applications.

Insulin resistance (IR), elevated HbA1c, and metabolic abnormalities such as abnormal BMI, dyslipidemia, and hyperuricemia are critical contributors to myocardial injury and increased MACEs risk, particularly in young PMI patients. IR leads to myocardial damage through impaired diastole, altered glucose utilization, and microvascular dysfunction, while metabolic abnormalities like elevated BMI and dysregulated lipid metabolism trigger inflammation and thrombosis through the release and accumulation of fat metabolites [24, 25]. The J-shaped relationship between BMI and MACEs observed in this study using RCS is consistent with the findings of some other studies [26, 27], Flegal et al.’s [28] study has shown that overweight individuals have a lower risk than normal-weight individuals. This study’s findings highlight the strong association of FFA with adverse cardiovascular events, FFA assessment alongside traditional risk factors to identify high-risk individuals requiring closer monitoring and intervention. Dyslipidemia, particularly elevated ApoB-containing lipoproteins such as LDL-C, lipoprotein(a), and triglyceride-rich lipoproteins, significantly contributes to MACEs risk. While LDL-C remains a primary target for lipid management, this study’s RCS analysis aligns with prior research showing a linear relationship between higher ApoB levels and increased MACEs risk, even in patients on high-intensity statin therapy [29, 30, 31]. These findings suggest that for younger PMI patients, early and aggressive management of ApoB levels may be crucial in reducing cardiovascular risk. Elevated uric acid levels also emerged as an important predictor of MACEs. Through mechanisms such as oxidative stress, endothelial dysfunction, and inflammation, uric acid exacerbates insulin resistance and promotes atherosclerosis [32, 33, 34, 35]. Managing uric acid levels may disrupt this pathological cycle, offering an additional avenue for intervention in younger patients.

Cardiac function plays a pivotal role in determining MACEs risk. Variables in the RSF model, such as elevated BNP, reduced LVEF, Killip $\ge$II, cardiogenic shock, and in-hospital diuretic use, reflect poor cardiac function during hospitalization. This study’s RCS analysis showed a linear relationship between decreasing LVEF and increasing MACEs risk, consistent with previous research linking reduced ejection fraction with poorer outcomes in PCI-treated patients [36, 37]. These findings emphasize the importance of targeted cardiac rehabilitation and monitoring strategies for PMI patients with compromised cardiac function.

A comprehensive strategy is essential for younger PMI patients to reduce MACEs risk, combining advanced predictive tools and tailored management interventions. Aggressive control of ApoB, FFAs, and uric acid levels is crucial to address inflammation, thrombosis, and oxidative stress, while individualized BMI management mitigates the J-shaped risk relationship observed with MACEs. Targeted cardiac rehabilitation and monitoring of BNP and LVEF further enhance outcomes in patients with compromised cardiac function. The RSF model demonstrated its strength by integrating these multifactorial risks into a comprehensive predictive framework. With the addition of SHAP values, the model provides individual-level explanations, helping clinicians identify key contributing factors for each patient’s risk. Combined with a personalized risk calculator, these tools enable dynamic and patient-specific intervention strategies targeting modifiable risk factors such as IR, dyslipidemia, hyperuricemia, and cardiac dysfunction. This approach supports more effective prevention and treatment, ultimately improving long-term outcomes and reducing MACEs incidence in young PMI patients.

This study has several limitations. Most importantly, it lacks external validation with independent cohorts, which is essential for confirming the generalizability and robustness of the algorithm. In future research, we will incorporate patients from diverse regions and hospitals to perform external validation, ensuring broader applicability across different populations. Additionally, this study primarily focuses on clinical characteristics, missing key factors such as lifestyle, dietary habits, and multi-omics markers. Expanding these variables in future studies could provide a more comprehensive understanding of risk factors and enhance the predictive accuracy of the model.

The RSF-based risk stratification tool demonstrated excellent performance, proving its capability to accurately predict MACEs risk in PMI patients. The model identified critical predictors such as FFA, cardiogenic shock, HbA1c, ApoB, diuretic use, LVEF, BNP, BMI, Killip $\ge$II, and UA, highlighting the multifactorial complexity of MACEs risk. Enhanced by SHAP values and a risk calculator, the RSF model provides a personalized framework to identify high-risk patients, pinpoint key risk factors, and guide targeted interventions. This approach enables early management of modifiable risks, improving outcomes and reducing MACEs in PMI patients.

ACS, acute coronary syndrome; ALT, alanine aminotransferase; ANC, absolute neutrophil count; ApoB, apolipoprotein B; AUC, area under the curve; BNP, brain natriuretic peptide; BMI, body mass index; C-index, concordance index; CABG, coronary artery bypass grafting; CK-MB, creatine kinase MB; COX, COX proportional hazards model; Cr, creatinine; CRP, c-reactive protein; DCA, decision curve analysis; ECG, Electrocardiogram; FFA, free fatty acid; FIB, fibrinogen; Glu, glucose; HbA1c, glycated hemoglobin; HR value, hazard ratio value; IABP, intra-aortic balloon pump; LAD, left anterior descending coronary artery; LASSO, least absolute shrinkage and selection operator; LM, left main coronary artery; LDH, lactate dehydrogenase; LDL-C, low-density lipoprotein cholesterol; LVEF, left ventricular ejection fraction; MACEs, major adverse cardiovascular events; ML, machine learning; PCI, percutaneous coronary intervention; PMI, premature myocardial infarction; RCS, restricted cubic spline; ROC, receiver operating characteristic; RSF, random survival forest; SD, standard deviation; SHAP, SHapley Additive exPlanations; ST-T, ST segment and T wave; TC, total cholesterol; TNT, troponin T; TyG, triglyceride-glucose; UA, uric acid; VIF, variance inflation factor; WBC, white blood cell count; XGBoost, extreme gradient boosting.

The datasets used and/or analysed during the current study are available from the corresponding author on reasonable request.

JG, CPL, and YLiu conceived the idea and designed the study. JXW and MML made substantial contributions to data analysis, interpretation, and article writing. JXW, PJL, ZC, MML, YLiang, YHW, ARJ, JW, and MLZ participated in the data collection. PJL, JG, CPL, YLiu, and ZC participated in the drafting and critical revision of the article. All authors contributed to editorial changes in the manuscript. All authors read and approved the final version of the manuscript. All authors have participated sufficiently in the work and agreed to be accountable for all aspects of the work.

The research protocol was approved by the Ethics Committee of Tianjin Chest Hospital (Ethic Approval Number: 2017KY-007-01) and strictly followed the Declaration of Helsinki. All enrolled patients signed informed consent before participating in the study.

This research was funded by the Tianjin Health Commission Science and Technology Project (No. TJWJ2021QN058), Key Discipline Project of Tianjin Health Science and Technology Project in 2022 (No. TJWJ2022XK032) and the Key Science and Technology Support Project of Tianjin Key Research and Development Plan in 2020 (No. 20YFZCSY00820).

The authors declare no conflict of interest.

Supplementary material associated with this article can be found, in the online version, at https://doi.org/10.31083/RCM31298.