This retrospective cohort study enrolled patients with new-onset STEMI undergoing PCI between January 2020 and June 2022 at the Affiliated Jiangyin Hospital of Nantong University (n = 120) and Xintai People’s Hospital (The Affiliated Hospital of Qilu Medical College) (n = 33). The pooled data from these 153 eligible patients were subsequently randomly allocated (7:3 ratio) into a derivation cohort (n = 107) for model development and an internal validation cohort (n = 46) for performance assessment.

The study protocol was approvaled by the ethics committees of both participating institutions (Affiliated Jiangyin Hospital of Nantong University, Approval No. 2025 Lun Shen Yan No. 014; Xintai People’s Hospital/The Affiliated Hospital of Qilu Medical College, Approval No. 2025-S021) and adhered to the Declaration of Helsinki. Given the retrospective nature of the study and expedited review, the requirement for written informed consent was waived by the ethics committees of both participating institutions.

STEMI diagnosis followed the 2021 American Heart Association/American College of Cardiology/American Society of Echocardiography/American College of Chest Physicians/Society for Academic Emergency Medicine/Society of Cardiovascular Computed Tomography/Society for Cardiovascular Magnetic Resonance (AHA/ACC/ASE/CHEST/SAEM/SCCT/SCMR) Guideline for Chest Pain [16]. Key inclusion criteria were first-onset STEMI, successful PCI (TIMI flow $\ge$2), and age $\ge$18 years. Principal exclusion criteria included severe comorbidities affecting survival (active malignancy, end-stage organ failure, or terminal illness with life expectancy 1 year), prior MI or revascularization, CMR contraindications, or incomplete data/follow-up. The patient selection flowchart is presented in Fig. 1 below.

Fig. 1.

Flowchart for patient selection. Of 308 STEMI patients post-PCI, 102 were excluded (prior coronary artery disease, no CMR, poor image quality). From 206 eligible patients, 53 were lost to follow-up. The remaining 153 patients were enrolled and randomized to a derivation cohort (n = 107) for nomogram development, and a validation cohort (n = 46) for performance assessment. CAD, coronary artery disease; CMR, cardiac magnetic resonance.

Upon approval from the institutional ethics committees and obtaining informed consent, data for eligible patients with new-onset STEMI undergoing PCI were retrospectively extracted from the electronic medical record systems of the participating centers.

2.2.2 Cardiac Magnetic Resonance (CMR) Protocol and Analysis

All patients underwent CMR imaging, typically 3–7 days post-PCI on a Philips Ingenia 3.0T scanner (Philips Healthcare, Netherlands) system. The protocol included ECG-gated steady-state free precession (SSFP) cine sequences for LV volumes, mass, LVEF, and regional wall motion, and late gadolinium enhancement (LGE) imaging 10–15 minutes post-contrast (Magnevist, Bayer, gadopentetate dimeglumine, 0.1–0.2 mmol/kg) to assess the myocardial scar [17].

CMR images were analyzed offline by two experienced radiologists (blinded to outcomes) using the Philips IntelliSpace Portal (ISP) workstation V6 (Philips Healthcare, Best, Netherlands). The key parameters quantified included Left Ventricular End-Diastolic Volume (LVEDV), Left Ventricular End-Systolic Volume (LVESV), LVEF, myocardial mass, mean Left Ventricular Wall Motion (LVWM) score (AHA 17-segment model), infarct size (IS), and microvascular obstruction (MVO). A representative case demonstrating comprehensive CMR assessment in a 57-year-old male patient with acute anterior STEMI post-PCI is illustrated in Fig. 2, showcasing the multiparametric evaluation approach employed in this study.

Fig. 2.

Cardiac magnetic resonance assessment of myocardial viability and function following primary PCI for acute anterior STEMI. Patient: 57-year-old male with acute anterior STEMI treated with primary PCI. (A) Short-axis LGE image with AHA 17-segment labeling (1–6). Hyperintense regions (arrows) indicate myocardial scar. (B) LGE transmurality bulls-eye plot: blue (0%), green (1–25%), yellow (26–50%), orange (51–75%), red (76–100%). Predominant blue indicates preserved viability. (C) LV functional parameters: EF 41%, EDV 119.4 mL, SV 48.8 mL, CO 3.2 L/min. (D) Wall motion analysis showing apical-to-basal gradient consistent with anterior wall abnormality. (E) First-pass perfusion with ROI 1–4 for microvascular assessment. PCI, percutaneous coronary intervention; STEMI, ST-segment elevation myocardial infarction; LGE, late gadolinium enhancement; AHA, American Heart Association; LV, left ventricular; EF, ejection fraction; EDV, end-diastolic volume; SV, stroke volume; CO, cardiac output; ROI, region of interest.

The primary outcome was MACE, which is a composite of all-cause death, recurrent myocardial infarction, heart failure hospitalization, clinically driven in-stent restenosis, and repeat unplanned revascularization [18]. All patients underwent systematic follow-up after discharge through outpatient visits, telephone interviews, and medical record reviews at 1, 3, 6, 12, 24, and 36 months post-PCI.

Continuous variables were described as the mean $\pm$ SD or median interquartile range (IQR), and categorical variables as n (%). Group comparisons utilized t-tests, Mann-Whitney U tests, or chi-square/Fisher’s exact tests. Univariate Cox regression was used to identify potential MACE predictors (p 0.10). A multivariate Cox model was developed using stepwise backward elimination (Akaike Information Criterion (AIC)-based), with p 0.05 defining independent predictors. Proportional hazards assumptions were checked (Schoenfeld residuals).

A nomogram was constructed from the final model to estimate the 1-, 2-, and 3-year risk of MACE. Model performance was assessed in both derivation and internal validation cohorts via Harrell’s C-index (discrimination), time-dependent AUCs, calibration plots (Hosmer-Lemeshow test for goodness-of-fit), and decision curve analysis (DCA) for clinical utility [19]. The analyses were performed with R software (version 4.2.2, R Foundation for Statistical Computing, Vienna, Austria) and SPSS software (version 26.0, IBM Corp., Armonk, NY, USA). A p-value of 0.05 (two-sided) was considered significant.

A total of 153 patients with new-onset STEMI who underwent PCI were included in this study. Patients were divided into a derivation cohort (n = 107) and an internal validation cohort (n = 46). The baseline clinical, laboratory, and CMR characteristics of the patients in both cohorts are detailed in Tables 1,2. Overall, the two cohorts were well-balanced for most baseline characteristics. However, statistically significant differences were found between the derivation and internal validation cohorts for diastolic blood pressure (85.83 $\pm$ 14.04 mmHg vs. 80.91 $\pm$ 11.89 mmHg, respectively, p = 0.04), mean corpuscular volume (91.26 $\pm$ 4.47 fL vs. 89.72 $\pm$ 3.82 fL, p = 0.043), lipoprotein(a) (69.47 $\pm$ 83.62 mg/dL vs. 38.51 $\pm$ 46.90 mg/dL, p = 0.02) and Global slope of first-pass perfusion (25.56 $\pm$ 7.39 vs. 28.7 $\pm$ 9.57, p = 0.049).

The median follow-up duration for the entire cohort was 35 months (IQR, 24–36 months). During the follow-up period, MACE occurred in 48/107 patients (44.86%) in the derivation cohort, and 23/46 patients (50.00%) in the internal validation cohort. The cumulative MACE incidence rates at 1-, 2-, and 3- years were 20.6%, 34.6%, and 44.9% in the derivation cohort, and 21.7%, 34.8%, and 50.0% in the internal validation cohort. Kaplan-Meier analysis showed no significant difference in overall MACE-free survival between the derivation and internal validation cohorts prior to model development (p = 0.73, as shown in Supplementary Fig. 1, Kaplan-Meier Curve for MACE in the Derivation and Validation Datasets).

Univariate Cox proportional hazards regression analysis was performed to identify potential predictors of MACE from a comprehensive set of baseline clinical, laboratory, and CMR parameters. This initial screening identified 34 variables as being associated with MACE (all p 0.1). The detailed results of this univariate analysis, including hazard ratios and p-values for all tested variables, are presented in Table 3.

Candidate variables from the univariate analysis were then included in a multivariate Cox proportional hazards regression model using a stepwise backward elimination strategy based on the Akaike Information Criterion (AIC). This process yielded a final parsimonious model that identified six independent predictors for MACE: Gensini Score (Hazard Ratio [HR]: 1.012, 95% Confidence Interval [CI]: 1.003–1.020, p = 0.006), albumin (HR: 0.849, 95% CI: 0.769–0.938, p = 0.001), LDL-C (HR: 1.377, 95% CI: 1.037–1.828, p = 0.027), LVEF (HR: 0.890, 95% CI: 0.833–0.951, p = 0.001), LVEDV (HR: 1.014, 95% CI: 1.003–1.025, p = 0.015), and Mean of LVWM (HR: 0.464, 95% CI: 0.288–0.747, p = 0.002). Final results from the multivariate model results are detailed in Table 4. The correlation matrix for these six predictors showed no significant multicollinearity (Supplementary Fig. 2, Correlation Matrix of the Six Independent Predictors Included in the Nomogram).

Based on these independent predictors, a prognostic nomogram was constructed to predict the 1-, 2-, and 3-year probability of MACE (Fig. 3).

Fig. 3.

Nomogram of Cox regression model for predicting the risk of MACE after PCI. Prognostic nomogram for predicting 1-, 2-, and 3-year probability of major adverse cardiovascular events (MACE) after percutaneous coronary intervention (PCI) in patients with new-onset ST-segment elevation myocardial infarction (STEMI). The nomogram was developed based on the final Cox regression model and incorporated six independent predictors: Gensini Score, albumin, LDL-C, Ejection Fraction (EF), End-Diastolic Volume (EDV), and Mean of Left Ventricular Wall Motion (Mean_of_LVWM).

The nomogram demonstrated good discrimination in the derivation cohort, with a Harrell’s C-index for predicting MACE of 0.803 (95% CI: 0.739–0.867). The time-dependent AUC values at 1-, 2-, and 3-years were 0.771, 0.835, and 0.795, respectively (Fig. 4A). In the internal validation cohort, the C-index was 0.693 (95% CI: 0.570–0.816), with time-dependent AUC values at 1-, 2-, and 3-years of 0.629, 0.619, and 0.667, respectively (Fig. 4B).

Fig. 4.

Time-dependent Receiver Operating Characteristic (ROC) curves showing the predictive performance of the nomogram for MACE at 12, 24, and 36 months. (A) ROC curves for the derivation cohort. (B) ROC curves for the internal validation cohort. The diagonal grey line represents an Area Under the Curve (AUC) of 0.5.

Calibration of the nomogram was carried out using calibration plots, which compared the nomogram-predicted probabilities with the Kaplan-Meier estimated probabilities of MACE. These demonstrated excellent agreement at 1-, 2-, and 3-years in both the derivation cohort (Fig. 5A) and the internal validation cohort (Fig. 5B). The concordance was further confirmed by quantitative analysis of the calibration slopes, which were found to be optimal in both cohorts (p $>$ 0.05 for all time points in both cohorts).

Fig. 5.

Calibration curves for the prognostic nomogram. These assessed the agreement between nomogram-predicted probabilities of major adverse cardiovascular events (MACE) and the observed rates at 12-, 24-, and 36- months. (A) Derivation cohort. (B) Internal validation cohort. The diagonal dashed line indicates perfect calibration.

DCA was performed to evaluate the clinical usefulness of the nomogram. In both the derivation (Fig. 6A1–A3) and internal validation cohorts (Fig. 6B1–B3), DCA revealed that the nomogram provided a superior net benefit for predicting 1-, 2-, and 3-year MACE risk across a wide range of threshold probabilities compared to treat-all or treat-none strategies.

Fig. 6.

Decision curve analysis (DCA) for the prognostic nomogram. Evaluating the net benefit for predicting major adverse cardiovascular events (MACE) at 12, 24, and 36 months across a range of threshold probabilities. Top row: DCA for the derivation cohort (traindata) at 12 months (A1), 24 months (A2), and 36 months (A3). Bottom row: DCA for the internal validation cohort (testdata) at 12 months (B1), 24 months (B2), and 36 months (B3). The red line represents the “Treat All” strategy, the green line represents the “Treat None” strategy, and the blue line represents the nomogram-based “Coxph” strategy.

Furthermore, the nomogram effectively stratified patients into distinct risk groups. Kaplan-Meier curves for high-risk and low-risk groups, categorized according to the median of the nomogram’s total points showed significantly different MACE-free survival rates in both the derivation cohort (p 0.001, Fig. 7A) and the internal validation cohort (p = 0.046, Fig. 7B).

Fig. 7.

Kaplan-Meier curves for MACE-free survival based on nomogram-defined stratification into high-risk and low-risk groups according to the median total nomogram score. Shaded areas represent 95% confidence intervals. Tables below each plot show the number of patients at risk at different time points. (A) Derivation cohort. (B) Internal validation cohort.

This study successfully developed and internally validated a prognostic nomogram that integrates six clinical and CMR-derived predictors (Gensini score, serum albumin, LDL-C, LVEF, LVEDV) and mean LVWM to estimate the 1- to 3-year risk of MACE in patients with newly diagnosed STEMI after PCI. The model demonstrated good discrimination in the derivation cohort (C-index: 0.803), and moderate predictive accuracy in the internal validation cohort (C-index: 0.693), with stable calibration. DCA supported the clinical utility of the nomogram and its capability for effective risk stratification [20].

A substantial body of evidence has consolidated the role of CMR for post-STEMI risk prediction, encompassing markers of irreversible injury (IS), microcirculatory damage (MVO), and quantitative function/remodeling metrics. Contemporary syntheses have highlighted that CMR captures complementary structural and functional phenotypes, thereby extending prognostication beyond bedside clinical scores (e.g., GRACE/TIMI) and conventional indices such as LVEF [7, 8, 9, 10, 11, 12]. Notably, recent multicenter analyses have refined earlier views by demonstrating that intramyocardial hemorrhage (IMH) is the decisive adverse phenotype of microvascular injury: patients with IMH⁺ carry a substantially higher risk of MACE, whereas those with MVO⁺/IMH^– exhibit outcomes comparable to patients without microvascular injury [21]. In parallel, longitudinal imaging has shown that persistent MVO can be detected beyond the subacute phase and portends unfavorable remodeling and events, reinforcing the importance of both injury severity and its temporal evolution [22, 23].

Beyond tissue injury, feature-tracking CMR (FT-CMR) strain provides incremental prognostic information. In particular, global longitudinal strain (GLS) confers risk discrimination that complements or surpasses LVEF in post-STEMI cohorts and related ischemic populations, supporting the use of regional/global deformation as a summary of spatially heterogeneous damage and recovery [24, 25]. Tissue characterization via mapping adds further nuance. T1/T2-based signatures of microvascular injury and edema, extending even to the non-infarcted myocardium, have been independently linked to subsequent cardiovascular events, suggesting that diffuse or peri-infarct changes convey vulnerability beyond scar size per se [26, 27]. Methodologically, the timing of acquisition materially affects the magnitude and stability of CMR biomarkers. Expert recommendations and contemporary reviews agree that a subacute window of around 3–7 days post-MI is optimal for quantifying key endpoints (e.g., IS, edema-defined area-at-risk, and microvascular injury), facilitating comparisons across studies [11, 28].

Against this backdrop, our approach of combining routine clinical factors (Gensini score, albumin, LDL-C) with quantitative measures of pump function and remodeling (LVEF, LVEDV), plus a regional functional summary (mean LVWM), aligns with the general direction in this field. A transparent Cox framework, together with a nomogram presentation, favor interpretability and clinical translation while preserving reproducibility. Our performance assessment (discrimination, time-dependent AUCs, calibration, and decision-curve analysis) mirrors contemporary guidance on model evaluation and reporting. The stable calibration across derivation and validation cohorts is particularly relevant for individualized risk estimation, an attribute not uniformly documented in prior CMR-integrated tools [5, 29, 30, 31, 32].

The six independent predictors in our final model map to complementary pathophysiologic axes, i.e., coronary disease burden (Gensini score), systemic milieu (albumin, LDL-C), and myocardial function/remodeling (LVEF, LVEDV, mean LVWM). From the CMR domain, reduced LVEF and increased LVEDV reaffirm well-established associations with adverse remodeling and MACE, while the strong protective association of higher mean LVWM underscores the incremental value of regional functional integrity beyond the performance of global ejection [25, 33, 34, 35, 36]. Among the clinical variables, lower albumin likely reflects systemic inflammation and poor physiologic reserve, consistent with literature linking hypoalbuminemia to adverse outcomes after PCI-treated STEMI, while higher LDLC denotes residual atherothrombotic risk despite revascularization [37, 38].

The absence of IS/MVO in the final multivariable model should be interpreted cautiously, rather than indicating negation of their biological importance. First, functional/remodeling measures often act as integrators of acute necrosis, microvascular damage (including IMH), and early remodeling, thereby attenuating the residual signal of tissue-centric variables previously considered jointly [21, 33, 34]. Second, the imaging window is important. Subacute scans ($\approx$ 3–7 days) may alter the relative weight of injury versus function metrics compared with very-early or late imaging, and persistent microvascular injury can differentially influence long-term risk [22, 23, 28]. Third, pragmatic modeling choices (AIC-guided parsimony in a modest event sample) and inter-study heterogeneity in analysis platforms/thresholds can determine which variables ultimately survive adjustment.

Overall, these considerations are consistent with our findings: a concise, interpretable model anchored in function and remodeling delivered good derivation-cohort discrimination, while maintaining calibration and clinical net benefit on internal validation. The converging literature emphasizes the prognostic salience of IMH, the trajectory of MVO, the incremental value of FT-CMR strain, and standardized subacute imaging, all of which support the biological plausibility of our selected predictors. This warrants prospective external validation and head-to-head augmentation with strain/mapping in larger, multi-center datasets [11, 21, 22, 23, 25, 26, 27, 28].

Key strengths of our study include: (1) a long-term (1–3 years) focus on MACE risk stratification after PCI in newly diagnosed STEMI patients addresses a critical clinical gap; (2) incorporation of high-resolution CMR-derived phenotypic parameters, particularly LVEDV and mean LVWM; (3) a graphical nomogram interface facilitating individualized risk estimation; (4) rigorous methodology, including systematic variable selection, multidimensional performance evaluation (discrimination, calibration, time-dependent AUC), and decision curve analysis; and (5) inclusion of data from two centers, enhancing generalizability.

The nomogram presented here may support tailored secondary prevention strategies post-PCI by identifying high-risk patients eligible for intensified pharmacotherapy or surveillance, while guiding appropriate follow-up frequency in lower-risk individuals. Its visual format may also enhance clinician–patient communication and promote treatment adherence. Future integration into clinical decision support systems could facilitate its application in routine practice.

This study has several limitations that warrant consideration. First, its retrospective design inevitably introduces the potential for selection and information bias, which could affect the generalizability of our findings. While rigorous efforts were made to ensure data completeness and consistency, further prospective studies are needed to validate these results and establish causality more definitively. Furthermore, the relatively small size of our internal validation cohort (n = 46) may limit the statistical precision of our performance estimates and could render them more susceptible to sampling variability. Despite this, the consistent calibration observed across both cohorts provides a reassuring indication of the model’s reliability in estimating actual risks [30].

Second, our study performed only internal validation. While this is a critical initial step in the development of predictive models, establishing true external generalizability requires rigorous validation across diverse, independent cohorts from different institutions or geographical regions. Such external validation is essential to confirm the robustness and transferability of our nomogram to broader clinical practice settings with varying patient demographics and treatment protocols [31].

Third, while our model successfully integrates a comprehensive set of clinical and CMR parameters that proved highly predictive, we did not assess the incremental value of all emerging biomarkers, or even more advanced CMR techniques such as detailed myocardial strain imaging or quantitative T1/T2 mapping. These advanced modalities may offer additional nuanced insights into myocardial tissue characterization and function. However, it is important to note that our current model effectively leverages readily available and clinically routine CMR parameters alongside clinical data. This provides a practical and robust tool for risk stratification that can be more easily implemented in many clinical settings, while also laying the groundwork for future investigations that incorporate these novel markers [5].

Fourth, it should be noted that CMR is not yet part of routine clinical practice for STEMI patients and remains a relatively expensive imaging modality. The practical implementation of our CMR-based nomogram may be limited by equipment availability, costs, and the need for specialized expertise. Future studies should evaluate the cost-effectiveness and feasibility of routine CMR-based risk stratification in diverse healthcare settings.

Future research should also aim to: (1) externally validate the model in large, prospective, multicenter cohorts; (2) explore the integration of novel biomarkers, genetic data, or advanced CMR parameters to refine predictive accuracy; (3) assess the clinical utility of dynamic risk prediction and develop digital tools for clinical implementation; and (4) conduct head-to-head comparisons with existing risk scores across diverse populations to determine incremental benefit [32].