Our study developed and validated a nomogram for predicting MACE in a cohort of 1,946 elderly patients who underwent primary PCI. The nomogram, based on 8 significant predictors, was assessed for its discriminatory ability using ROC analysis, which yielded an AUC of 0.71. Kaplan-Meier analysis further evaluated its prognostic reliability by stratifying patients into four risk groups based on predicted MACE risk. This analysis, illustrated in both the train and test sets, revealed that very high-risk patients (predicted risk > 0.75) had significantly higher hazard ratios for MACE compared to lower-risk groups, affirming the nomogram's clinical reliability.
Effective prognostic information is often obtained through scoring systems. The TIMI risk score is a straightforward tool originally designed to predict 30-day mortality in STEMI patients treated with fibrinolytics, and it has also proven effective for predicting in-hospital mortality in STEMI patients undergoing primary PCI [13, 14]. The Controlled Abciximab and Device Investigation to Lower Late Angioplasty Complications (CADILLAC) risk score was developed to predict both 30-day and one-year mortality in STEMI and NSTEMI patients following primary PCI [15]. Compared with the mentioned and other traditional risk scores, including TIMI, PAMI, CADILLAC, and GRACE, which offer a modest discrimination in both the general STEMI (AUCs ranging from 0.68 to 0.70) [16] and elderly population, with an AUC of 0.70 and 0.69 for in-hospital and post-discharge mortality, respectively [17] our nomogram model demonstrates improved predictive power in the patient population aged ≥ 65 years, a population being rarely investigated in the existing literature. [18, 19]
The incorporation of the most robust clinical predictors for the elderly population following PCI, including age, diabetes, prior MI, LVEF, STEMI presentation, and eGFR, which were identified by Jalali et al.'s meta-analysis [20], alongside real-time procedural characteristics like post-PCI TIMI flow, stent diameter, and post-PCI shock, into our nomogram, has led to the development of a more bedside-friendly and homogeneous risk assessment tool for clinicians and patients, as other machine learning models such as iPROMPT [21] and XGBoost with a large sample size [22], rely more on heterogeneous predictors and “black-box” algorithms while showing more discriminative ability.
Zhu et al. [18] have developed a similar nomogram in the elderly population post-PCI, but lacked the distinction between STEMI and NSTEMI populations, underestimating the possible clinical discernibility between these two. Although more discriminative ability was achieved by Zhu’s model with an AUC of 0.787, the high reliance of this model on biomarkers, overlooking the mechanical and hemodynamic dimensions of risk, which are addressed by procedural variables, and a smaller sample size of nearly 1000 patients compared with our nomogram, demonstrates the high clinical utility of the current model. Unlike our model, which addressed the risk of MACE following PCI in a specific age group, Yao et al. [23] developed another risk-prediction nomogram model, incorporating six variables, for 1-year readmission due to MACE in the general STEMI patients’ population with a considerably smaller sample size, lowering the generalizability of this tool in the clinical environment.
Our model addresses some methodological shortcomings mentioned in a study that systematically explored the prediction models for MACE following PCI. With the inclusion of nearly 2,000 patients aged ≥ 65, we minimized heterogeneity and increased applicability for the elderly population while providing an events-per-variable (EPV) above the ≥ 20 threshold. The incorporation of real-time procedural characteristics into our model alongside the implementation of multivariable analysis addressed the lack of multivariable modeling for the selection of MACE predictors and the underestimation of procedural variables for risk stratification, as mentioned by Deng et al. [19] Moreover, we presented both calibration plots and Hosmer-Lemeshow statistics, directly addressing what 74% of models failed to report in Deng’s review. [19]
We have enabled the risk stratification of elderly patients into four distinct risk groups based on the model’s predicted risk, which can inform individual care by directing high-risk patients to more intensive treatment plans, including earlier referral to specialized cardiology and heart failure clinics as well as extended monitoring and more intensive post-discharge follow-up. DCA is conducted to evaluate the clinical value of a prognostic model by quantifying the net benefit across a range of risk thresholds for developing a certain outcome, at which a clinician might choose to intervene. In our model, DCA delivers a positive net clinical benefit in a broad range of risk thresholds (23–89%), which translates into the applicability of the current model for both clinicians who only treat extremely high-risk patients and those who approach patients more conservatively in real-world settings. Chen et al.’s iPROMPT [21] shows a positive net benefit in a much narrower threshold range of 30–60% and Zhu’s model [18] demonstrates net benefit only when the threshold exceeds 15% which does not clarify how their net gain changes beyond moderate thresholds. Hamilton et al. [22] did not include a DCA, leaving clinicians and patients uncertain about whether this model leads to better clinical decision support in real-world settings. Fang et al.’s [24] DCA showed net benefit in a wide threshold range of 10–99% in their model, but included only 466 STEMI patients.
Importantly, although we have not yet implemented our nomogram in the institution’s electronic medical record (EMR), the model is designed for integration into EMR and real-time use at the bedside. Its variables are readily available after PCI, and its point-based graphic format supports bedside decision-making and shared discussions with patients and families. This implementation plan reinforces the model’s potential utility for cardiologists and explains how we envision its applicability in daily care alongside its direct benefit to readers and clinicians.
Our nomogram was constructed based on eight key factors identified as independent predictors of MACE in elderly STEMI patients undergoing PCI. These factors include LVEF, serum creatinine, hemoglobin, FBS, presence of VHD, post-PCI TIMI flow grade less than 2, diameter of the stent placed in the culprit lesion, and the presence or absence of shock in the post-PCI setting.
LVEF, as a critical measure of cardiac function, is particularly important in the context of STEMI, as it helps to assess the severity of cardiac damage and guides treatment decisions. Previous studies have shown that low LVEF is associated with higher rates of MACE following STEMI [25–27]. VHD is both a risk and a complication of ACS, which is associated with a worse prognosis [28, 29]. Our results align with the study of Hasdai et al., in which STEMI patients with pre-existing VHD had a worse prognosis compared to those without VHD [30].
Our nomogram incorporates easily obtainable blood test parameters, including creatinine, FBS, and baseline hemoglobin levels. Various studies have shown that higher creatinine and worse kidney function are associated with higher rates of MACE after acute MI [31, 32]. Many patients with elevated creatinine levels have comorbid conditions such as diabetes, hypertension, and chronic kidney disease (CKD), which independently contribute to worse outcomes in STEMI. Moreover, renal impairment can exacerbate the inflammatory response during MI [33]. Additionally, renal dysfunction can lead to fluid overload and hypertension, further straining the heart during an acute event [34].
Furthermore, studies have shown that stress hyperglycemia is significantly associated with an increased risk of mortality in STEMI patients treated with PCI, regardless of diabetic status [35, 36]. Stress hyperglycemia triggers the production of inflammatory factors, which can exacerbate atherosclerosis through various intracellular pathways [37]. Additionally, stress hyperglycemia increases thrombogenic activity, leading to a hypercoagulable state [38]. Anemia is another strong and independent predictor of MACE in patients with ACS, showing a notable dose-response relationship [39]. It significantly reduces oxygen delivery to the myocardium beyond coronary stenoses and increases myocardial oxygen demand by requiring a higher stroke volume and heart rate to ensure sufficient systemic oxygen delivery [40]. These factors may explain the progressively poorer outcomes seen in ACS patients with lower baseline hemoglobin levels. In the study of Liu et al., a higher hemoglobin level in the anemic group was associated with a decreased risk of 1-year mortality; however, a higher hemoglobin level in the erythrocytosis group was linked to an increased risk of 1-year mortality [41].
It is well established that peri-procedural complications during PCI can significantly impact mortality [42]. In our study, we observed that the characteristics of PCI and post-PCI events play a crucial role in patient outcomes. Notably, post-PCI TIMI flow, the occurrence of post-PCI shock, and the diameter of the PCI stent were significantly meaningful factors. Previous studies have shown that achieving TIMI flow grade 3 following PCI is linked to better short-term and long-term outcomes in these patients, including a decrease in MACE and cardiac mortality [43, 44]. Although peri-procedural cardiogenic shock is reported to be low, it was linked to a significantly higher mortality [45]. Likely, in our study, post-PCI shock was the most powerful predictor of MACE following PCI. Stent diameter was another key factor in predicting MACE in this population. The benefits of stent oversizing on procedural and clinical outcomes have been documented. Specifically, small vessels treated with smaller stents experienced more adverse events [46]. Shugman et al. reported that deploying bare-metal stents in STEMI patients with infarct-related arteries measuring 3.5 mm or larger was associated with low rates of target vessel revascularization [47]. This indicates that choosing larger stents could enhance long-term outcomes.
Our study has several strengths, including the use of a large cohort of 1,946 elderly STEMI patients, which enhances the robustness and generalizability of the findings. The rigorous analytical approach, employing both univariate and multivariate methods, allowed for the accurate identification of significant predictors of MACE. The nomogram was validated with an AUC of 0.71, demonstrating its effective discriminatory ability across train and test cohorts. Additionally, the inclusion of key clinical factors such as LVEF and post-PCI TIMI flow underscores its practical relevance. The DCA further confirmed the nomogram’s broad applicability and net benefit for clinical decision-making. However, the study has limitations such as the potential lack of generalizability due to being conducted in a single center, a follow-up period that may not capture long-term outcomes, and variability in some data points between cohorts. Other relevant factors not included in the nomogram might also influence MACE risk stratification.
Limitation
There are several important limitations to consider. Firstly, it was conducted at a single center, which may influence the generalizability of the findings to broader populations, as our results may not fully represent diverse patient demographics or different clinical settings. Additionally, the sample size of fewer than 2,000 participants, while adequate for preliminary findings, suggests that a larger cohort could enhance statistical power and improve the detection of significant differences or associations. The absence of external validation with a larger dataset indicates an opportunity for future research to strengthen the applicability of our results across various populations. Additionally, some clinically relevant parameters were not available in our dataset. These include details of antiplatelet therapy, anatomical vessel involvement (e.g., single- vs. multi-vessel disease), and post-discharge data such as attendance at cardiac rehabilitation or adherence to guideline-directed medical therapy (GDMT). While these variables are known to influence long-term outcomes, they could not be assessed in the current study. However, our model incorporates several intra-procedural predictors, including post-PCI TIMI flow, stent diameter, and post-PCI shock, which offer insight into procedural success and clinical severity. Notably, a recent systematic review of PCI-related MACE prediction models [19] identified these procedural variables as underexplored components of existing models. The inclusion of such factors in our nomogram provides added value and reflects an effort to address this methodological gap. Importantly, we recognize that important clinical parameters, including echocardiography and electrocardiogram, were not included in this investigation, and future studies that incorporate these factors may provide a more comprehensive understanding of the outcomes. Although the model demonstrated satisfactory discrimination (an AUC of 0.71, sensitivity of 72%, specificity of 63%), further improvement may be achievable through external validation in larger multicenter cohorts, incorporation of additional anatomical and post-discharge predictors (e.g., multi-vessel disease, adherence to guideline-directed medical therapy), and the application of rigorously validated machine-learning approaches. In addition, our Cox model was specified to include only main effects. Although this approach prioritizes interpretability and stability, it assumes no effect modification between predictors. Evaluation of clinically plausible interactions in larger or external datasets is an important direction for future research. Lastly, some study limitations, including the lack of external validation in multicenter studies and standardization across different systems, are potential barriers to the implementation of our nomogram model into the EMR. Assessment of clinician and care team acceptance, evaluation of workflow impact, and feasibility testing with medico-legal considerations will enable the integration of this nomogram into the EMR and real-time bedside settings. We will focus on pursuing each step to underline the practical implications of clinical applicability and to ensure the integration of this model into the EMR.