Abstract: Background　Acute myocardial infarction (AMI) is a leading cause of death worldwide. Patients remain at risk of in-hospital major adverse cardiovascular events (MACE) after percutaneous coronary intervention (PCI), and early identification of high-risk patients is essential for improving short-term prognosis. Although an increasing number of risk prediction models have been developed in recent years, substantial differences exist across studies regarding predictors, model performance, and risk of bias, highlighting the need for a systematic evaluation. Objective　To systematically evaluate the research status and predictive efficacy of risk prediction models for in-hospital MACE after PCI in patients with AMI. Methods　Databases including CNKI, Wanfang Data, VIP, CBM, PubMed, Embase, Cochrane Library and Web of Science were searched from inception to January 2026 to collect studies concerning risk prediction models for in-hospital MACE in AMI patients after PCI. Two investigators independently screened the literature and extracted data. The risk of bias was assessed using the PROBAST tool. Meta-analysis of common predictive factors was performed using Stata 18 software, and the area under the curve (AUC) of model predictive performance was statistically analyzed using MedCalc software. Results　A total of 14 studies with 14 prediction models and 15 225 patients were included. The incidence of in-hospital MACE ranged from 5.03% to 47.02%, with an overall incidence of 13.96% (2 126/15 225). Among the 14 studies, 12 were rated as high risk of overall bias and 2 as low risk of overall bias. The AUC of the models ranged from 0.666 to 0.921. Quantitative pooling of studies with sufficient AUC data showed a pooled AUC of 0.833 (95%CI=0.794-0.871). Meta-analysis showed that increased left ventricular ejection fraction (LVEF) was associated with a reduced risk of in-hospital MACE after PCI in patients with AMI, whereas higher Killip class, increased blood urea nitrogen (BUN), older age, and smoking history were associated with an increased risk of in-hospital MACE (all P<0.05). Sensitivity analysis using the leave-one-out method showed no obvious change in the direction of the pooled effects of major predictors, including LVEF, age, Killip class, and BUN, suggesting that the main results were relatively stable. After pooling the AUC effect sizes, the funnel plot showed an asymmetric distribution of studies, indicating potential publication bias. Conclusion　Risk prediction models for in-hospital MACE in AMI patients post PCI showed satisfactory overall predictive performance and discrimination. However, most included studies carried high risk of bias, and the pooled AUC presented significant heterogeneity as well as publication bias, suggesting that the actual predictive efficiency of these models might be overestimated. Therefore, pooled AUC cannot be simply equated to the stable predictive capacity of models under real clinical conditions. Further multi-center, large-sample studies with adequate internal and external validation are needed in the future. Optimized research design and reporting standards will help construct more robust risk prediction models suitable for clinical practice, so as to realize early identification and prevention of in-hospital MACE after PCI.

Key words: Acute myocardial infarction, Percutaneous coronary intervention, Major adverse cardiovascular events, Risk prediction model, Systematic review

摘要： 背景　急性心肌梗死（AMI）是全球主要的死亡原因之一，患者在经皮冠状动脉介入治疗（PCI）后仍存在院内主要不良心血管事件（MACE）发生风险，早期识别高危患者对改善短期预后具有重要意义。近年来相关风险预测模型不断增多，但不同研究在预测因子、模型性能及偏倚风险方面存在差异，尚需系统评价。目的　系统评价AMI患者PCI后院内发生MACE风险预测模型的研究现状及预测效能。方法　计算机检索中国知网、万方数据知识服务平台、维普网、中国生物医学文献数据库、PubMed、Embase、Cochrane Library、Web of Science等数据库中有关AMI患者PCI术后院内发生MACE风险预测模型的研究。检索时间为建库至2026年1月。由2名研究者独立筛选文献并提取数据，采用PROBAST工具进行偏倚风险评价，使用Stata 18软件对共同预测因子进行Meta分析，采用MedCalc软件对模型预测性能AUC进行统计学分析。结果　共纳入14篇文献，建立14个预测模型，涉及15 225例患者，院内MACE发生率为5.03%~47.02%，总体发生率为13.96%（2 126/15 225）。14篇文献中，12篇文献总体偏倚风险为高，2篇文献总体偏倚风险为低。纳入研究中模型的受试者工作特征曲线下面积（AUC）为0.666~0.921，对报告充分AUC数据的研究进行定量合并后，合并AUC为0.833（95%CI=0.794~0.871）。Meta分析结果显示，左室射血分数（LVEF）升高是AMI患者PCI术后院内发生MACE的保护因素（P<0.05）；Killip分级、血尿素氮（BUN）升高，年龄增加及合并吸烟史是AMI患者PCI术后院内发生MACE的危险因素（P<0.05）。采用逐一剔除法进行敏感性分析后，LVEF、年龄、Killip分级、BUN等主要预测因子的合并效应方向未发生明显改变，提示主要结果总体具有一定稳定性。合并AUC效应量值后绘制漏斗图，结果显示：各研究左右分布不对称，表明可能存在发表偏倚。结论　AMI患者PCI术后院内发生MACE风险预测模型整体表现出较好的预测性能和区分能力，但多数纳入研究存在较高偏倚风险，且AUC合并结果存在明显异质性和发表偏倚，提示模型真实预测效能可能被高估。因此，合并AUC不宜直接等同于模型在真实临床环境中的稳定预测能力，未来仍需开展多中心、大样本、充分进行内外部验证的研究，进一步优化研究设计与报告流程，以构建更稳健、更加适用于临床实践的风险预测模型，从而实现PCI术后院内MACE的早期识别与预防。

关键词: 急性心肌梗死, 经皮冠状动脉介入治疗, 主要不良心血管事件, 风险预测模型, 系统评价