Development and Validation of a Risk Prediction Model for Ischemic Stroke in Acute Myocardial Infarction Without Comorbid Atrial Fibrillation

Title: Development and Validation of a Risk Prediction Model for Ischemic Stroke in Patients with Acute Myocardial Infarction without Atrial Fibrillation

Background:

Patients with acute myocardial infarction (AMI) who do not have atrial fibrillation (AF) still exhibit a significantly higher risk of ischemic stroke compared to the general population. A multicenter retrospective study by Ferreira et al. found that among patients with reduced left ventricular ejection fraction (LVEF ≤ 35%), the incidence of stroke was 5.97% in those with AF and 2.9% in those without AF over a median follow-up of 1.9 years. However, this study had important limitations: the population was restricted to patients with LVEF ≤ 35%, limiting generalizability; the endpoint was not clearly differentiated between ischemic stroke and intracerebral hemorrhage (although the latter accounted for less than 10%); and the risk of stroke in non-AF patients was underestimated at 1.5% annually, which is significantly higher than in healthy individuals. Ischemic stroke can severely worsen the prognosis of AMI patients. Evidence from clinical studies shows that the in-hospital mortality rate among patients with concomitant stroke is as high as 46.2%, compared to 6.3% in those without stroke, with a hazard ratio of 7.3. Even if stroke does not lead to death, it significantly impairs quality of life and increases public health expenditures.

The pathophysiology of ischemic stroke in non-AF AMI patients is complex and involves four major pathways:

Cardiogenic embolism: Detachment of left ventricular mural thrombi or spontaneous echo contrast (SEC) in the left atrium; Arterio-arterial embolism: Thrombosis resulting from rupture of atherosclerotic plaques in cerebral arteries; Paradoxical embolism: Deep vein thrombosis from the lower extremities passing through a patent foramen ovale; Hypercoagulable state: Endothelial dysfunction driven by systemic inflammatory responses.

For patients with sinus rhythm (excluding AF), anticoagulant therapy combined with antiplatelet treatment may offer significant benefits. The COMPASS trial demonstrated that in patients with stable atherosclerosis (including 22% with a history of MI), rivaroxaban (2.5 mg twice daily) combined with aspirin significantly reduced the risk of composite cardiovascular events (cardiovascular death, MI, or ischemic stroke) by 24% (HR 0.76; 95% CI 0.66-0.86; P < 0.001), with a 40% reduction in the risk of ischemic stroke. However, this regimen also increased the risk of major bleeding by 70%, highlighting the need for precise risk stratification.

Current evidence indicates that advanced age, heart failure, left atrial enlargement, renal insufficiency, prior stroke history, hypertension, and diabetes are independent predictors of stroke in non-AF AMI patients. However, current clinical practice faces a critical gap: although well-established models for predicting stroke in AF patients (such as CHA₂DS₂-VASc) are widely used, there is no specific risk prediction tool for ischemic stroke in non-AF AMI patients. Given that non-AF AMI accounts for more than 85% of all AMI cases and contributes to more than 75% of ischemic strokes in AMI patients, this lack leads to under-recognition of high-risk patients and challenges in decision-making regarding anticoagulation. Therefore, the development of a dedicated predictive model is urgently needed to enable precise intervention.

Innovation and Rationale:

Based on the "multiple pathways of thrombosis after MI," this study aims to integrate the following novel predictive factors:

Imaging features: Left atrial diameter Biomarkers: D-dimer, lipoprotein(a) (LP(a)) Treatment parameters: Duration of dual antiplatelet therapy

Research Objectives:

To develop and validate a specialized risk prediction model for ischemic stroke in non-AF AMI patients (CAMI-Stroke Score), with the following goals:

Screening for high-risk patients for in-hospital stroke (positive predictive value >85%); Stratifying the risk of ischemic stroke within 3 months and 1 year post-discharge (target C-index >0.80).

Research Content:

A retrospective analysis of hospital records of AMI patients admitted between January 1, 2014, and December 31, 2023, will be conducted. Cases meeting the diagnostic criteria for AMI will be selected, excluding those with AF. Baseline data, clinical information, and follow-up data will be collected. After retrieving and verifying the original medical records, the study will analyze the follow-up data of 3,832 AMI patients (follow-up completed).

Model Development:

Predictive variables: Significant factors will be identified using LASSO regression, followed by multivariate Cox regression modeling.

Validation methods: The model will be developed using 70% of the sample as a training set and validated using 30% as a test set. Discrimination (AUC, C-index) and calibration (Hosmer-Lemeshow test) will be assessed. Decision curve analysis (DCA) and clinical net benefit analysis will be performed. Key performance metrics will be reported with uncertainty intervals (e.g., confidence intervals). If public database data are available, external validation may be considered. The development of the model and feature selection must take into account the feasibility of clinical practice and the availability of data. The impact of the model on patient outcomes (e.g., mortality, morbidity, quality of life) and healthcare costs in real-world clinical settings will be evaluated. The goal is to establish a model development ecosystem that is based on data quality, methodological rigor, clinical needs, ethical fairness, and continuous improvement.

Quality Control and Assurance:

This study is a retrospective study, and all patient identity information will be anonymized to protect privacy. To prevent the leakage of patient identity information and ensure confidentiality, electronic medical records should be closed and screen savers activated when leaving the workstation. Medical records will be handled in accordance with the "Medical Records Management Regulations." When using patient imaging data, all personal identifiers will be removed. During the model validation phase, the predictive performance across subgroups (e.g., age ≥75 years, eGFR <60 ml/min) will be analyzed. Fairness indicators (e.g., equal opportunity difference) will be used to assess bias, and the algorithm will be adjusted to reduce discriminatory risks.

Data Security Measures:

Storage location: Encrypted server within the hospital's internal network (physically isolated area) Access control: Dual authentication (employee ID + dynamic password) + role-based access (researchers can only export aggregated data) Data transmission: Secure SSL-VPN transmission; prohibited to use WeChat or email for data transfer Device management: Research-specific computers will be equipped with data loss prevention (DLP) systems, with USB ports disabled

Ethics of Clinical Research:

This study complies with Article 39 of the "Ethical Review Measures for Biomedical Research Involving Human Subjects" (2023), which provides exemptions under the following conditions: (1) the study uses existing medical records without identifying individuals; (2) the study does not involve secondary use of sensitive personal information; (3) the study protocol has been approved by the ethics committee. This study only uses biochemical marker concentrations and does not perform genetic sequencing or genetic analysis, complying with Article 21 of the "Regulations on Human Genetic Resources." All study protocols and related materials must be submitted to the Ethics Committee for approval before the study begins. Researchers must submit annual reports to the Ethics Committee and notify them in writing upon completion or termination of the study. Any changes during the study must be reported to the Ethics Committee, and modifications cannot be implemented without their approval, unless they are necessary to eliminate immediate and direct risks to participants, in which case the Ethics Committee will be notified.

Conflict of Interest Statement:

The researchers declare no commercial conflicts of interest. This study has no funding support.

Supplementary Provisions:

This model is intended solely as a research tool and is not to be used for clinical decision-making. If the model is applied in clinical practice in the future, it must undergo prospective validation and a process for informed consent between physicians and patients.