Predicting cardiovascular and neurological complications after COVID-19 vaccination: Insights from a comprehensive review

The global COVID-19 vaccination campaign has raised important questions regarding rare cardiovascular and neurological complications. While systematic surveillance has identified these events, predictive models remain limited. Despite billions of vaccine doses administered worldwide, there remains a critical need for individualized risk assessment tools that can move beyond population-level estimates to guide personalized clinical decision-making, particularly for vulnerable subgroups such as young males and individuals with autoimmune conditions. This study aimed to systematically review and meta-analyze cardiovascular and neurological complications following COVID-19 vaccination, develop artificial intelligence (AI) models to predict individual risk, and identify modifiable risk factors for these complications across different vaccine platforms and demographic groups. We conducted a comprehensive systematic review of PubMed, Embase, Web of Science, and Cochrane databases from December 2019 to December 2024. Studies reporting cardiovascular complications including myocarditis and pericarditis and neurological complications including Guillain-Barré syndrome, Bell’s palsy, and cerebrovascular events were included. Random-effects meta-analyses calculated pooled incidence rates and risk ratios. Novel machine learning algorithms including XGBoost, Random Forest, and neural networks were developed using individual-level data sourced from large-scale pharmacovigilance databases (e.g., VAERS, EudraVigilance) and electronic health records, incorporating demographic data, medical history, and vaccine characteristics. Data heterogeneity across studies was managed through rigorous harmonization protocols, multiple imputation using chained equations for missing data, Winsorization and robust scaling for outlier management, and batch effect correction using ComBat harmonization. Model performance was evaluated using AUC-ROC, sensitivity, specificity, precision, recall, F1-score, and calibration metrics. From 12,847 identified records, 156 studies comprising 847 million vaccine doses were included. Meta-analysis revealed myocarditis incidence of 18.5 cases per million doses (95% CI: 14.2–22.8) with highest rates in males aged 12–29 years after second mRNA vaccine dose (54.8 per million, 95% CI: 42.1–67.5). Pericarditis incidence was 8.2 per million doses (95% CI: 6.1–10.3). Guillain-Barré syndrome occurred at 3.8 cases per million (95% CI: 2.9–4.7), predominantly with adenovirus-vectored vaccines. The XGBoost model achieved highest predictive performance for myocarditis (AUC 0.847, 95% CI: 0.831–0.863) with age, sex, vaccine type, and dosing interval as key predictors. Neural network models showed superior performance for neurological complications (AUC 0.812, 95% CI: 0.795–0.829). Specifically, the XGBoost myocarditis model achieved a sensitivity of 81.3%, specificity of 78.9%, precision (positive predictive value) of 24.7%, recall of 81.3%, and an F1-score of 0.764. For neurological complications, the neural network model yielded a sensitivity of 79.8%, specificity of 77.2%, precision of 70.1%, recall of 79.8%, and F1-score of 0.746. Feature importance analysis identified modifiable risk factors including dosing intervals and vaccination timing. COVID-19 vaccines demonstrate rare but serious cardiovascular and neurological complications with identifiable risk patterns. AI-based prediction models can effectively stratify individual risk and inform personalized vaccination strategies. These findings support the development of clinical decision support tools to optimize vaccine safety while maintaining population-level protection against COVID-19.