Tracking mRNA Vaccine Side Effects Beyond Manual Surveillance: Machine Learning Filters 17% of Redundant Signals
Background
During the COVID-19 pandemic, the large-scale emergency deployment of messenger RNA (mRNA) vaccines revealed structural limitations in existing pharmacovigilance systems. Traditional passive surveillance, which relies on voluntary reports from healthcare providers and vaccine recipients, suffers from structural flaws such as frequent reporting delays and high underreporting rates. The difficulty in precisely calculating the actual incidence of vaccine-related adverse events within the general population has also challenged health authorities. Whenever new vaccine platforms are introduced, surveillance channels are flooded with unstructured descriptive data and duplicate complaints. To early detect abnormal signs and clearly prove safety, a new approach was needed to rapidly integrate data from multiple sources. Machine Learning (ML) has emerged as a significant alternative for processing Real-World Data (RWD), encompassing vast Electronic Health Records (EHR), medical billing data, and adverse event reporting systems.
Key Findings
The researchers comprehensively collected English-language literature indexed in PubMed, Embase, and Web of Science from the inception of the database up to June 2026. Two reviewers independently reviewed the literature, and risk of bias assessment was based on a modified QUADAS-2 tool for assessing the quality of diagnostic accuracy studies. Due to the high level of heterogeneity across machine learning tasks (signal detection, text extraction, risk prediction, prognosis stratification), algorithms, and evaluation metrics, a narrative synthesis approach was inevitable. A total of 43 studies passed the strict criteria to be included in the final analysis.
In the field of adverse event prediction, tree-based models in the decision tree family recorded stable values with an Area Under the Curve (AUC) between 0.85 and 0.87. However, it is difficult to determine direct superiority because the clinical settings and datasets used vary across studies. When Natural Language Processing (NLP) was applied to descriptive adverse event records in the vaccine reporting system, duplicate signals decreased by 17%. By filtering out unnecessary duplicate reports, the workload of surveillance personnel was reduced, while the speed at which meaningful safety signals are identified was significantly enhanced. In the case of myocarditis, cited as a major adverse reaction to mRNA vaccines, the prediction performance of ML models trained on cardiovascular cohorts reached an AUC of up to 0.899. However, studies directly validating these models in actual mRNA vaccine-vaccinated patient groups remain rare, and most are limited to retrospective analyses. For next-generation platforms such as self-amplifying mRNA vaccines or vaccines for cancer treatment, ML applications are identified as being in the proof-of-concept stage due to a lack of post-marketing real-world data.
Implications and Outlook
This systematic review clearly demonstrates that vaccine safety surveillance systems, which previously relied on post-event reporting, are evolving into active and intelligent real-time monitoring systems. Algorithms that learn from vast medical big data in real-time have the potential to detect subtle signs of rare side effects, which are difficult to identify during the early stages of vaccination. However, significant hurdles remain before integration into actual clinical practice. Discrepancies in data formats and quality across medical institutions, and the 'black box' problem where the derivation process of algorithms is difficult to explain clearly, remain obstacles to gaining trust in clinical settings. The current lack of specific licensing guidelines and certified validation standards from regulatory agencies is also cited as a factor delaying commercialization. Only by ensuring data transparency and establishing explainable artificial intelligence models will intelligent pharmacovigilance systems be positioned as a reliable shield for public health.
BACKGROUND: The rapid deployment of mRNA vaccines during the COVID-19 pandemic exposed limitations in traditional pharmacovigilance systems, including delayed reporting, high underreporting rates, and inability to calculate true incidence. Machine learning (ML) offers new pathways to overcome these challenges by integrating multi-source real-world data. METHODS: We systematically reviewed English-language studies from database inception to June 2026. Searches were performed in PubMed, Embase, and Web of Science. Two reviewers independently screened records. Given substantial heterogeneity across ML tasks (signal detection, text extraction, risk prediction, prognosis stratification), algorithms, data sources, and metrics, we performed narrative synthesis. Risk of bias was assessed using adapted QUADAS-2. RESULTS: We identified 43 studies. For adverse-event prediction, tree-based models reported AUCs of 0.85-0.87, though estimates derive from heterogeneous settings. NLP reduced redundant signals by 17% in vaccine reporting systems. For myocarditis, ML models reached AUCs up to 0.899 in cardiovascular cohorts, but direct mRNA vaccine applications remain limited and retrospective. Emerging platforms (self-amplifying and tumor mRNA vaccines) lack post-marketing data, rendering ML applications largely conceptual. CONCLUSION: ML-assisted pharmacovigilance enables a shift from passive to active, intelligent monitoring. Despite challenges in data quality, model interpretability, and regulatory approval, intelligent pharmacovigilance systems will become essential infrastructure for safeguarding public health.
The results of this study suggest specific directions for improvement in vaccine administration sites and health authority safety management systems. In medical institutions, high-risk groups for myocarditis can be precisely identified by pre-analyzing the underlying diseases, past drug reaction history, and cardiovascular risk factors of vaccine recipients using EHR-based ML models. This involves linking a Clinical Decision Support System (CDSS) that automatically provides intensive observation schedules and customized post-vaccination management guidelines to selected high-risk vaccine recipients. Health regulatory agencies can significantly reduce the administrative burden on dedicated personnel by introducing NLP algorithms into adverse event reporting systems to rapidly eliminate 17% of duplicate signals. This makes it a reality to establish a nationwide active surveillance network that captures statistical anomalies without missing them during the initial phase of new vaccine distribution.