Abstract
We report robust baseline rates of prioritised AESI across 13 databases. Age, sex, and variation between databases should be considered if background AESI rates are compared to event rates observed with COVID-19 vaccines.
Introduction
As the global community deployed COVID-19 vaccines at an unprecedented scale, the ability to distinguish between vaccine-induced side effects and coincidental medical events became a matter of urgent public health priority. To accurately identify a "safety signal," regulators and researchers must first understand the "background rate"—the frequency at which specific medical conditions occur naturally in a population without vaccination.
This study, conducted by the Observational Health Data Sciences and Informatics (OHDSI) community, provides a massive-scale baseline for vaccine safety monitoring. By analyzing data from over 126 million people across eight countries, the researchers established background rates for 15 adverse events of special interest (AESI), demonstrating that what is considered "normal" varies drastically depending on a person's age, sex, and geographic location.
The Problem: Defining the Baseline
When a rare medical event, such as a blood clot or a neurological condition, occurs shortly after vaccination, it is often reported to passive surveillance systems like VAERS (US) or EudraVigilance (EU). However, these reports are difficult to interpret without a standardized denominator. Historically, background rates have been difficult to compare because different studies use inconsistent definitions, varying timeframes, and diverse populations. Without a harmonized, international baseline, it is nearly impossible to determine if the observed number of events exceeds the expected number.
Methodology: The Power of Standardization
The researchers conducted a multinational, multi-database network cohort study utilizing the OMOP Common Data Model (CDM). This approach allowed for a high degree of standardization across 13 different databases without requiring the sharing of sensitive patient-level data.
Key methodological pillars included:
- Standardized Phenotypes: The team focused on 15 AESIs (e.g., acute myocardial infarction, stroke, deep vein thrombosis, Bell’s palsy) using standardized algorithms based on diagnostic codes.
- Large-Scale Analysis: They calculated incidence rates per 100,000 person-years, stratified by age (eight groups) and sex.
- Meta-Analysis: Random-effects meta-analysis was used to pool results, and event frequencies were classified using the Council of International Organizations of Medical Sciences (CIOMS) categories, ranging from "very common" to "very rare."
Data Infrastructure
The study utilized a diverse data landscape encompassing 126,661,070 unique individuals and over 227 million person-years of follow-up between 2017 and 2019. The network included:
- 13 Databases: Comprising 8 electronic health records (EHR) and 5 administrative claims datasets.
- Global Reach: Data originated from Australia, France, Germany, Japan, Netherlands, Spain, the UK, and the US.
- Diverse Settings: Sources ranged from primary care records in the UK (CPRD) to massive US claims datasets (Optum, MarketScan) and hospital-linked data in Spain (SIDIAP).
Key Insights
1. Universal Heterogeneity
Even with standardized definitions, incidence rates varied significantly between databases and countries. For instance, the rate of deep vein thrombosis (DVT) in women aged 35–54 was found to be five times higher in one US database than in the Spanish database. This suggests that local healthcare practices, coding behaviors, and population genetics play a significant role in reported incidence.
2. The Age and Sex Gradient
Most AESIs showed a profound correlation with age and sex. Acute Myocardial Infarction (AMI) was classified as "very rare" (<1/10,000) in women under 35, but became "common" (>1/100) in both men and women aged 85 and older. Conversely, conditions like appendicitis and anaphylaxis were more frequently observed in younger populations.
3. The Danger of "Single Estimates"
Because rates can shift through multiple CIOMS frequency categories (e.g., from "rare" to "uncommon") as a person ages, the authors argue that using a single "global" background rate is misleading. Safety monitoring must be precisely tailored to the specific demographics of the vaccinated population to avoid false positives or missed signals.
Limitations and Future Directions
While the study provides a robust foundation, it is subject to the inherent limitations of observational data. The reliance on diagnostic codes (ICD-10, etc.) without manual validation introduces the risk of measurement error. Furthermore, some databases were limited to primary care settings, which may underreport conditions typically requiring hospitalization, such as strokes.
Additionally, the "healthy user bias" must be considered; the general population baseline may differ from the health profiles of early vaccine recipients (e.g., frontline healthcare workers). The authors suggest that future safety evaluations might benefit from "within-database" comparisons, such as self-controlled case series (SCCS), where a patient acts as their own control, to further mitigate the effects of population heterogeneity.
Conclusion
This OHDSI study underscores the necessity of a nuanced, stratified approach to vaccine safety. By providing a comprehensive map of background AESI rates, it equips regulators with the context needed to interpret post-vaccination events accurately, ensuring that safety signals are identified with precision while maintaining public confidence in immunization programs.
Summary updated