The study, led by Nikola Panajotovikj and Jordi Mestres of Chemotargets SL and the University of Girona, examined VigiBase, the World Health Organization’s global database maintained by the Uppsala Monitoring Centre; the FDA Adverse Event Reporting System, known as FAERS; the Vaccine Adverse Event Reporting System, or VAERS, co-managed by the FDA and the Centers for Disease Control and Prevention; and Japan’s Adverse Drug Event Report database, called JADER, run by the Pharmaceuticals and Medical Devices Agency. As of December 31, 2025, these four systems collectively held roughly 72.3 million reports, with VigiBase alone containing approximately 42.4 million, FAERS 26.3 million, VAERS 2.7 million, and JADER 0.9 million. The findings were published in Frontiers in Drug Safety and Regulation.

One of the most technically demanding parts of the work involved deduplication, the process of identifying reports that appear in more than one database. Because national regulators forward cases to VigiBase while also filing directly with systems like FAERS, the same adverse event can exist in multiple repositories simultaneously. The researchers used a content-matching algorithm that compared drug codes, adverse event terms, time to onset, indication, basis of report, event date, sex, region, and age, harmonizing drug names through the WHO Drug Dictionary and the Anatomical Therapeutic Chemical classification system, and matching adverse events at the MedDRA Preferred Term level. Duplicates were classified into three tiers: exact matches on every field, near-matches differing by a single missing element, and near-matches differing by a single character in one field.

The results of this deduplication effort were striking. After removing duplicates, VigiBase retained 39.5 million reports, of which 27.0 million, or 68.4 percent, were exclusive to that database. FAERS retained 20.3 million reports with 9.1 million, or 44.8 percent, being unique. VAERS was left with 2.3 million reports, 41.5 percent of which were exclusive, while JADER kept nearly all of its 0.9 million reports, with 92.4 percent being unique. Critically, the analysis revealed that VigiBase incorporates more than half of the entire contents of both FAERS, at 54.7 percent, and VAERS, at 56.5 percent. This substantial overlap means that researchers who naively combine data from multiple databases without accounting for duplicates risk double-counting the same adverse events, potentially inflating statistical signals and distorting safety conclusions.

The COVID-19 pandemic left an indelible mark on these databases, though its impact varied dramatically depending on each system’s regulatory mandate. In VAERS, which is specifically designed to capture vaccine-related reports, a staggering 57.8 percent of all reports relate to a COVID-19 vaccine. VigiBase saw 13.9 percent of its reports linked to COVID-19 vaccines, while JADER recorded 3.7 percent. FAERS, by contrast, remained almost untouched at just 0.3 percent, because the US regulatory framework directs vaccine reports to VAERS rather than to the general drug safety system. The pandemic’s influence on database composition is so pronounced that the researchers excluded all COVID-19 vaccine reports from their temporal trend analyses to avoid obscuring underlying patterns in drug safety reporting.

Temporal analysis revealed that these databases are growing at an accelerating pace. Approximately 40 percent of all reports in VigiBase were received in just the past five years, and 70 percent within the past ten years. FAERS collected 40 percent of its reports in the last eight years and 70 percent in the last fourteen years, with similar patterns observed for JADER. VAERS experienced the most dramatic acceleration, accumulating 70 percent of all its reports in just seven years, driven largely by the pandemic. When COVID-19 vaccine reports were excluded, the timeline for reaching these milestones extended considerably, particularly for VAERS, where the 70 percent threshold shifted from seven years to seventeen years.

Perhaps the most culturally revealing findings concern who files these reports. In FAERS, consumers dominate, submitting 53.4 percent of all reports, followed by physicians at 25.0 percent. In stark contrast, JADER is overwhelmingly physician-driven, with doctors filing 70.5 percent of reports, reflecting a Japanese pharmacovigilance framework that relies heavily on healthcare professionals. VigiBase occupies a middle ground, with a nearly balanced distribution between consumers at 41.2 percent and physicians at 39.7 percent. Interestingly, lawyer-reported cases, while rare across all databases, are most prominent in FAERS at 2.5 percent, potentially reflecting litigation-driven reporting patterns in the United States.

Demographic patterns also diverged significantly across systems. Female patients are over-represented in VigiBase, FAERS, and VAERS, accounting for 60.4, 60.3, and 64.8 percent of reports respectively, a pattern that may reflect women’s higher healthcare engagement and greater susceptibility to certain adverse events. JADER stands out as a notable exception, with an almost equal sex distribution at 48.9 percent female and 51.1 percent male. Age distributions tell equally distinct stories: JADER reports skew heavily toward elderly patients, with 42.3 percent in the retirement category of 75 years and older, consistent with Japan’s ageing population and associated polypharmacy burden. VAERS shows a higher proportion of paediatric reports, reflecting routine childhood immunization schedules in the United States.

When the researchers compared reporting patterns before and after January 1, 2015, excluding COVID-19 vaccine reports, they found consistent shifts across all four systems. Consumer reporting increased markedly in VigiBase, rising from 29.7 to 41.6 percent, while physician reporting declined from 53.7 to 36.4 percent. Similar trends appeared in FAERS and JADER, suggesting a global movement toward greater patient involvement in pharmacovigilance. The proportion of reports involving antineoplastic and immunomodulating agents rose substantially across all databases, reflecting the surge in oncology and immunotherapy approvals in recent years. Meanwhile, reports of medication errors and product use issues increased dramatically, from 1.4 to 5.1 percent in FAERS and from 1.3 to 5.0 percent in VigiBase, a trend the researchers attribute to the increasing formalization of medication error concepts within pharmacovigilance coding practices rather than a true rise in error frequency.

The study also quantified each database’s capacity to generate statistical safety signals using the proportional reporting ratio, a disproportionality measure that flags drug-event pairs reported more frequently than expected. FAERS produced the largest absolute number of signals at 1,616,441, followed by VigiBase at 1,203,070, VAERS at 107,390, and JADER at 96,174. When normalized by the size of each database’s drug-event vocabulary, FAERS again led with a signal yield of 21.1 percent, followed by VigiBase at 16.1 percent, VAERS at 11.0 percent, and JADER at 9.8 percent. Notably, of the 2,021,522 unique signals identified across all four databases, only 0.1 percent were detected in all four simultaneously, while 54.6 percent were exclusive to a single database. This striking lack of overlap underscores the complementary nature of these systems and reinforces the argument that multi-database signal detection strategies, when properly deduplicated and interpreted in light of each system’s unique characteristics, offer the most comprehensive picture of post-marketing drug safety worldwide.

Subject of Research: Comparative analysis of global spontaneous adverse event reporting systems for drug and vaccine safety surveillance

Article Title: Global patterns and trends in spontaneous reporting systems: VigiBase, FAERS, VAERS, and JADER

Article References: Panajotovikj, N., & Mestres, J. (2026). Global patterns and trends in spontaneous reporting systems: VigiBase, FAERS, VAERS, and JADER. Frontiers in Drug Safety and Regulation, 6, Article 1919822. https://doi.org/10.3389/fdsfr.2026.1919822

Image Credits: AI Generated

DOI: 10.3389/fdsfr.2026.1919822

Keywords: pharmacovigilance, VigiBase, FAERS, VAERS, JADER, adverse drug events, drug safety signals, COVID-19 vaccines, deduplication, disproportionality analysis, spontaneous reporting, post-marketing surveillance

The story begins in December 2020, when the US Centers for Disease Control and Prevention prepared to roll out the first mRNA covid-19 vaccines. Before the first doses were administered, the CDC assured healthcare professionals and the public that the Vaccine Adverse Event Reporting System, or VAERS, could quickly identify potentially harmful reactions following vaccination. VAERS, a long-standing national early-warning system operated jointly by the CDC and the Food and Drug Administration, depends on voluntary reports from clinicians, vaccine manufacturers, and members of the public, and it was intended to serve as the first line of defense in spotting unexpected patterns of harm.

To mine that flood of incoming reports for danger signals, the CDC planned to deploy two complementary data-mining techniques. The first was the proportional reporting ratio, or PRR, a statistical measure that compares how often a particular adverse event appears in reports for a given vaccine against how often it appears for all other vaccines in the database. The second was an empirical Bayesian method supplied by the FDA, which uses prior expectations to determine whether observed reporting frequencies are unusually elevated. According to the original plan, the two agencies would run both methods, share their results, and discuss any discrepancies. That plan, the investigation found, was quietly abandoned.

In a letter cited by The BMJ, then CDC Director Rochelle Walensky acknowledged that the agency did not perform PRR analyses until 2022, more than a year after the vaccines had been given to hundreds of millions of Americans. In the interim, both the CDC and the FDA chose to rely entirely on the FDA’s Bayesian method. That decision would prove consequential, because the FDA’s algorithm carried a structural flaw that, according to internal records, agency officials understood before the pandemic even began: it was effectively blind to adverse events that were elevated by both major mRNA vaccines at the same time.

The technical reason for this blindness lies in the mathematics of disproportionality analysis. By early 2021, more than 90 percent of all VAERS reports concerned the new mRNA vaccines made by Pfizer and Moderna. If both vaccines raised the risk of an adverse event, such as myocarditis, in roughly equal amounts, then the observed frequency of that event in the database would closely match the expected frequency calculated against the background of all other vaccines. The two values would cancel each other out, and no automated alert would ever fire. In other words, the very dominance of the new vaccines in the reporting stream neutralized the statistical machinery designed to detect harm.

Internal government documents show that FDA officials were aware of this limitation before and during the pandemic, and that CDC officials were informed of the problem while the vaccination campaign was underway. In early 2021, FDA medical officer Dr Ana Szarfman, working with statistician William DuMouchel, the original developer of the Bayesian algorithm, warned top officials about the flaw and proposed an updated algorithm that would have flagged the hidden signals. Instead of being heeded, Szarfman was told to cease and desist. Meanwhile, agency officials continued to cite the absence of system alerts as they reassured clinicians and the public about the vaccines’ safety profile.

The consequences of that silence became measurable when the CDC finally ran its PRR analyses in 2022. Walensky publicly stated that the results revealed no additional unexpected safety signals. But when The BMJ examined the underlying analyses, they showed hundreds of adverse events that met the agency’s own alert criteria, including myocarditis, the inflammation of the heart muscle; pericarditis, inflammation of the fluid-filled sac surrounding the heart; Bell’s palsy; and tinnitus. These were precisely the kinds of associations that the FDA’s Bayesian method had failed to trigger, and their appearance in the PRR data suggests that a properly functioning dual-surveillance approach might have surfaced them far earlier.

The investigation also documents that the FDA’s own pharmacovigilance leadership acknowledged the problem internally. In October 2023, the agency’s pharmacovigilance chief wrote in an email to colleagues that the FDA had known of the detection deficiency since the vaccines first rolled out more than two years earlier. The FDA did not respond to The BMJ’s requests for comment on the email or on the broader findings. That admission, coming from within the agency responsible for the flawed method, adds weight to the central claim that the surveillance gap was understood at senior levels and never corrected in public.

The scientists who tried to raise the alarm have now spoken on the record. Szarfman insisted to The BMJ that she had never sought to undermine public confidence in the covid vaccines, expressing her frustration with the observation that very few people understand the statistics, and that this, in her view, was the core of the problem. DuMouchel said he could not explain the officials’ resistance to switching to the updated method, stating simply that he believed they were wrong. He also expressed regret that the FDA had rejected Szarfman’s proposed fixes for VAERS, adding that if the agency had paid attention to her, the system would have performed better.

The investigation reignites a debate that has simmered throughout the pandemic era: what steps were actually taken to safeguard public health from any unintended consequences of the fastest vaccination rollout in history? The documented record now shows an early-warning system whose primary detection tool was structurally incapable of flagging the most relevant signals, a backup method that was postponed for more than a year, internal warnings that were met with a cease-and-desist order, and public reassurances that leaned on the very absence of alerts the flawed system was guaranteed to produce. Whether the myocarditis and other signals eventually identified could have been detected, communicated, and acted upon sooner remains the question that the agencies’ own documents now force into the open.

Subject of Research: Suppressed warnings over a flawed US vaccine adverse event surveillance algorithm during the covid-19 vaccination rollout

Article Title: US officials suppressed warnings over flawed covid vaccine safety algorithm

Article References: US officials suppressed warnings over flawed covid vaccine safety algorithm. (n.d.). Original publication

Image Credits: AI Generated

DOI: Not provided

Keywords: VAERS, CDC, FDA, mRNA vaccines, myocarditis, vaccine safety surveillance, proportional reporting ratio, empirical Bayesian method, The BMJ investigation, adverse event detection, public health transparency, covid-19 vaccination