FDA Officials Rejected Superior Statistical Tool for Detecting COVID-19 Vaccine Safety Signals in VAERS, Documents Show
BY JEFF SKINNER
STATEWIDE - Federal health officials at the Food and Drug Administration knew of a more advanced statistical method to detect potential safety problems with COVID-19 vaccines in the nation’s primary adverse event reporting system but directed researchers to stop using it, according to internal emails and a new Senate report released Tuesday.
The documents, obtained by Sen. Ron Johnson, R-Wis., through a subpoena and detailed in a Senate Permanent Subcommittee on Investigations majority staff interim report, indicate that FDA personnel were concerned the improved tool could generate findings that might fuel “anti-vaccination rhetoric.” Critics say the decision left the agency relying on an older method known to have limitations, including failure to account for “masking” effects where reports for one vaccine can obscure signals for another.
The Vaccine Adverse Event Reporting System (VAERS), co-managed by the FDA and Centers for Disease Control and Prevention, serves as an early warning system for potential vaccine safety issues. Anyone can submit reports of adverse events following vaccination, but the FDA has maintained the data cannot prove causation and requires further investigation through other systems, providing an aura of plausible deniability.
The dispute centered on empirical Bayesian (EB) data-mining techniques used to scan VAERS for disproportionate reporting of adverse events. The FDA had long relied on the Multi-item Gamma Poisson Shrinker (MGPS), a method developed in 1999 by statistician William DuMouchel. A newer approach, the Regression-Adjusted Gamma Poisson Shrinker (RGPS), was designed to adjust for masking and other confounding factors.
Ana Szarfman, an FDA medical officer and safety data-mining expert in the Center for Drug Evaluation and Research, advocated for adopting the RGPS method as early as late 2020, according to the emails. Szarfman and DuMouchel, then working as a contractor for the FDA through Oracle, tested the updated tool.
In a July 12, 2021, analysis using RGPS, DuMouchel identified statistically significant safety signals for “death and sudden death” associated with the Pfizer-BioNTech and Moderna mRNA COVID-19 vaccines, the documents show. The older MGPS method did not flag those signals as strongly.
FDA officials pushed back. On May 7, 2021, Szarfman was told to “hold off on creating and sending data mining reports and analyses.” Peter Marks, then director of the FDA’s Center for Biologics Evaluation and Research, later described Szarfman’s work as a “major distraction” in a Sept. 15, 2021, communication. Marks said she had been asked to “cease and desist” because her strategy “could create erroneous conflicts that feed in to anti-vaccination rhetoric.
Dr. Peter Stein, director of the Office of New Drugs in the Center for Drug Evaluation and Research, reinforced the directive, instructing Szarfman to focus on assigned work and not share internal analyses externally.
In 2022, Szarfman and DuMouchel published a peer-reviewed paper in the journal Drug Safety demonstrating the RGPS method’s advantages. It identified roughly 25 statistically significant signals for COVID-19 vaccines that the older method missed, including sudden cardiac death, Bell’s palsy and pulmonary infarction. Szarfman forwarded the paper to then-FDA Commissioner Robert Califf, who replied “Thanks. These are good,” but there is no indication the agency adopted the updated approach.
Richard Forshee, deputy director of the Office of Biostatistics and Pharmacovigilance in the Center for Biologics Evaluation and Research, emailed Marks expressing concerns about the paper and indicating discussions on how to respond.
The Senate report, titled “Unmasked: How Biden Health Officials Purposely Turned a Blind Eye Toward COVID-19 Vaccine Safety Signals,” was released Tuesday alongside a hearing.
Karl Jablonowski, senior research scientist at Children’s Health Defense, testified about shortcomings in VAERS signal detection for COVID-19 vaccine injuries.
Johnson’s office obtained approximately 600 pages of emails through a January 2025 subpoena to the Department of Health and Human Services. The documents also show internal awareness that the older EB method had “blind spots” and was described in some contexts as “mostly useless” for COVID-19 vaccine signal detection, yet continued to be used in analyses.
FDA officials took steps to limit distribution of data-mining reports, including stopping weekly transmissions to the CDC in September 2022 to “reduce email traffic.” Some CDC staff suggested the pause related to Freedom of Information Act requests from Johnson and groups including Children’s Health Defense and the Informed Consent Action Network.
Szarfman retired from the FDA in 2025. Johnson’s staff attempted to contact her for the investigation, but it remains unclear whether she will testify.
The FDA did not immediately respond to requests for comment Tuesday. The agency has previously stated that it continuously monitors VAERS and other systems, using multiple layers of surveillance including the Vaccine Safety Datalink and BEST system for deeper analysis when signals emerge. VAERS itself is a hypothesis-generating tool, not a system for establishing causality.
Public health experts have long noted VAERS limitations, including underreporting of events, variable report quality and the inability to calculate incidence rates without denominator data on doses administered, but has opted not to move towards correction.
During the COVID-19 vaccination campaign, which administered hundreds of millions of doses in the U.S., the system detected known risks such as myocarditis with mRNA vaccines and thrombosis with thrombocytopenia syndrome, prompting discussion on updates to guidance, but no official action to investigate or halt vaccination campaigns that led to countless deaths.
Critics, including Johnson and vaccine safety advocates, argue the episode reveals a reluctance to pursue signals that could undermine public confidence in the COVID-19 shots. Supporters of the agency’s approach counter that adopting unvalidated methods or lowering signal thresholds risks generating false positives that could erode trust or divert resources from confirmed risks.
The peer-reviewed 2022 paper by Szarfman and DuMouchel highlighted RGPS’s ability to reduce confounding. However, internal FDA communications suggested officials believed the paper had methodological issues.
It remains unclear precisely which EB data-mining approach the FDA currently employs for routine VAERS monitoring of COVID-19 or other vaccines. The agency has faced separate scrutiny over transparency, including alleged efforts to manage FOIA responses related to safety analyses. Tuesday’s Senate report adds to ongoing congressional and public debate over COVID-19 vaccine oversight during the Biden administration. Johnson has conducted multiple investigations into pandemic response policies, including vaccine safety monitoring.
As of Tuesday evening, neither the FDA nor CDC had issued a detailed public response to the specific allegations about the RGPS tool and directives to Szarfman.
The episode underscores longstanding tensions in pharmacovigilance: balancing rapid signal detection with the need to avoid premature conclusions from passive surveillance data. Experts agree that any potential signal flagged in VAERS requires rigorous follow-up in active surveillance systems with better control for confounders before informing policy.
