An investigation published September 16 in The BMJ reports that U.S. health officials relied on a data-mining method to screen COVID-19 vaccine safety reports even after being warned it could miss certain signals, and that an FDA medical officer who pushed for a fix was asked to "cease and desist." Within days, physicians and statisticians disputed how the journal framed those findings.
The dispute centers on the Vaccine Adverse Event Reporting System, or VAERS, the early warning database run jointly by the CDC and FDA. Both sides agree on a key technical point: a statistical problem called "masking" can hide signals. They disagree on whether officials' choices amounted to suppression or a defensible trade-off during a vaccine rollout of unprecedented size.
The investigation does not present new clinical data showing that the vaccines caused new harms. It concerns how early warning signals were screened in 2021 and 2022.
What the BMJ Report Documents
The BMJ investigation, by Pulitzer Prize-winning journalist David Willman, is based on government emails released under the Freedom of Information Act and in response to Senate investigators, along with interviews. According to the BMJ's press summary, the CDC had planned to use two methods to scan VAERS, but a letter from then-CDC Director Rochelle Walensky said the agency did not run one of them, proportional reporting ratios, until 2022. Both agencies instead relied on an FDA method known as empirical Bayesian data mining.
The report says that method failed to flag potential links between mRNA vaccines and myocarditis, pericarditis, Bell's palsy, and tinnitus. It explains that more than 90% of early VAERS reports involved the Pfizer and Moderna vaccines, so if both raised the risk of an event by similar amounts, comparing one vaccine against a database dominated by the other could produce no alert.
According to the BMJ, FDA medical officer Dr. Ana Szarfman, working with statistician William DuMouchel, who developed the FDA's Bayesian algorithm, warned officials in early 2021 and proposed an updated method but was asked to stop. The report also quotes the official who headed the FDA's vaccine pharmacovigilance division writing more than two years later that the agency had been aware of the limitation before and during the pandemic.
"Very few people understand the statistics," Szarfman told the journal. DuMouchel said of officials, "I think that they were wrong."
Where Critics Say the Story Goes Too Far
In a detailed critique in Science-Based Medicine, surgical oncologist Dr. David Gorski argued that the investigation documents a real methodological debate but frames it as a cover-up. He relied heavily on analysis by statistician Jeffrey Morris, who has written a white paper on U.S. vaccine safety monitoring.
Their central point is that the main problem was the comparison group, not the Bayesian method itself. According to Morris, the FDA's flagging threshold was deliberately more conservative, producing fewer false alarms at the cost of lower sensitivity. A method that flags more events is not automatically better, he argued, unless its false-positive and false-negative rates are compared under realistic reporting conditions.
Critics also noted that the CDC's 2022 analysis pooled the two mRNA vaccines and compared them against non-COVID vaccines, so the two could not mask each other there. They said the jump in alerts reflected both a different comparison group and a looser threshold, and that the myocarditis signal emerged in spring 2021 as younger people, who face the highest risk, became eligible for vaccination.
Gorski also highlighted a disclosure in the article: the investigation is adapted from an upcoming book by Willman and Peter Doshi, a senior editor at the BMJ's investigations unit whom Gorski has long criticized. The BMJ says the piece was externally peer-reviewed.
Safety Signals Are Starting Points, Not Verdicts
A safety signal is a statistical flag suggesting that an event is reported more often than expected after a particular vaccine. It is a prompt for further study, not proof of harm. The CDC says that a report to VAERS does not mean a vaccine caused an adverse event, and that the CDC and FDA investigate further when an association appears possible.
VAERS accepts reports from anyone, which makes it fast but noisy. During the COVID-19 rollout, officials actively encouraged reporting, and health care providers had specific reporting requirements, which sharply increased report volume. Both the BMJ and its critics agree that this surge shaped how the algorithms behaved.
When VAERS suggests a possible problem, the CDC and FDA follow up using other monitoring systems, a step designed to separate coincidence from true side effects.
Every screening method balances sensitivity against specificity. If a method is too sensitive, investigators chase many false alarms; if it is too strict, real problems surface later. That trade-off is a legitimate scientific question, and it is at the heart of this disagreement.
Questions That Remain for Regulators
Several points remain unresolved. It is not clear whether federal agencies have changed how they screen VAERS data since 2022, and the investigation does not quantify how much earlier any specific risk might have been recognized under a different method.
For patients, the practical meaning is limited. The report concerns early screening decisions, not new clinical evidence, and myocarditis is already recognized as a rare risk after mRNA vaccination, particularly in young males. People with questions about COVID-19 vaccination should talk with a clinician about their own age, health, and risk.
Congressional interest may also grow, since the BMJ said its reporting drew partly on records released in response to U.S. Senate investigators.
Key Questions Answered
What did the BMJ investigation find? It reported that the FDA and CDC relied on a VAERS screening method that officials had been warned could miss signals, and that an FDA officer who proposed a fix was told to stop.
What is masking? When nearly all reports involve two similar vaccines, comparing one against the other can hide a risk both share.
What do critics say? They argue the main issue was the comparison group and a deliberate trade-off to reduce false alarms, not a cover-up, and that the framing overstates wrongdoing.
What is a vaccine safety signal? A statistical flag that an event is reported more often than expected. It calls for further study and does not prove the vaccine caused harm.
Does this change vaccine guidance? The investigation does not present new clinical data, and no agency has announced a guidance change based on it.
Is myocarditis a known risk of mRNA vaccines? Yes. It is recognized as a rare risk after mRNA COVID-19 vaccination, seen most often in young males.
Published by Medicaldaily.com
