HHS Develops AI Tool to Analyze Vaccine Injury Claims

The US Department of Health and Human Services is advancing a new generative artificial intelligence tool designed to analyze data within a national vaccine safety database. This system aims to identify patterns and formulate potential hypotheses regarding adverse events following immunization. According to a recent agency inventory of AI projects planned for the coming year, the tool is currently in development and has not yet been deployed. However, its potential application has raised significant concerns among public health experts, who fear its outputs could be leveraged to support an anti-vaccine political agenda.

The initiative has been in development since late 2023, as noted in previous agency reports. Critics are particularly worried that HHS Secretary Robert F. Kennedy Jr., a long-standing vaccine skeptic, might use the AI-generated predictions to further policies that undermine public confidence in vaccines. During his tenure, Kennedy has already made substantial changes to the nation’s childhood immunization schedule, removing recommendations for several vaccines, including those for Covid-19, influenza, and RSV.

Kennedy has also been vocal about his desire to overhaul the current vaccine safety monitoring framework. He has criticized the Vaccine Adverse Event Reporting System (VAERS), claiming it suppresses information about the true incidence of side effects. Furthermore, he has proposed modifications to the federal Vaccine Injury Compensation Program that could lower the legal threshold for claiming vaccine-related injuries, even for events not scientifically proven to be linked to immunization.

VAERS, a system jointly managed by the CDC and FDA, was launched in 1990 as an early warning mechanism to detect potential safety signals after a vaccine’s approval. It operates as a passive reporting system where anyone, healthcare providers or the public, can submit a report of an adverse health event following vaccination. A critical caveat is that reports to VAERS are not verified, and the data alone cannot establish that a vaccine caused any specific adverse event.

“VAERS, at best, was always a hypothesis-generating mechanism,” explains Dr. Paul Offit, a pediatrician and vaccine expert at Children’s Hospital of Philadelphia. “It’s a noisy system. Anybody can report, and there’s no control group.” The CDC’s own website explicitly states that a VAERS report does not confirm causation. Despite this, anti-vaccine activists have frequently misrepresented VAERS data over the years to falsely argue that vaccines are unsafe.

The move toward using advanced AI is not entirely unprecedented. Government scientists have utilized traditional natural language processing models to sift through VAERS data for several years. “It’s not surprising that HHS would move toward the adoption of more advanced large language models,” notes Leslie Lenert, formerly of the CDC and now at Rutgers University.

A well-known limitation of VAERS is its lack of denominator data, it does not track how many people received a vaccine. This absence can make reported events appear more common than they truly are in the vaccinated population. For this reason, Lenert emphasizes that information from VAERS must be paired with other robust data sources to accurately assess any potential risk.

The use of large language models introduces another layer of complexity, as these systems are prone to generating convincing but false information, or “hallucinations.” This underscores the necessity for human experts to rigorously vet any hypotheses produced by the AI tool. Lenert cautions that while “VAERS is supposed to be very exploratory,” there is a risk some officials may begin to treat its outputs as more definitive than they are, a shift that could have serious public health consequences.

(Source: Wired)