On March 11, the FDA launched the Adverse Event Monitoring System (AEMS), consolidating several fragmented reporting databases into a single real-time public dashboard with artificial intelligence processing covering all FDA-regulated products. The new system promises future advances, including data analytics tools. But the agency made a consequential omission: It launched a powerful public-facing tool without building in the infrastructure to help people understand what the data can and cannot tell them. Without proper training, the adverse drug event information presented in the dashboard might mislead users and could lead to confusion over the safety of FDA regulated products.
Further, this change was made without opening a formal comment period where pharmacoepidemiologists and drug safety scientists could have weighed in before launch. Solicited feedback on the operability of the public dashboard could prevent future misapplication and misinterpretation of the adverse drug event data presented in the new format.
In this piece, we lay out the issues with AEMS, highlighting the need for proper contextualization and understanding of underlying data-generating processes. We close with a call to FDA leadership to invest in public comment solicitation and investment in public-facing training materials.
Adverse Drug Event Information Needs Context
Adverse event reporting systems are, by design, hypothesis-generating tools for researchers. A report in AEMS suggests that there is a suspected connection between a product and an outcome. Reports vary in completeness and quality as the system accepts submissions from manufacturers, clinicians, and the public, with no uniform verification of the clinical details. Thus, causal claims should not be made based on the number of reports without additional context.
For example, AEMS, applied specifically for medication reports from the legacy FDA Adverse Event Reporting System’s (FAERS), can tell you how many adverse event reports were filed for a given drug. It does not tell you total drug usage. Without that number, an incidence rate cannot be calculated, and without an incidence rate, you cannot compare the safety profiles of two products from this data alone.
In the United States, drug usage data exists in fragmented systems across insurers, pharmacy benefit managers, and health systems. That is a structural problem that the new AEMS still does not solve. Unfortunately, the dashboard does not highlight or explain this important caveat, so users could incorrectly interpret the information from the database without understanding this important limitation.
To overcome the issue of contextualizing data without knowing drug use, researchers have developed analytical methods like disproportionality analyses which identify safety signals of drug-event pairs that appear more frequently than expected relative to other drugs in the database. While these methods do not require specific knowledge of number of users of the medications, these tools do require methodological training to apply and interpret correctly. Also, this calculation is unavailable in AEMS and requires researchers to perform it themselves. The results of this calculation can be inflated if deduplication of cases is not conducted. The AEMS data contains duplicate cases that require clean up before disproportionality analyses.
Reporting bias makes the problem of potential misinterpretation bigger. According to experts only about 10 percent of adverse events are ever reported. That single number should give any end user of the system some pause before drawing conclusions from these raw counts. Importantly though, reporting rates are responsive to media attention: When a drug or vaccine receives press coverage, event reporting spikes, producing the statistical appearance of a safety signal where there may be none.
We already know what happens when the public encounters uncontextualized adverse event data. During the COVID-19 pandemic, anti-vaccine groups cited raw death counts from VAERS (the vaccine-specific reporting system now folded into AEMS) as evidence that vaccines were killing thousands of people. VAERS reports are unverified, and a death reported after vaccination may have nothing to do with the vaccine itself. But in an August 2022 nationally representative survey conducted by the Annenberg Public Policy Center, only about one in four US adults understood that.1
To its credit, the FDA includes data limitation disclaimers on the AEMS dashboard public announcement and published an FAQ page addressing common misinterpretations. Nevertheless, these disclaimers are not automatically apparent to new users who might not intuitively recognize the need for drug usage data for context. The press release described AEMS as one component of a broader surveillance program and noted that reports do not confirm causation. Notably, though, VAERS carried similar warnings during the COVID-19 pandemic. These warnings did not prevent widespread public misinterpretation of the data.
Expert Input and Additional Training Can Improve the Public Dashboard
The lessons learned from decades working with these adverse drug event data would have been an important addition in the redesign of the entire adverse event surveillance infrastructure. Section 3075 of the 21st Century Cures Act specifies that best-practices guidelines for using the adverse event system must be developed "with input from experts qualified by scientific training and experience to evaluate the safety and effectiveness of drugs." Pharmacoepidemiologists and pharmacovigilance scientists have extensive experience working with these data systems and training others to interpret them responsibly. Unfortunately, the FDA did not solicit scientific input on how the public-facing dashboard should communicate the data's limitations in an overt manner.
A dashboard displays raw material but does not provide a transparency system. Transparency also requires adequate dissemination and translation. We strongly suggest that the FDA open a formal public docket soliciting expert and client input on how AEMS data should be contextualized for non-specialist audiences, what additional limitations should be disclosed at the point of query, and how the system's AI tools for signal detection should be documented and validated.2
Beyond the dashboard itself, the agency needs to invest in active public education including a webinar series explaining what adverse event reports can and cannot tell you, short video modules designed separately for clinicians and for the public, and partnerships with professional societies to disseminate interpretive guidance.
While democratizing data is important, doing it intentionally is essential. When people encounter raw adverse event counts without context, the most intuitive conclusion is often the wrong one. In a media environment built on speed, that misinterpretation can spread faster than any correction. The AEMS needs a careful user manual, and the FDA is best situated to produce it.
- Jamieson KH, et al. Misinformation and the Vaccine Adverse Event Reporting System. JAMA. 2024;331(12):1005–1006.
- Andeson M, et al. A comment about AI methods that may help advance pharmacovigilance. Clin Pharmacol Ther. 2026; in press.
