Your Next Risk Signal Is Already in MAUDE: Why Safety Signaling Starts With the Narrative
The Signal is Already Public – Just Hard to Uncover
Post-market teams rarely suffer from a lack of data. FDA's MAUDE database holds adverse event reports from every manufacturer in each product code, and every month an average of 250k to 300k new reports are added. The challenge sits with what is inside each report. The most informative part, the narrative describing what happened is unstructured free text; and free text is slow to read, difficult to search and nearly impossible to extract trends manually.
As a result, most teams fall back on what is easy to count. They filter by device problem code, run keyword searches, and track the totals. Not surprisingly, this approach has a blind spot. A single real-world problem is generally spread across a dozen codes and described in many different ways (often depending upon the reporter), so it never resolves into one clear signal.
Safety Signaling is designed to close that gap. This article looks at what changes when analysis starts with the narrative rather than the code, and why that shift carries even more weight now that QMSR is in effect.
Why Codes and Keyword Searches Fall Short
FDA receives more than three million medical device adverse event reports each year and uses them to monitor device performance and detect potential safety issues (FDA). Manufacturers are expected to keep watch over the same data. Three factors make that harder than it sounds:
The data has known limits. FDA describes MAUDE as a passive surveillance system. Because of under-reporting, inaccuracies and the lack of verification that a device caused an event, the database alone cannot establish incidence, prevalence, or cause (FDA MAUDE). None of that makes it less useful. It simply means the reports need to be read carefully and interpreted in context.
The same problem is coded in many different ways. Assigning a code is a judgment call, and manufacturers make different calls. For example, in catheter guide wires (listed under product code DQX), reports describing “guide wire fracture” were coded as: Difficult to Remove (114); Material Separation (95); Difficult to Advance (73); Break (27); and several others. A search on any one of those codes returns only a slice of the problem. A keyword search misses every report that used different language.
Public data does not mean clean data. MDR reports are full of misspellings, abbreviations, inconsistent brand and manufacturer names, and reports misfiled under an unexpected product code. In our work with manufacturers, we see that roughly 30% of relevant, publicly available reports never surface in a standard keyword search. The reports are in the database. They simply don't match the terms that teams would generally search for.
The most complete picture can be drawn from the narrative. That is where effective signaling must begin.
How Narrative-based Safety Signaling Works
Basil Systems' Safety Signals follows four steps, each building on the one before it.
1. Select a product code. In this example, DQX (catheter guide wires).
2. Read every narrative in context. Our models inform LLMs what each report describes rather than relying on the codes alone. For DQX, that meant examining roughly 3,500 report narratives filed between October 2024 through July 2026.
3. Group reports into problem topics. Similar events are grouped into topics, and each grouping carries a risk level, a report count, and an injury count. The DQX reports in our example resolves into 70 distinct topics.
4. Trend the topics and compare the codes. A heatmap shows which problems are rising and which are declining – along with the severity of each grouping – month over month. An IMDRF view then shows which code each manufacturer chose for the same problem.
Using our “grouped problem” analysis, the largest DQX topic provides the output:
On the heatmap, this topic reached 19 reports in May, the highest monthly count in the period reviewed. Because each report remains linked to the codes its manufacturer selected, the comparison across companies is direct and specific.
That is the practical difference between a large volume of reports and a signal a team can act on.
Why this matters more now
Several regulatory guidelines and requirements now converge on the same point – external post-market data should be monitored, trended, and carried into risk management; and that work should be documented.
QMSR. Since February 2, 2026, FDA's Quality Management System Regulation has incorporated ISO 13485:2016 by reference. On the same date, FDA retired its long-standing inspection technique and moved to a risk-based approach under Compliance Program 7382.850 (FDA QMSR FAQ). Investigators now use a manufacturer's risk management file to decide where to focus. FDA lists feedback or post-market information that is not used as input into risk management among its examples of an Official Action Indicated finding (King & Spalding).
ISO 14971, clause 10. The risk management standard requires manufacturers to actively gather production and post-production information, including publicly available information about similar devices. That information is then reviewed for new hazards, identifying risks that are no longer acceptable and changes in the state of the art (BSI).
EU MDR, Article 88. Manufacturers must report any statistically significant increase in the frequency or severity of non-serious incidents that could significantly affect the benefit-risk analysis (MDR Article 88). A trend can only be reported if someone is actively looking for it.
Regulators are applying AI as well. In December 2025, FDA deployed agentic AI (ELSA) for its staff, identifying post-market surveillance, inspections, and compliance among the primary intended uses (FDA).
Along with everything above, the public data a manufacturer reviews and incorporates is increasingly being reviewed by the agency as well.
From Signal to Action
A signal has value only when it leads to a decision that can be shown and defended. The table below compares how each step typically works today with how it works when signaling starts from the narrative.
The aim is not to chase every report. It is to make sure that every meaningful signal ends in a documented, defensible decision.
To see which problems are rising in your product codes, contact the Basil Systems team for a walkthrough.
Frequently Asked Questions
What is safety signaling for medical devices? Safety signaling is the process of identifying possible new or changing risks in post-market data, such as adverse event reports, before they grow into larger problems. A signal indicates a pattern that deserves closer review by a manufacturer. It is not, by itself, proof that a device caused harm.
Why not search MAUDE by IMDRF code? The same problem is frequently coded and described in many different ways across manufacturers and the reports themselves. Additionally, the data itself is inconsistent, so a code or keyword search captures only part of it. Our customers have documented and shared with us that roughly 30% of relevant reports never appear in standard keyword searches in MAUDE. Using Basil to analyze the narratives allows reports to be grouped by what actually happened.
How does Safety Signaling relate to QMSR? Under QMSR, effective February 2, 2026, FDA inspections follow a risk-based approach. Post-market and feedback data that is not used as input into risk management is one of FDA's examples of a serious finding. Basil's Safety Signals provides the data the FDA seeks.
Does ISO 14971 require reviewing information about other companies' devices? Clause 10 asks manufacturers to collect publicly available information about similar devices as part of their production and post-production information.
How does this support EU MDR trend reporting? Article 88 requires manufacturers to report statistically significant increases in non-serious incidents. Monitoring problems month over month makes it easier to identify increases early and to understand them in context.
Sources
- FDA, Medical Device Reporting (MDR): How to Report Medical Device Problems
- FDA, MAUDE: Manufacturer and User Facility Device Experience
- FDA, Quality Management System Regulation: Frequently Asked Questions (updated October 2, 2026)
- King & Spalding, The QMSR Goes Live and FDA Implements a New Medical Device Inspection Technique
- BSI, Production, post-production and the new ISO 14971
- EU MDR, Article 88: Trend reporting
- FDA, FDA Expands Artificial Intelligence Capabilities with Agentic AI Deployment (December 1, 2025)
- Basil Systems Safety Signals, product code DQX (October 2024 to July 2026)