FDA adverse-event reports in AEMS and MAUDE: signals are not rates

A report in FDA’s Adverse Event Monitoring System (AEMS) or Manufacturer and User Facility Device Experience (MAUDE) data means someone submitted information about an event involving a named product. It does not prove the product caused the event, show how often the event occurs, or establish that one product is riskier than another. An incidence estimate needs a validated numerator and compatible exposure denominator; comparative or causal conclusions need appropriate study design, verification, and additional evidence.¹³

This distinction matters when a clinic, influencer, plaintiff advertisement, or manufacturer cites “thousands of FDA reports.” A large count can reflect wider use, longer time on market, mandatory reporting, duplicate follow-ups, stimulated publicity, coding practices, or a genuine safety issue. The count alone cannot tell which.

AEMS consolidation and MAUDE are not interchangeable views

FDA describes AEMS as a consolidation platform for medical products, vaccines, devices, tobacco, food, cosmetics, and veterinary medicines.¹ MAUDE remains a current public source for Medical Device Reports.³ Treat each query as a dated view with named product scope rather than assuming AEMS and MAUDE are permanently disjoint universes or interchangeable counts.

FDA’s 2026 postmarket-surveillance explainer describes how CDER uses spontaneous reports alongside active surveillance, literature, clinical trials, product utilization, and other data to investigate potential signals.² A public dashboard is one window into that process, not the final benefit-risk review.

A report is not a verified case narrative

FDA states that AEMS reports may be incomplete, duplicative, unverified, and influenced by other diseases or medicines, and that submission does not prove causation.¹ MAUDE has similar limits: MDR information can be incomplete, inaccurate, untimely, unverified, or biased, and causality often cannot be determined from the report alone.³

Read the reporter’s verb. “Patient received product and later experienced X” is a temporal association. “Reporter believes device caused X” is an opinion submitted to the system. “FDA determined” would require a separate FDA record.

Important missing fields can include medical history, exact product or model, lot or serial, dose and timing, procedure technique, operator, other exposures, diagnostic testing, treatment, outcome, explant analysis, and alternative explanations. An empty field does not mean “no.”

Duplicates can look like independent events

One event may generate a patient report, clinician report, facility report, manufacturer report, and follow-up updates. Manufacturer reports can reference earlier report numbers, but public searches do not always make every duplicate obvious. A corrected or supplemental report may resemble a new case.

Before counting:

• compare event dates, age, sex, location, product, lot or serial, narrative, and reporter;
• follow manufacturer and user-facility report numbers;
• identify initial, supplemental, corrected, and follow-up submissions;
• distinguish multiple devices in one event from multiple events; and
• document a deduplication rule before looking at outcomes.

Do not silently discard records because they look similar. Report raw count, probable event count, and uncertainty separately.

The missing denominator prevents incidence estimates

Incidence requires a numerator of defined events and a denominator of people, doses, devices, procedures, or time at risk under comparable ascertainment. AEMS and MAUDE do not contain a reliable complete exposure denominator.¹³ Underreporting and stimulated reporting further distort the numerator.

Examples of invalid calculations include:

• reports divided by units sold without matching geography, time, reuse, indications, and reporting capture;
• comparing two brands’ raw counts without exposure and time on market;
• using search-result totals as confirmed injuries; and
• treating a fall in reports after publicity fades as lower underlying risk.

A risk estimate needs a designed study or surveillance analysis with defined population, exposure, outcome validation, follow-up, comparator, confounding plan, and sensitivity analyses.

Report coding organizes narratives; it does not prove them

Medical Device Reports use codes for device problems, patient problems, evaluation results, and related fields. FDA maintains coding resources and updates them over time.⁵ Codes help group records but can vary by reporter interpretation, manufacturer evaluation, software version, and available information.

A “no known device problem” code can coexist with an injury report; it does not mean no event occurred. A device-problem code does not prove the device caused the outcome. Read the narrative and evaluation fields and note whether the device was returned, inspected, or unavailable.

For longitudinal analysis, preserve the code version and query date. A term can be added, retired, or mapped differently. FDA’s downloadable MDR data files support reproducible analysis but require joining multiple tables and handling updates.⁴ Dashboard screenshots alone are fragile evidence.

A better query starts with a falsifiable question

“Is this device dangerous?” is too broad. A useful question might be: “What reports between these dates mention thermal injury after this exact model and handpiece in this procedure, and what information would distinguish device malfunction from technique, maintenance, or patient factors?”

Predefining the query reduces cherry-picking after a striking narrative appears.

Search results can miss the product or overcapture neighbors

Drug reports can list brand names, generic names, concomitant products, or suspect roles. Device reports can use trade names, model families, product codes, manufacturer aliases, or component names. A broad query can include unrelated versions; a narrow brand query can miss generic descriptions.

Build a product identity table first. For a drug, use active ingredient, formulation, route, application number, manufacturer, and dates. For a device, use manufacturer, proprietary name, model, product code, 510(k) or PMA number, handpiece or accessory, and recall history.

Do not assume a product named in a concomitant field was suspected. Distinguish primary suspect, secondary suspect, concomitant, interacting, and unspecified roles when the database provides them.

Follow the signal into evaluated records

After identifying a pattern, look for FDA safety communications, labeling changes, recalls, warning letters, advisory-committee materials, post-approval studies, literature, Sentinel analyses, manufacturer notices, and updated instructions. The public-action guide explains what each record means.

The absence of a public FDA action does not prove no risk. Signal evaluation takes time, evidence can remain insufficient, and some risks stay within labeling rather than triggering a recall. Conversely, a safety communication may address a meaningful risk without establishing the incidence suggested by raw reports.

Date every conclusion: “As of August 29, 2026, FDA’s current page says…” This prevents an old dashboard analysis from outranking a newer label or communication.

Individual decisions need product and clinical context

A person reading an adverse-event narrative may recognize a symptom or procedure. The report cannot diagnose the reader or determine whether to stop a medicine, remove a device, dissolve filler, or repeat treatment. Use the exact product and treatment record with an appropriate clinical evaluation.

Preserve product name, lot or serial, dose or settings, dates, facility, professionals, consent, symptoms, photographs, laboratory or imaging results, treatments, and outcome. FDA reporting can accept incomplete information, but detailed records make a report more useful.

Use reports as a signal, then build the missing evidence

1. Name the product precisely. Map ingredient or device model, manufacturer, formulation or component, identifiers, approval record, and market dates.
2. Write the query before counting. Fix database version, date window, terms, roles, outcomes, exclusions, and duplicate rules.
3. Read narratives and follow-ups. Separate temporal sequence, reporter opinion, medical confirmation, manufacturer evaluation, and FDA conclusion.
4. Refuse an unsupported rate. Do not divide reports by an incompatible sales or procedure estimate; identify a valid exposure denominator or state that none exists.
5. Triangulate the signal. Look for active surveillance, trials, epidemiology, label changes, recalls, safety communications, and mechanism evidence.
6. Preserve and report the individual record. Keep treatment and product identifiers, clinical details, outcome, and follow-up so regulators can evaluate the submission.

The decisive question is: “Does this database result show a submitted signal, a verified event pattern, a causal finding, or a rate—and what denominator and follow-up evidence would be required to move to the next level?”