How the site counts reports
Every figure on this site is a count of reports in the openFDA files. Nothing is modelled or estimated. This page says how a report becomes a count, and what the counts leave out.
- Drug reports: the openFDA copy of FAERS, release of , with 20,646,523 reports. The newest report was received on .
- Device reports: the openFDA copy of MAUDE, release of , with 26,326,888 reports. The newest report was received on .
FDA is implementing a new system, AEMS, that consolidates its reporting systems. FDA's public dashboard for drug reports now has the name AEMS. The openFDA files still use the name FAERS, and so does this site.
Editorial policy
- A sentence of FDA is a quotation. It is shown with quotation marks, the name of the FDA page and the date on which the site read the page. The site never changes it.
- A sentence of the site comes from a fixed rule. It gives a fact and a count with its period. It never says that a product caused an event.
- The sentence "A report is not proof that the drug was the cause of the event" is the site's own. It stands above the words of FDA and is not a quotation.
- The site gives no advice. It does not say to take, stop or change a medicine.
- The site gives no rate, because the number of people who use a product is not in the data.
- The site does not rank products. A table of several products is in order of name and has no sort by count. FDA's words about counts of reports stand beside it.
- A page title has no count and not the words "side effects".
- The site never shows a name that reads as the name of a person.
The site reads the FDA pages again when it changes a quotation. Corrections says how to report an error.
What one report is
A drug report is one submission to FDA about one patient. It lists the drugs, the reaction terms, the outcomes, the age and sex of the patient, and the type of reporter.
A device report is one submission about one event. It gives the brand name, the manufacturer, the product code, the event type and coded problems.
The site counts reports. It counts a report once for each drug, device or term that the report names. So a report that records five reaction terms is in the count of each term.
How the site names a drug
A reporter writes the product name. The site matches the name to the National Drug Code Directory, after it removes salt abbreviations and dose form words.
A name that is a brand name in the directory gets a brand page. A name that is a generic name gets a generic page. A name that matches nothing stays as the reporter wrote it.
A brand and its generic are separate pages, because reporters name them separately. A drug page shows the other names of the same active ingredient with a sum. The sum adds separately reported names. A report that names two of these names is in the sum two times.
Some reported "drug names" are not the names of medicines. These names never take the label or the class of a product with the same spelling: Radiation, Radiation therapy, Radiotherapy, Chemotherapy, Drug, Unknown, Placebo, None, Unspecified. A page that the site published earlier for such a name stays on the site. It is marked noindex and says so.
Suspect, concomitant and interacting drugs
A report records the role of each drug. A drug page counts the reports in which the drug is suspect or interacting. The reports that name the drug only as concomitant are in a separate count. They are in no other figure.
FDA says about the role:
Reported role of the drug in the adverse event report. These values are not validated by FDA.
Source: openFDA, Drug adverse event fields (YAML reference). The site read the page on .
Any number of the drugs may be marked as suspect if thought to be responsible for one or more of the reactions, but that information is not validated.
Source: openFDA, Drug Adverse Event Overview. The site read the page on .
Drug classes
A drug class is the Established Pharmacologic Class (EPC) on the label of the drug. A drug is in the class that its reports name most often. When the reports name none, the site uses the class of the National Drug Code Directory. A class page needs 2 drug pages.
Reaction lists
A reaction list shows each reaction term that 5 or more reports record for a drug or a class. A term with fewer reports is not in the list.
The site keeps reaction terms only for a drug and a year in which the drug has 12 or more reports. For such a year it keeps each term that two or more reports record, and the 20 most frequent terms.
So a count in a list can be lower than the count of FDA. A term that one report records in each year is not in the list.
- If all the years of a drug have terms, the page says so. It says that a count can be lower than the count of FDA.
- If the years with terms hold under 90 percent of the reports, the page says that the list is partial. It gives the share of the reports that the list covers.
- If they hold under 50 percent, the page says that the list is not the full count of FDA.
The CSV file of a drug or a class has the same rows and the same limits. Its last row is a note with these words.
A reaction term has a plain meaning under it. The meaning is a gloss written by the site in its own words. It explains the word and adds nothing to a report. The terms come from MedDRA, the Medical Dictionary for Regulatory Activities. The site shows them as openFDA publishes them, and it quotes no definition from MedDRA.
Months, years and quarters
A month is the month in which FDA first received the report. The report may describe an event of an earlier year. 0 records with no usable receive date are left out.
The table of outcomes on a drug page is by the year of the openFDA file that holds the latest version of the report. It is not by the year of first receipt. The table says which year it uses.
The newest months are not complete. FDA releases drug data every quarter, and reports can arrive months after an event. The latest year is not complete, and a chart draws it as an outline.
A quarter is closed for a dataset when three things are true:
- The export has the file of the quarter.
- The newest receive date is the last day of the quarter or later.
- The total of the quarter is at least half of the total of the same quarter one year before.
The drug data and the device data are both closed to 2026-Q2.
Duplicates and follow-up reports
One event can have more than one report. An update, a second reporter or a manufacturer can each add one. The public files do not carry enough data to link them. The site counts the records as openFDA publishes them, without removing duplicates.
So a count is a count of reports. It is not a count of events or of patients. FDA says:
There are many instances of duplicative reports and some reports do not contain all the necessary information.
Source: FDA, FDA Adverse Event Monitoring System (AEMS) Public Dashboard. Content current as of . The site read the page on .
Thresholds and the index
| Page | The site builds it from | Search engines may index it from |
|---|---|---|
| Drug name | 200 reports | 5,000 reports, with 100 or more of them in the latest 12 months and a report in the last five years |
| Device brand name | 150 reports | 3,000 reports, with 50 or more of them in the latest 12 months and a report in the last five years |
| Device type (product code) | 300 reports | 5,000 reports, with a report in the last five years |
| Drug class | 2 drug pages with 2,000 reports in all | 20,000 reports and 3 or more drug pages |
| Reaction term | 5,000 reports | 20,000 reports |
| Manufacturer | 1,000 reports | 5,000 reports and 3 or more brand pages |
| Month | 300 drug reports | The newest 13 months |
| Findings, lists, guides and the pages about the site | Always | Always |
The aim is that search engines list only the pages with enough figures of their own. A page under its index threshold keeps its address and its content. It carries a noindex mark, and the sitemap leaves it out. The index thresholds are higher than the thresholds for a page to exist, and the site can change them.
A drug name is in the A to Z directory from 500 reports. A device brand name is in it from 1,000 reports. Each needs a report in the last five years. The lists of the directory link every name in it.
A name under that threshold is in a second list. Search engines do not index the pages of that list.
Every page that search engines do not index has a link. The link is in a list of drugs, reaction terms, drug classes, manufacturers or months, or on the page of the device type.
Held pages
A held page is a page that the site published earlier and keeps. The new FDA data no longer meet the condition for a new page. A held page is marked noindex. It shows the date of the FDA data and the condition that the data no longer meet. The address of a page does not change, and the site never gives an address twice.
A drug or a device with no report in the last five years is also marked noindex. Its page says that FDA received no report in those five years.
Tables of several products
A table of reports for several products lists each product with a stated number of reports or more. The page states the number. The rows are in order of name. A product with fewer reports is not in the table.
The number is the smallest of 5, 10, 25, 50, 100, 250 and the larger steps that leaves a table of 150 rows or fewer. So the number differs from page to page. The table of drug classes on a reaction page has up to 60 rows. A table of device types of a manufacturer has up to 60 rows.
The table of other drugs in the same reports is a selection. It has the drug names that the reports name most often. The site tracks 2,500 drug names, and a pair needs 5 or more reports. The page states this.
FDA says:
Additionally, it is important to remember that adverse event reports represent a small percentage of total usage numbers of a product. Common products may have a higher number of adverse events due to the higher total number of people using the product.
Source: openFDA, Drug Adverse Event Overview. The site read the page on .
MDR data alone cannot be used to establish rates of events, evaluate a change in event rates over time or compare event rates between devices.
Source: openFDA, Manufacturer and User Facility Device Experience. The site read the page on .
Counts by quarter and updates
The site compares each build with the build before it. It compares the reports that FDA received in the latest closed quarter, for each drug page and device page.
A new closed quarter is shown as two counts and their quarters. When FDA changes the count of a quarter in a later export, the site shows the count before and the count now. It records a change only when the count moves by 5 reports and by 2 percent or more.
The update pages and the feed list a page only if search engines may index it and one of its counts is 10 or more. A different count is not a finding about a product. FDA says:
MDR data alone cannot be used to establish rates of events, evaluate a change in event rates over time or compare event rates between devices.
Source: openFDA, Manufacturer and User Facility Device Experience. The site read the page on .
Many factors can influence whether or not an event will be reported, such as the time a product has been marketed and publicity about an event.
Source: openFDA, Drug Adverse Event Overview. The site read the page on .
The information in these reports cannot be used to estimate the incidence (occurrence rates) of the events reported.
Source: FDA, FDA Adverse Event Monitoring System (AEMS) Public Dashboard. Content current as of . The site read the page on .
Names of people
The site reads no narrative, no patient detail and no reporter name. A product name, a device brand name or a manufacturer name that reads as the name of a person is not published. The site checks the names against patterns. Examples are a surname before a given name, a title, or a short name with no mark of an organisation.
When the brands of a manufacturer include such a name, the site counts them under the name (NAME WITHHELD). The filter is cautious. It can leave out a product name that is not a person.
Advertising
Only the pages of a drug, a drug class, a device, a device type, a manufacturer, a month and a reaction term can carry advertising. No other page carries it. This includes the home page, the lists, the guides and all text pages.
These 8 reaction pages never carry advertising. They show the 988 line:
- Completed suicide
- Intentional drug misuse
- Intentional overdose
- Intentional product misuse
- Intentional self-injury
- Poisoning deliberate
- Suicidal ideation
- Suicide attempt
Known gaps
- A count in a reaction list can be lower than the count of FDA. See the section on reaction lists.
- The data have duplicate reports. The site does not remove them.
- The data have no count of the people who use a product. No page gives a rate.
- A device can have more than one spelling of its brand name. Each spelling can be a separate page.
- A manufacturer name is free text. A subsidiary, a plant or a different spelling can be a separate name.
- FDA releases drug data every quarter. The newest months are not complete.
- The match of a drug name to a brand name or a generic name can be wrong. Corrections says how to report an error.
- The plain meaning of a reaction term is the site's own gloss. It is not a definition of FDA or of MedDRA.
- FDA is changing its reporting systems. A change of the files can change the counts of the site.
The page Data sources lists the FDA files and the FDA pages that the site quotes.
Last checked against FDA pages: .