Open Altio
Twenty years of FDA adverse event reports, cleaned and deduplicated, free to use, with the code and every rule in the open.
Available now
What comes with the cleaned FAERS release today.
The method and match rules are published with the evidence behind them. Probable duplicates are found even when a case is missing a field such as the event date or age. Duplicates are flagged, never deleted, and each data release is versioned by FDA quarter with its own counts and changelog.
Active ingredients are split out, and combinations keep every ingredient. Mappings to RxNorm codes and WHO ATC classes are built in.
EBGM with EB05 and EB95, PRR and ROR, plus RRR and IC. Each one has a methods note, and the common thresholds are built in, with settings you can change.
Each reaction is checked against the drug’s current FDA label from DailyMed and marked as on the label or not on the label.
For organizations that give their MedDRA subscriber ID, checked with MSSO: reaction terms in the current MedDRA version, with the MedDRA hierarchy and SMQs. Everyone else sees the preferred terms FDA published.
How we clean FAERS
FDA publishes every version of every report it receives. We keep one version per case, take out the cases FDA withdrew, and set probable duplicates aside, with every step and its counts on the methods page.
What we found in FDA’s files, and what each cleaning step did, is on the methods page, and in more detail inside Open Altio once you sign in. Disagree with a rule? Change it in your own copy and rebuild. The release records how yours differs from ours.
Access
Use the hosted service at open.altioai.com, or run Open Altio yourself from the open-source code on GitHub.
Hosted
Search, export and download the cleaned release at open.altioai.com. Apply for yourself, or for your organization. Altio reads every application and answers by email.
Only you. You can join an organization later, if its admin invites you.
Your organization joins once. You become its admin and invite your colleagues. Apply with a work email on its domain; free mail is not accepted.
Read the data license and the privacy notice.
Open source
The pipeline, the rules, the evidence and a local explorer are on GitHub under the Apache 2.0 license. Open Altio runs only on your machine, with no sign-in and no telemetry.
Install the code, then try the demo:
uv tool install open-altio
open-altio demoFor data scientists and engineers. Needs Python 3.12 or later, on macOS, Linux or Windows.
Coming next
FAERS is the first public safety database we clean. These follow, with the same open method.
The US Vaccine Adverse Event Reporting System, run by CDC and FDA. Vaccine reports, deduplicated and coded the same way as FAERS.
Health Canada’s database of adverse reaction reports for drugs and health products sold in Canada, normalized to the same case model.
Japan’s adverse drug event database, published by PMDA in Japanese. We plan to normalize it and translate drug names and fields into English, so it can sit alongside FAERS.
License and contributing
FAERS reports do not establish that a product caused an event.
Thanks
We are grateful to the DiAna project at the University of Bologna, whose drug dictionary we build on; to Banda and colleagues, whose AEOLUS resource (Scientific Data, 2016) gave us our first duplicate rule; and to Erasmus MC, whose faersdbstats and aioli projects we checked our cleaning steps against.