AI pharmacovigilance case processing, with a person on every field

Pharmacovigilance case processing is the work between "we received a report" and "a qualified person is willing to say what this case is." Seriousness, coding, causality, expectedness, and the reporting clock are that work. AI pharmacovigilance case processing, in the only sense this site uses the phrase, is a draft of those fields from the source text. It is not a safety database, not a governance layer inside someone else's database, and not a machine that files the report while the team is asleep.

The judgments, named plainly

A processed case can answer a short list of questions. If it cannot, it is still intake, which is fine, but it is not done.

• Seriousness. Serious or not, and which ICH criterion if it is serious. Severity grades are not a substitute.
• Coding. A MedDRA preferred term that matches the reporter's words, sitting in a system organ class. The verbatim stays visible next to the term.
• Causality. A WHO-UMC style call — certain, probable, possible, unlikely, or unassessable — with the time course and the obvious alternative causes.
• Expectedness. Listed or not in the Investigator's Brochure or company core data sheet, at the right specificity, on a named version.
• Reporting. Whether an expedited report looks required, to whom, and on a 7-day clock, a 15-day clock, or neither.
• Narrative. A few sentences that match the structured fields.

Aggregate writing, signal detection, and literature screening are other jobs. They use processed cases. They are not the processing of one case. PSUR, PBRER, and DSUR stay outside the workspace on purpose.

What the workspace actually drafts

You open the AE Connect workspace, paste the narrative, or upload a PDF, Word file, text, email, CSV, or image. The file is parsed in your browser. The model sees extracted text, not the original file. It returns the fields above, each with a confidence and a short reason, plus references it thinks apply. You accept or override. You can download the case. There is no e-signature step on this site. Sign-off, if your quality system requires a Part 11 signature, belongs in the system your validation package covers.

When the model is unsure, the draft should look unsure. Expectedness is the clearest example. The workspace does not contain your brochure. A suggestion of "unexpected" means "go check the document," not "the core data sheet has been consulted." MedDRA suggestions are the same kind of aid. Confirm the term. Do not paste a suggested code into an official case without looking it up in the MedDRA version you are licensed to use.

If no API key is configured, the workspace falls back to a deterministic mock so the screens can be tried. A mock is not a medical assessment. Production use needs the real model and a reviewer who knows that.

What stays human

The reviewer is the drug-safety decision. The draft can be faster than a blank form and still be wrong about hospitalization, about which product was suspect, or about a second event buried in paragraph four. Override is the product working, not the product failing. A team that accepts every field without reading the source has automated a risk, not a process.

The same limit applies to SUSAR calls and to E2B(R3) content. The draft can show that a 7-day clock might apply. It cannot start the clock, unblind the subject, or send the file. AE and SAE reporting remains the sponsor's duty.

Where this sits relative to a safety database

AE Connect integrates with any safety database, including Argus, LifeSphere, and Veeva Vault Safety. Those products remain the system of record, including submission and aggregate reports. Read how that works for Argus, LifeSphere, and Vault Safety.

A company email starts a 14-day sandbox trial. Annual subscriptions are Starter, Growth, and Scale at $36,000, $72,000, and $120,000 per year. Pricing has the Contact sales and Book a demo links.