Drop a screenshot here:assets/screenshots/signal-watch.png
An early-warning tool for drug safety, built to surface signals in FDA adverse-event data that simple counts would miss.
Drop a screenshot here:assets/screenshots/signal-watch.png
Overview
Every drug on the market can cause harms that clinical trials never caught. The FDA collects these as adverse-event reports in FAERS, a public database with over 20 million entries, but popular drugs accumulate large report counts simply because more people take them. Raw report counts alone are not a reliable signal.
Signal Watch computes the same disproportionality statistics used by pharmacovigilance teams to address this problem: PRR, ROR, and IC025, calculated across roughly 40,000 drug and reaction pairs pulled from the openFDA FAERS API. An analyst can triage each flagged pair, inspect the contingency table, track the quarterly reporting trend, and leave notes, all inside a full-stack workbench built on Aurora PostgreSQL and Next.js.
To keep signals reliable, the pipeline applies a term stoplist, minimum-count thresholds, and requires agreement across all three statistical methods before flagging a pair as confirmed. As a validation check, it also recovers several well-established drug warnings directly from the raw data.
Built with