How to monitor FDA without drowning

Most device companies monitor FDA the same way: a Google Alert, the RAPS newsletter, and a colleague who forwards things. It works until it doesn’t — and when it doesn’t, the miss is rarely obscure. It was sitting in one of four public, official, machine-readable sources the whole time. This post is the monitoring method we built FDA Radar around. You can run it yourself, by hand, for free; the product exists because most people who try that stop by week three.

The four sources that carry the risk

The Federal Register is where regulation actually changes — proposed and final rules, guidance availability notices, classification orders. If a requirement that binds you moved, it moved here first, with a docket number and a comment deadline attached.

Warning letters are FDA showing its enforcement hand. A letter to a competitor in your device class is not gossip; it is the agency telling you, in public, which failure it is currently looking for. Read a quarter’s letters in your space and the inspection theme is unmistakable.

Recall enforcement reports matter twice. Once for your own supply chain — components, suppliers, contract manufacturers — and once for a reason people miss: a recall on a device you cited as a predicate can bear directly on the comparison your clearance stands on.

510(k) clearances are the slow-moving map of your competitive and regulatory neighbourhood: who just got cleared in your product code, against which predicates, through which pathway. Nothing about your obligations changes when a clearance posts — but your next submission’s predicate landscape just did.

Why the naive version fails

All four sources are public. Three have official APIs. Any competent engineer can wire them into a feed in a weekend — we say this from experience, because that weekend version is roughly where every “AI regulatory intelligence” tool starts and most of them stop. The problem is volume: the 510(k) database alone publishes thousands of clearances a year, and routine Class II recall entries arrive in batches that would bury any inbox. A feed that forwards everything is a firehose pointed at the one person in the company least able to spare the time. Firehoses do not get read. They get filtered to a folder, and the folder is where the material item goes to die.

The discipline that makes monitoring work is not collection. It is refusal: deciding, explicitly and in advance, what does not deserve your attention.

Our version of that refusal is structural. Routine recall entries and the 510(k) firehose route to a searchable corpus — present, queryable, out of the way. What remains for human review is the material residue: roughly eighteen items in a typical week. That number is the entire point. Eighteen items is an amount a professional actually reads.

Running it by hand

If you run this method yourself, the weekly loop looks like this: search the Federal Register for your CFR parts and the word “device” in agency “Food and Drug Administration”; scan the quarter’s CDRH warning letters for your device class; query the enforcement report database for your product codes and your suppliers’ firm names; and check new clearances in your product codes. Keep a log of what you decided not to act on — the decision not to act is the part an auditor asks about.

The loop takes a disciplined hour a week when nothing happens, and considerably more the week something does. FDA Radar is that loop, run daily, with the join against your actual product portfolio done by identifier rather than by memory — and a human reviewer deciding what is material before anything reaches you. Either way, the method is the same. What we sell is the discipline of it, on the weeks you don’t have the hour.