The Two Clocks

You built the working version over a weekend. It has been in review since the spring. Your clock says this took two days; their clock says nothing has happened. Nobody is lying — both instruments are reading correctly, and the danger isn't the review board. It's that you eventually adopt the outside view of yourself.

FDA is the most legible version of this problem, because the outside clock is published. Every 510(k) submission has a date FDA received it and a date FDA decided. Subtract, and you get the number everyone quotes.

It is also, as FDA will tell you itself, the wrong number.

Every specialty, one year at a time

Each dot is one FDA medical-specialty review panel in the selected year. Horizontal position is the median calendar days from receipt to decision — log scale, because the tail is long. Vertical position is the share of that panel's decisions that took more than 180 days. Dot area is the number of decisions.

The two dashed lines are FDA's own commitments under MDUFA V: a 90-day goal measured in FDA Days, and an average total time to decision of 128 calendar days for submissions received in FY2023, falling to 112 by FY2025. Everything on this chart is calendar days. Hold that thought.

Thirty-five years of the outside clock

The same measurement, every decision year since 1990. Red is the median; navy is the 90th percentile — the wait that one submission in ten exceeds. The selected year above is marked.

Two things are true at once here. The horizon moved — the median calendar wait roughly doubled between the 2005 trough and 2024. And the horizon has moved before and come back: the 1993–95 spike was worse than anything since, and it collapsed.

A horizon isn't a wall. It's two clocks disagreeing — and the reading on either one is a fact about the observer as much as about the thing being observed.

Now drag the sample-size floor

Set the year to 2023 and leave the floor at 1. There is a dot far out on the right at 914 days — Pathology. It is the most alarming point on the chart and it is a single submission. One device, one number, one panel that in most years handles between one and four 510(k)s.

Drag the floor up and the horror show empties out:

Every value computed from the FDA 510(k) database, decision-year cohorts, substantive-decision records with a receipt date on or before the decision date.

At n ≥ 50 the cloud collapses to a tight band and the story changes completely: there is no catastrophically slow specialty. There is one system, drifting.

This cuts both ways, and the direction matters. Small samples can invent a pattern — the 914-day panel that is one device. They can also bury a real one in noise. The only way to know which you are looking at is to move the floor and watch which direction the number goes.

The clock this chart cannot see

Here is the part that should make you distrust every number above, including the ones on this page.

FDA does not grade itself on calendar days. Its performance goal — a MDUFA decision within 90 FDA Days for 95% of standard 510(k)s — counts only the days the submission is on FDA's desk. When the agency issues an Additional Information request, the clock stops. It restarts when the sponsor responds. Weeks or months of a submission's calendar life can belong entirely to the company that filed it.

So the drift in the chart above is real, and it is not a measurement of FDA getting slower. It is the sum of two things that the public database cannot separate: FDA review time, and sponsor hold time. The most likely reading of a rising calendar median is more AI requests — which is partly an FDA behaviour and partly a submission-quality problem.

The same decomposition exists for your project. "It has been in security review since the spring" almost always contains a fortnight where the questionnaire sat in your inbox. That is not a moral failing. It is the reason the outside clock reads the way it does, and the reason it is a bad instrument for deciding whether you are still moving.

Where this chart is thin

• Calendar days, not FDA Days. The public 510(k) database publishes a receipt date and a decision date and nothing in between. Every duration here therefore includes sponsor response time that FDA's own metric excludes. This chart is the outside clock by construction.
• Decision cohorts, not receipt cohorts. MDUFA reports group submissions by the fiscal year FDA received them. This page groups by the calendar year FDA decided them, because receipt cohorts near the end of a data extract are missing their slowest members. The two framings do not produce the same number and should not be compared directly to MDUFA's published figures.
• A panel is not a technology. "Radiology" in 2024 contains ultrasound transducers and autonomous image-triage algorithms in the same dot. Specialty medians hide the thing most readers actually want to know.
• Cleared devices only, and cleared devices survive. Withdrawn and never-decided submissions are absent. The submissions that fell in and did not come back out are, appropriately, invisible.
• Fiscal versus calendar year. FDA's goals are stated on a fiscal year; this page uses calendar years. The 128-day and 112-day reference lines are approximate markers here, not a scoring line.
• One vintage. A single August 2026 extract. Records with implausible or inverted dates were dropped; 2025 is excluded as incomplete.