78.4% of medical device types are classified differently depending on the market. Here is the data, the method that produced it, and what it means for a submission plan.
Of 3,093 device types registered in 4 or more markets, 2,424 — 78.4% — carry a different risk class in at least one of them. 391 device types manage the full spread: High in one market, Medium in another, Low in a third.
This is not published anywhere, because it is not any single registry's fact. Each authority states its own classification and has no reason to state anyone else's. The comparison only exists if you hold the same device type across many registries at once.
Risk class drives the submission route, the evidence package, the review timeline and often whether a local licence holder is required at all. A device type that is Medium at home and High in a target market is not a translation exercise — it is a different submission, usually a more expensive one, and frequently the thing that turns a six-month plan into an eighteen-month one.
The direction is not consistent either, which is what makes it hard to intuit. Brazil is stricter than the European Union on some implants and more permissive on some instruments. Assuming your home market is the conservative one is the error that costs a year.
A single percentage invites the question of whether it was chosen. It was not: the finding holds, and strengthens, however narrowly you draw the population.
| Registered in | Classified differently | Device types | Rate |
|---|---|---|---|
| 2 or more markets | 2,841 | 4,288 | 66.3% |
| 3 or more markets | 2,602 | 3,513 | 74.1% |
| 4 or more markets | 2,424 | 3,093 | 78.4% |
| 5 or more markets | 2,228 | 2,758 | 80.8% |
We publish the 4-market figure because a device type present in 4 markets is one somebody has actually taken international, which is the population this question is asked about.
These are not the widest spreads in the data. They are the ones where a person has read the actual device records behind every cell, in every market shown, and confirmed that the records really are the device type named. That is a deliberately small number, and the reason is the most useful thing on this page.
Matching a device type across registries that each use their own vocabulary is the hard part of this work, and it is where the errors live. An accessory can end up filed under the code of the device it serves — the hand instruments used to place an implant carrying the implant's own code — and the result looks exactly like that implant being classed Low in that market. We checked a dozen candidate rows selected purely by how wide their spread was, and had to reject nine of them for precisely this. So we publish only what survived that check, with the evidence, rather than the twelve that looked most striking.
Each cell below rests on at least 15 registrations with 60% agreement, and the rows are re-checked against the data on every rebuild — if one changes shape it is pulled until a person looks again.
| Device type | Classed High in | Classed Medium in | Classed Low in |
|---|---|---|---|
| Bone fixation screw The clearest illustration of why a class label alone misleads. This is Class III in BOTH China and the European Union -- but China runs three tiers and the EU runs four, so the same label is the top of the scale in one and not in the other. Fourteen markets place it mid-scale, including the EU at 96% agreement over 8,417 records and South Korea at 100%. Only China treats it as its highest class. | China | Canada, Colombia, European Union, Hong Kong, India, Indonesia, Japan, Malaysia, Saudi Arabia, Singapore, South Korea, Taiwan, Thailand, Vietnam | — |
| General internal orthopaedic fixation system implantation kit The device type IS the instrument kit, not the implant, and the records match it -- Australia's are labelled fixation system implantation kit, internal. China classing the same kit High is the divergence. | China | Canada, Colombia, Hong Kong, Japan, Malaysia, Saudi Arabia, Singapore, Taiwan, United States, Vietnam | Australia, Brazil, Indonesia |
| Immunochromatographic antigen rapid test Records are Strep A, LH and M. tuberculosis rapid tests -- the device type is right. Three markets place it at the bottom of their scale and nine put it mid-scale -- nobody calls it top-tier. The European and Saudi records sit in the old IVD Directive's General IVD category, which was self-declared and genuinely the lowest tier; Taiwan's are Class I of three. | — | Brazil, Canada, China, Hong Kong, India, Indonesia, Singapore, Thailand, Vietnam | European Union, Saudi Arabia, Taiwan |
| In vitro diagnostic immunological reagent Records are immunoassay reagents (Alinity C barbiturates and methadone; SAA immunofluorescence assay). Correct device type in every market shown. The European and Saudi records are General IVDs under the old IVD Directive -- self-declared, the lowest tier -- against Class III in Colombia and South Korea, which is about as wide as this gets. | Colombia | Brazil, Canada, China, Costa Rica, India, Indonesia, Singapore, South Korea, Taiwan, Vietnam | European Union, Mexico, Saudi Arabia |
The same data is queryable for any device type, by GMDN code
(gmdn_code), GMDN term (device_type, partial is fine) or FDA
product code (fda_product_code) — each answer carrying the number of
records behind it. This call returns the first row of the table above:
curl -s https://meridiantrace.com/v1/mcp \
-H "content-type: application/json" \
-d '{"jsonrpc":"2.0","id":1,"method":"tools/call",
"params":{"name":"classify_device","arguments":{"device_type":"Bone fixation screw"}}}'
It returns the device-type hierarchy, the FDA product codes the type maps to, and the observed risk class in every market where we hold enough records to say — each with the sample size behind it. This tool answers without an API key.
These are observed classifications, drawn from the risk class recorded on actual registration records in each market, not from reading each country's classification rules. For every device type and market we report the modal class, the share of records agreeing with it, and the number of records it was drawn from. A market showing n=15 is a hint; n=3,000 is a pattern. Both travel with the answer so you can tell them apart.
Three exclusions apply, and they are worth stating because each one changes the answer:
Why the aggregate can be trusted further than any single row. Device type matching carries error, and we can measure some of it: where a registry publishes its own classification name we compare ours against it, and 193 device type × market answers currently disagree in the 1 market where that check can be run. That is a real problem for naming an individual device type, which is why the examples above are hand-checked. It is not a problem for the headline: removing every one of those failing answers moves it by 0.3 points. A percentage over thousands of device types absorbs matching noise; a single named row is destroyed by one bad cell. We would rather say which of those two we are doing.
That makes this a good way to find out where your assumptions are likely wrong, and a poor substitute for a regulatory determination. Registries also differ in how they name classes, which we normalise to Low / Medium / High — useful for comparison and lossy at the edges. Confirm against the market's own classification rules before you file.
Data as of 27 August 2026, drawn from 1,262,350 registration records across 22 markets, covering 6,679 device types and 33,818 device type × market answers. This page is generated from that data rather than written by hand; the figures above and the method described here come from the same build.