The lowest defensible MDR class for your device is the one with the smallest regulatory burden you can honestly justify against a specific Annex VIII rule, a stable intended purpose under Article 2(12), and a reading of Annex VIII the Notified Body will accept. Minimising classification is not gaming the system. It is applying the rules precisely, with evidence, where they allow legitimate judgment — and accepting the higher class where they do not.

By Tibor Zechmeister and Felix Lenhard. Last updated 10 April 2026.

TL;DR

  • Minimising classification under Regulation (EU) 2017/745 means finding the lowest class that is defensible — not the lowest class that is wished for, and not the highest class a cautious consultant would recommend.
  • The leverage sits in three places: the intended purpose under Article 2(12), the specific Annex VIII rule and its sub-clauses, and the interpretive guidance in MDCG 2021-24 and MDCG 2019-11 Rev.1.
  • A defensible lower class requires a written argument that cites the rule, quotes the intended purpose, and addresses every escalation trigger the rule contains. No hand-waving.
  • Minimising fails — and should fail — when the lower class would compromise patient safety, when the intended purpose has to be distorted to fit, or when MDCG guidance explicitly places the device higher.
  • The final check is the Notified Body reviewer: if your argument would not survive a careful reading of Annex VIII against your own intended purpose, the class is not defensible, regardless of how much cheaper it would be.

What "lowest defensible" actually means

The phrase is deliberate. "Lowest defensible" is not "lowest possible." It is not "lowest we can talk a reviewer into." It is the lowest class that holds up when someone who reads Annex VIII every day walks through your intended purpose, maps it to the rule you cite, and checks the result against MDCG 2021-24 and the Borderline Manual.

The reason the framing matters is that founders hear "minimise" and think "negotiate." That is the wrong mental model. You are not negotiating with the Notified Body about how much regulation applies to you. You are applying the Regulation carefully, and the careful application produces a specific class. If the careful application produces Class IIa, Class IIa is your answer. If it produces Class I, Class I is your answer. The work is to do the application honestly, in detail, and with documentation — not to push for a lower number.

The distinction also matters for patient safety. The MDR classifies devices by risk for a reason: higher-risk devices carry higher obligations because the consequences of failure are more severe. A founder who "wins" a lower class by distorting the intended purpose is not saving money. They are carrying an unclassified risk forward into the market, and the PMS data will eventually expose it — usually at the worst possible time. Lowest defensible is the defence against that outcome. It forces the class to match the actual risk, as expressed in the actual intended purpose, as governed by the actual rule.

Where the rules give legitimate flexibility

Classification under Article 51 and Annex VIII of Regulation (EU) 2017/745 is partly a lookup and partly a judgment. The lookup parts — long-term implantables contacting the central circulatory system, for example — do not flex. Either the device does the thing the rule describes or it does not. No amount of careful argumentation moves a Rule 8 implantable out of Class III if it meets the criteria.

The judgment parts are where minimising is legitimate. Several places in Annex VIII leave genuine interpretive space:

  • Rule 11 on software. The distinction between software that "provides information used to take decisions with diagnosis or therapeutic purposes" and software whose output is merely informational depends on a careful reading of the intended purpose, the user interface, and the downstream clinical workflow. MDCG 2019-11 Rev.1 (June 2025) is the reference for this reading.
  • Rule 10 on active diagnostic and therapeutic devices. Whether a device is active in the Annex VIII sense, whether it supplies energy directly to the patient, and whether it has the specific characteristics that trigger escalation are all questions that turn on a precise reading of both the intended purpose and the Annex VIII Section 1 definitions.
  • The measurement function question. Whether a Class I device has measurement function (and therefore becomes Im) is not always obvious. The Annex VIII Section 1 definition of measurement function is narrower than common English usage, and many devices that "measure" in a colloquial sense do not meet the regulatory definition.
  • Invasiveness and duration boundaries. Transient versus short-term versus long-term use. Body orifice versus surgically invasive. The boundaries are defined, but a given device can sit close to a boundary, and the reading of the intended purpose determines which side.

Where the rules give flexibility, they give it so that the classification matches the actual device, not so that manufacturers can pick the class they prefer. The flexibility is the Regulation's acknowledgement that devices are not always clean cases. The obligation is to apply the flexibility precisely.

The intended purpose lever

Intended purpose is defined in Article 2(12) of Regulation (EU) 2017/745 as "the use for which a device is intended according to the data supplied by the manufacturer on the label, in the instructions for use or in promotional or sales materials or statements and as specified by the manufacturer in the clinical evaluation" (Regulation (EU) 2017/745, Article 2, point 12).

Every word of that definition matters for minimising classification. The intended purpose is not what the founder wishes the device to do. It is what the label, the IFU, the promotional materials, and the clinical evaluation together say the device is for. If those four surfaces are not consistent, the intended purpose is not stable, and no classification argument built on it will survive a Notified Body review.

The Graz wellness pivot is the purest example of the intended purpose lever at work. A Graz-based team had spent most of a year building a Class IIa regulatory strategy around their product. The technology was competent, the QMS work was underway, and the Notified Body prep was in progress. When we pulled the intended purpose apart against the Article 2(1) definition of a medical device, it turned out the genuine use of the product was wellness, not medical. Nothing about the technology needed to change. What changed was the honest regulatory positioning: as actually used, the product did not meet the Article 2(1) criteria at all. A disciplined re-framing moved the product out of MDR scope entirely. Market entry became much faster. The lesson was not that every device can do this — most cannot — but that the Purpose Pass was where the discovery became possible.

The lever here is honesty, not creativity. The Graz team did not talk the product out of being a medical device. They looked carefully at what the product actually did and wrote an intended purpose that matched. The old intended purpose had been aspirational — the founders had assumed a medical framing because it sounded more serious. The new one was accurate.

The general rule: the most leveraged thing you can do to minimise classification, ethically and defensibly, is to write an intended purpose that is exactly as broad as the device genuinely needs, and no broader. Extra claims that sound good in a pitch deck but add nothing the customer uses are pure classification cost. Strip them, or accept the higher class they trigger.

The evidence-based argument

A defensible lower class requires a written argument, and the argument has a specific shape. Two to five pages. Structured around four components.

One — the intended purpose paragraph. One page, at most. The exact wording that will appear on the label, in the IFU, on the website, and in the clinical evaluation. Consistent across all four surfaces. Specific enough that a reviewer can read it once and understand what the device is for, and narrow enough that it does not invite escalation.

Two — the specific Annex VIII rule. Not "Rule 10" but "Annex VIII, Section 3, Rule 10, second indent." The specific sub-clause. The specific paragraph. If any implementing rules from Annex VIII Section 2 might apply, they are addressed here too.

Three — the mapping. Element by element, how each part of the rule maps to the intended purpose. If the rule contains an escalation trigger — "unless the device is intended to..." — the argument must explicitly address why the trigger does not apply. Leaving it to inference is how lower-class arguments fail.

Four — the guidance cross-check. MDCG 2021-24 (October 2021) for general classification, MDCG 2019-11 Rev.1 (June 2025) for software, and the Borderline Manual v4 (September 2025) for borderline cases. If the guidance supports your position, cite it. If the guidance is silent on your specific case, say so. If the guidance appears to contradict you, address it head-on.

Tibor's hardware-software defence is what this looks like in practice. A combined hardware-software device where the software side was unambiguously Class IIa and the hardware side was disputed. The question was whether the hardware carried measurement function in the Annex VIII sense, which would have pulled it into the Class I measuring (Im) sub-category and triggered additional conformity assessment obligations. The team went through the Annex VIII Section 1 definition of measurement function, then through what the hardware actually did, what quantitative output it produced, how that output was used downstream, and whether the intended purpose referred to measurement at all. The intended purpose did not. The raw signal was consumed by the Class IIa software, which already handled the clinically meaningful interpretation. The hardware on its own produced a raw signal, not a measurement in the Annex VIII sense. The Notified Body reviewer read the argument, matched it against Annex VIII and internal guidance, and accepted the Class I position without measurement function.

This is what defensible lower-class minimising looks like. Not rhetoric, not pushing, not hoping the reviewer will not check. A careful written argument that a careful reviewer can read and agree with.

When minimising fails patient safety

Some lower-class arguments should fail, and the framework has to be honest about which ones.

A lower class fails patient safety when the rule clearly escalates and the argument depends on ignoring the escalation trigger. If Rule 11 places the software at Class IIb because the information is used for decisions that could cause serious deterioration of a person's state of health, "we think it is IIa" is not an argument — it is a hope. The Regulation wrote the trigger for a reason. Overriding it would put patients downstream of a device at higher risk than the classification system assumes.

A lower class fails patient safety when the intended purpose has to be distorted to fit. If the device, as actually used, will make claims or support decisions that the intended purpose excludes, the gap between marketing reality and regulatory text is a trap. The marketing claims will eventually catch up — through a website review, a promotional audit, an incident report — and the classification will be exposed as built on a fiction.

A lower class fails patient safety when MDCG guidance explicitly places devices like yours higher. The guidance documents exist because the Regulation's drafters knew the rules needed interpretation, and the interpretation has authority. Arguing against MDCG 2019-11 Rev.1 on software classification, or against MDCG 2021-24 on general classification, without a stronger counter-argument rooted in the MDR text itself, is not minimising — it is gambling.

The test is: if a patient harmed by a device failure asked why the device was in the lower class, would the argument hold up? If the answer is "because the Annex VIII rule, read precisely against the intended purpose, placed it there" — hold the class. If the answer is "because we found a way to argue it down" — raise it.

Common mistakes

The failure modes cluster around a few patterns:

  • Copying a competitor's class without reading Annex VIII against your own intended purpose. Your device is not their device. Your intended purpose is not their intended purpose. The class is specific to the file, not to the category.
  • Starting the minimisation work after the product is built. The leverage is highest before architecture decisions are locked. Post-hoc minimising fights the physical product as well as the Regulation.
  • Treating minimising as negotiation with the Notified Body. The Notified Body is not your counterparty. Annex VIII is. The reviewer is reading the same Regulation you are.
  • Writing the intended purpose to fit the desired class. This reverses the chain. The intended purpose should reflect the device's actual use, and the class follows. Reverse-engineering the intended purpose from the class is how manufacturers end up with claims that cannot be supported in the market.
  • Ignoring escalation triggers. Most Annex VIII rules contain "unless" clauses or implementing rules that pull devices up. Skipping past these because they are inconvenient is the single most common reason lower-class arguments fail at Notified Body review.
  • Confusing "lowest defensible" with "lowest wished-for." Defensible means a reviewer will accept it. Wished-for means the founder hopes for it. Only the first is a strategy.

The Subtract to Ship angle

The Subtract to Ship framework for MDR compliance treats classification as the second of four passes, and the discipline of the pass is exactly this: find the lowest defensible class, then subtract every activity that would only be required at a higher one. Subtraction is only safe when the class is honest. A founder who minimises the class by distorting the intended purpose is not subtracting — they are building on a foundation that will collapse. A founder who minimises by reading Annex VIII precisely against a stable intended purpose is subtracting correctly, because the work that gets cut was never required in the first place.

The same framework tells you when to stop minimising. When the next lower class would require bending the intended purpose, ignoring an escalation trigger, or arguing against explicit MDCG guidance, the pass is done. Subtract to Ship cuts waste. It does not cut the requirements the Regulation actually imposes.

Reality Check — Is your class minimised or distorted?

  1. Is your current intended purpose consistent with what your label, IFU, website, and clinical evaluation say — or will say?
  2. Can you cite the specific Annex VIII rule and sub-clause that supports your target class, by number and paragraph?
  3. Have you explicitly addressed every escalation trigger in the applicable rule, rather than leaving them to inference?
  4. Have you checked MDCG 2021-24 and, if your device is software, MDCG 2019-11 Rev.1 — and does the guidance support your position or contradict it?
  5. If a patient harmed by a device failure asked why the device was in your chosen class, does the answer rest on the rule and the intended purpose, or on a creative argument?
  6. Would your written classification argument survive a careful reading by a Notified Body reviewer who reads Annex VIII every day?
  7. If the honest answer to the class question is one level higher than you want, are you prepared to accept the higher class — or will you keep pushing until something bends?

Frequently Asked Questions

What does it mean to minimise device classification under the MDR? It means finding the lowest class that is defensible against a specific Annex VIII rule, a stable intended purpose under Article 2(12), and the relevant MDCG guidance. Minimising is legitimate where Annex VIII allows interpretive judgment — on software rules, on measurement function, on active device sub-rules. It is not legitimate where the rule clearly places the device higher or where the intended purpose has to be distorted to support the lower class.

Is minimising classification the same as gaming the system? No. Gaming the system means pushing for a lower class by distorting the intended purpose, ignoring escalation triggers, or arguing against MDCG guidance without substance. Minimising correctly means applying Annex VIII precisely, with evidence, where the rules allow legitimate judgment. The difference is whether the argument holds up under a careful reading of the Regulation — if it does, the minimising is defensible; if it does not, it is a risk masquerading as a strategy.

How much class reduction is actually possible for a typical startup device? Usually less than founders hope, but more than consultants assume. The biggest moves come from the Purpose Pass — changing a device from medical to wellness by fixing an inaccurate intended purpose, or from Class IIa to Class I by removing claims the device did not actually need. Within a given intended purpose, moves between adjacent classes are possible when Annex VIII allows judgment and the argument is strong. Multi-class jumps within a stable intended purpose are rare and usually indicate the original classification was wrong.

What is the single biggest leverage point for minimising classification? The intended purpose. Article 2(12) defines it as the use stated across the label, the IFU, the promotional materials, and the clinical evaluation. Every word in that paragraph has classification consequences. Extra claims that sound good in a pitch deck but add nothing the customer uses are pure classification cost. Writing the intended purpose exactly as broad as the device needs — and no broader — is where most legitimate minimising lives.

What happens if my minimised class fails at Notified Body review? The reviewer disputes the class and you either produce a stronger written argument, accept the higher class, or escalate under Article 51(2) to the competent authority. In practice, a well-built minimisation argument rarely fails outright — it either holds or results in a partial win. Arguments that fail badly almost always fail because the intended purpose was distorted or an escalation trigger was ignored. The fix is to rebuild on an honest intended purpose and a precise rule reading.

Sources

  1. Regulation (EU) 2017/745 of the European Parliament and of the Council of 5 April 2017 on medical devices, Article 2(1) (definition of medical device), Article 2(12) (intended purpose), Article 51 (classification), Annex VIII (classification rules, definitions, implementing rules, Rules 1-22). Official Journal L 117, 5.5.2017.
  2. MDCG 2021-24 — Guidance on classification of medical devices, October 2021.
  3. MDCG 2019-11 Rev.1 — Guidance on Qualification and Classification of Software in Regulation (EU) 2017/745 — MDR and Regulation (EU) 2017/746 — IVDR, October 2019, Revision 1 June 2025.

This post is part of the Device Classification & Conformity Assessment series in the Subtract to Ship: MDR blog. Authored by Felix Lenhard and Tibor Zechmeister. Minimising classification is one of the highest-leverage decisions a startup makes — and one of the easiest to get wrong. When the rule reading is genuinely contested, or the intended purpose is genuinely unstable, a working session with a sparring partner who has built these arguments dozens of times is usually faster than another month of internal debate.