MDR Rule 22: When a Diagnostic Function Forces Class III

Annex VIII of Regulation (EU) 2017/745 groups its active-device classification rules as Rules 9 through 13, and most classification training stops there. One more active-device rule sits further down, filed among the special rules: Rule 22. It reaches active therapeutic devices with a built-in diagnostic function — and where that function significantly determines the patient management the device delivers, Rule 22 classifies the device as Class III regardless of how the device would otherwise have been classified.

Where Rule 22 sits, and why it gets missed

MDR's Annex VIII organizes its twenty-two classification rules into four groups: non-invasive devices, invasive devices, active devices, and a set of special rules layered on top. The active-device group runs Rules 9 through 13 — which is where most classification exercises for powered devices stop looking. Rule 22 belongs to the special-rules group instead, introduced by the MDR and absent from the old Medical Device Directive entirely. A team that classifies a connected or software-driven device by working through Rules 9 to 13 and the relevant software classification rule, then stops, can reach a defensible-looking classification that never asked Rule 22's question at all.

What the rule actually targets

Rule 22 is narrower than its reputation. It only reaches devices that are already active therapeutic devices — a device built purely for diagnosis or monitoring, with no therapeutic function of its own, sits outside it entirely and is classified under the rules that already govern diagnostic and monitoring devices. What the rule adds is a test for devices that both treat and diagnose in a single system: where the diagnostic or sensing function doesn't just inform a clinician on the side, but significantly determines the therapeutic action the device itself delivers, the device is Class III.

• Closed-loop systems. A device that senses a physiological parameter and automatically adjusts its own therapeutic output in response — an automated insulin-dosing system is the paradigm case — is the kind of design the rule was written for.
• Automated external defibrillators. An AED's rhythm-analysis function doesn't just report a reading; it determines whether and how the device delivers therapy, which is exactly the dependency Rule 22 is testing for.
• Devices with an incidental diagnostic feature. A therapeutic device that also displays a diagnostic reading a clinician separately acts on, without that reading driving the device's own therapeutic function, sits in a genuinely harder zone the rule's text doesn't resolve on its own.
• The connected-device overlap. Teams building connected or AI/ML-enabled closed-loop products are the ones most likely to have a Rule 22 question and least likely to have asked it, because the classification work tends to stop at the software rule.

The hard part of Rule 22 isn't reading the rule. It's building a clinical case for where your device's diagnostic function sits on a spectrum the regulation never put numbers on. Why the classification rationale carries the weight

An undefined threshold is still a threshold

"Significantly determines the patient management" is not a defined term in Regulation (EU) 2017/745, and the European Commission's MDCG 2021-24 classification guidance is the reference point for working through the broader rule set, but it does not settle every edge case Rule 22 raises on its own. That ambiguity has drawn public pushback from manufacturers who argue the rule's plain wording could sweep in devices most clinicians would not instinctively think of as combined diagnostic-therapeutic systems. Whatever side of that debate a given device falls on, the practical consequence is the same: a Notified Body reviewing a borderline Rule 22 case is going to ask for the manufacturer's reasoning, not just its conclusion, and a classification file that doesn't document the diagnostic function's actual role in the therapy is the file that gets questioned late, during conformity assessment rather than during design.

None of this is exotic once a team knows to ask the question. It is a bounded piece of classification work: confirm the device is an active therapeutic device, map every diagnostic function it carries, and build the clinical argument for where that function sits relative to the device's own therapeutic action — in writing, before a Notified Body does it for you. Teams that fold Rule 22 into the same classification exercise that already covers borderline and mode-of-action questions are the ones who reach conformity assessment with a classification nobody has to relitigate.