How to write an intended purpose for software as a medical device?

Writing an intended purpose for software as a medical device under the EU MDR

The intended purpose is the statement in which a manufacturer defines what its device is for. For software as a medical device it is the single most consequential sentence in the technical documentation, because it decides whether the software is a medical device at all, which class it falls into, which conformity assessment route applies and how much clinical evidence the manufacturer has to produce.

Startups often write it late, borrow wording from a pitch deck, or keep it broad in the hope of covering future features. Each of these choices tends to surface later as a classification dispute, a request for further clinical data or a gap between what the product claims and what it has been assessed to do.

In June 2025 the Medical Device Coordination Group revised its guidance on the qualification and classification of software, MDCG 2019-11, and placed particular weight on a clearly crafted intended purpose. Notified bodies specialised in software, Scarlet among them, have since written about the same point from the assessor side.

This article explains what the MDR requires, which elements a strong intended purpose for SaMD contains, how individual words move a product between risk classes, and how to keep the statement consistent across the documentation and the market.

Why the intended purpose decides the regulatory pathway

Article 2(12) of Regulation (EU) 2017/745 defines intended purpose 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.

Two consequences follow from that definition. The intended purpose is not only what the manufacturer writes in the technical file, since a claim on a website or in an app store listing forms part of it. And the clinical evaluation has to support the statement as written, so every element of the statement creates a corresponding need for evidence.

Qualification, classification, the general safety and performance requirements that apply, the clinical evaluation, the risk management file and the post market surveillance plan all start from this statement. A change to it later in development usually means reopening several of those documents at once.

What MDCG 2019-11 revision 1 changed

The revised guidance was published on 17 June 2025 on the European Commission MDCG guidance page. It did not change the decision logic for qualification or the structure of Rule 11, but it clarified how both should be applied.

The revision stresses that qualification and classification depend exclusively on the statement, irrespective of the platform on which the software runs or the way it is integrated and delivered. It also confirms that a low level of risk does not take software outside the MDR, so a product with a medical purpose is a device even when its potential for harm is small.

For modular products, the revision reinforces that each module with a medical purpose needs its own clearly defined intended purpose, and that the product as a whole must also have one. It adds examples, including software intended to treat, and clarifies the treatment of products listed in Annex XVI that have no medical purpose.

The elements of a strong intended purpose

Annex II of the MDR requires the technical documentation to describe the statement together with the intended patient population, the medical conditions to be diagnosed, treated or monitored, the indications, contraindications and warnings, and the intended users. For software, the following elements give the statement the precision a notified body expects.

Medical purpose and clinical condition

The statement names the medical purpose in the terms used by the MDR, such as diagnosis, monitoring, prediction or treatment, and the specific condition concerned. A statement that the software supports cardiovascular health says almost nothing. A statement that it estimates the ten year risk of a major cardiovascular event in adults without known cardiovascular disease can be classified, tested and evaluated.

Patient population and intended users

The population is described by age, condition, setting and any relevant exclusions, and the users by role and level of training. Whether the output is read by a specialist, a general practitioner or the patient affects the usability engineering under IEC 62366, the risk analysis and the clinical evidence. Our article on IEC 62366 usability engineering covers how user profiles feed into validation.

Role of the output in clinical decisions

The statement makes clear whether the output informs a decision, drives it, or performs a diagnostic or therapeutic function on its own. This element carries most of the weight in classification, and it is where vague verbs such as support, assist or help cause the most difficulty, since they do not tell the reader how much the clinician relies on the software.

Inputs, use environment and limitations

The data the software accepts, the conditions under which it is used and the situations in which it must not be used complete the statement. For AI devices this includes the input modalities and acquisition conditions on which the model was trained and validated, since performance outside those conditions has not been demonstrated.

How wording moves software between classes under Rule 11

Rule 11 of Annex VIII places software that provides information used to take decisions with diagnostic or therapeutic purposes in class IIa. The class rises to IIb where those decisions may cause a serious deterioration of health or a surgical intervention, and to III where they may cause death or an irreversible deterioration of health.

Software intended to monitor physiological processes is class IIa, and class IIb where it monitors vital physiological parameters and variations in them could result in immediate danger to the patient. All other software falls into class I.

Because the rule turns on the consequences of the decisions the software informs, the description of the clinical condition and of the role of the output largely determines the class. The same algorithm can be class IIa when it prioritises routine cases for review and class IIb or III when it is intended to rule out a time critical condition. A manufacturer that wants a particular class has to write an intended purpose that genuinely limits the device to that use, and then build and market the product accordingly.

Where the statement leaves the medical purpose itself in doubt, the question becomes one of qualification rather than class. Our article on whether an app is a mental health SaMD or a wellness product works through that boundary for one common category.

Keeping the intended purpose consistent across documents and markets

The same statement should appear, word for word or in a controlled derivative, in the instructions for use, the clinical evaluation plan and report, the risk management file, the software requirements, the usability engineering file and the post market surveillance plan. Inconsistencies between these documents are among the most frequent reasons a notified body raises questions on software submissions.

Marketing deserves the same control. Article 7 of the MDR prohibits claims that ascribe to a device functions or properties it does not have, or that create a false impression about its benefits. MDCG 2025-4 on making software available through online platforms adds that information shown in app stores has to match the statement of the device. A website that promises earlier diagnosis for a product certified to inform triage creates a compliance problem regardless of the technical file.

Startups that plan to enter the United States should also compare the EU statement with the indications for use they intend to propose to the FDA. The concepts differ in detail, but a single well structured description of purpose, population, users and role of the output can be adapted to both.

Intended purpose for AI and modular software

For a modular product, the manufacturer decides which modules have a medical purpose and writes an intended purpose for each of them, together with an overall statement for the product. Modules without a medical purpose remain outside the device boundary, provided the interfaces between them are documented and the medical modules do not depend on them in a way that affects safety.

For AI devices, the statement also defines the limits within which the model may later be updated. Retraining on additional data from the same population and the same inputs can remain within it, while extending the device to a new population, a new clinical condition or a stronger claim changes the statement and requires a new assessment. This is also the boundary within which a predetermined change control plan can operate, so writing the statement with the product roadmap in view saves work later.

Examples of intended purpose statements for SaMD

The examples below are simplified for illustration. They show how the same elements produce an intended purpose that a notified body can classify and assess, and how a weak version of each statement reads.

Clinical decision support

A weak version describes software that helps clinicians manage patients with heart failure. A stronger intended purpose states that the software estimates the risk of hospital readmission within 30 days for adult patients discharged after an admission for heart failure, using data from the hospital electronic health record, in order to inform follow up planning by cardiologists and heart failure nurses. It is not intended to replace clinical judgement or to be used for patients under 18 years of age.

Remote monitoring

A weak version describes an app that tracks symptoms. A stronger intended purpose states that the software collects symptoms and spirometry readings entered by adult patients with diagnosed asthma, and alerts the treating respiratory team when values fall outside thresholds set by the clinician. It does not monitor vital physiological parameters in real time and is not intended for emergency use.

Digital therapeutic

A weak version promises better sleep. A stronger intended purpose states that the software delivers a structured programme of cognitive behavioural therapy for insomnia to adults diagnosed with chronic insomnia, as an adjunct to care by a physician, and that it is not intended for patients with untreated sleep apnoea or bipolar disorder.

In each case the stronger intended purpose names the medical purpose, the population, the users, the role of the output and the main limitations. That is what allows classification under Rule 11, the clinical evaluation and the risk analysis to rest on the same foundation.

Common weaknesses in SaMD intended purpose statements

  • Verbs that leave the role of the output open, such as support, assist, optimise or empower, without stating which decision the software informs.
  • A population described only by the condition, with no age range, setting or exclusions.
  • Users described as healthcare professionals in general, when the device has only been validated with a specific group of specialists.
  • Claims in the clinical evaluation, the website or investor materials that go beyond the statement in the instructions for use.
  • A single statement for a modular product, with no distinction between medical and non medical modules.
  • Inputs that are not described, so that the notified body cannot tell which data the performance figures apply to.

Steps for writing an intended purpose that holds up

A startup can approach the statement as a design input rather than a regulatory formality.

  • Draft the intended purpose before the architecture is fixed, and revisit it at each design review as features are added or removed.
  • Describe the medical purpose, condition, population, users, role of the output, inputs and limitations separately, then combine them into a statement of a few sentences.
  • Test the draft against Rule 11 and the decision trees in MDCG 2019-11, and record the classification rationale alongside it.
  • Check that each element can be supported with evidence that is available or can realistically be generated, and narrow the statement where it cannot.
  • Place the statement under document control and derive every other description of the device from it, including marketing and app store text.
  • Discuss the statement with the notified body at the pre application stage, since agreement on it early removes the most common source of later questions. Our overview of SaMD regulatory compliance describes how this fits into the wider pathway.
Writing an intended purpose for software as a medical device under the EU MDR

Frequently asked questions

What is the intended purpose of a medical device?

Under Article 2(12) MDR it is the use for which the device is intended according to the label, the instructions for use, promotional or sales materials and statements, and the clinical evaluation. It determines qualification, classification and the evidence the manufacturer has to provide.

What changed in MDCG 2019-11 revision 1?

The June 2025 revision kept the qualification logic and the structure of Rule 11, but stressed that qualification and classification depend only on the intended purpose, clarified the treatment of modular software and Annex XVI products, and added new examples.

Can a website or app store listing change the intended purpose?

Yes. The MDR definition includes promotional and sales materials, so claims made in marketing form part of the intended purpose. Article 7 MDR prohibits misleading claims, and MDCG 2025-4 requires app store information to match the intended purpose.

How does the intended purpose affect the class of SaMD?

Rule 11 classifies software by the consequences of the decisions its output informs. Describing the clinical condition and the role of the output determines whether the software is class I, IIa, IIb or III.

Does retraining an AI model change the intended purpose?

Not necessarily. Retraining on data from the same population and inputs can stay within the existing intended purpose, while a new population, condition or claim changes it and requires a new conformity assessment.

Define your intended purpose with Aura Health

Aura Health helps digital health and SaMD startups write intended purpose statements, document qualification and classification under Rule 11, and align the clinical evaluation, risk management and marketing claims with them before the first submission to a notified body.

To review the statement of your product and its classification, write to partnerships@aurahealth.ch.