Clinical Evaluation Versus Clinical Investigation, and Why Medical Software Needs Both

A clinical evaluation gathers the evidence that already bears on a device and judges whether it supports the claim being made. A clinical investigation runs a fresh study to produce new evidence when what exists falls short. One appraises, the other generates. The two are easy to confuse, since both carry the word clinical and both end in a written report, yet they answer different questions, and a serious medical product usually needs both.

Key points#

• A clinical evaluation reads and weighs the evidence that already exists; it does not run an experiment.
• A clinical investigation is a designed study, the medical-device counterpart of a clinical trial, run to fill a gap the evaluation has identified.
• The evaluation comes first and defines the question; the investigation is one possible answer to it.
• Under EU MDR, the evaluation is a living file that is updated as real-world data arrives, not a box ticked once at launch.
• Software raises the stakes: models can perform well on their training data and stumble on populations they never met, so the evidence has to be measured rather than asserted.

Start with the claim, not the code#

Before either process makes sense, it helps to name the thing they both serve: the claim. A medical product asserts that it does something useful and safe for a defined purpose. Everything downstream, the paperwork, the studies, the regulatory pathway, exists to test whether that assertion is true. Clinical evaluation and clinical investigation are simply the two ways of putting a claim under pressure. The first reads the record; the second adds to it.

Keeping the claim in view also explains why teams cannot regulate their way out of the work by calling a product a wellness app. If software is intended to inform a diagnosis or a treatment decision, it is a medical device, and the same evidence logic follows the claim wherever it lands.

The evaluation: a structured argument about existing evidence#

A clinical evaluation is best understood as a case built from sources already on hand. You state what the device is for, then assemble every relevant piece of evidence and weigh whether it supports that purpose. The material includes published studies on the device itself, studies on genuinely equivalent devices, data from earlier versions, and findings collected once the product is in use.

The operative verb is appraise. No experiment is being run. Someone is reading what exists, deciding how much each piece is worth, and writing down plainly where the evidence holds and where it stays thin. A credible evaluation is candid about its own gaps rather than papering over them, because the gaps are what tell you whether a new study is needed.

Under EU MDR Article 61, this is neither optional nor finished at launch. The evaluation is a living file, refreshed as new data arrives, because a device that looked well supported on its first day can drift out of alignment with reality as it meets new populations and new uses.

The investigation: making evidence that does not yet exist#

A clinical investigation is a study designed and run precisely because the evidence needed cannot be found in the literature. It is the device world's version of a clinical trial. A protocol is written, the participants and the measurements are defined in advance, ethics approval is obtained, people are enrolled, and data is collected under rules fixed before anyone looks at a result.

The purpose is to create one fresh, trustworthy piece of evidence. When an evaluation runs into a question the published record cannot answer, an investigation is how that question gets answered.

This is also where the heaviest obligations sit, because the study involves real people. Protocol design, informed consent, participant safety, and pre-specified outcomes stop being good practice and become formal duties. The credibility of any finding rests on the discipline of the method that produced it, which is why these duties are written down and audited.

How the two lock together#

Picture the evaluation as the question and the investigation as one possible answer. The evaluation asks whether the current evidence supports the claim. If the honest answer is yes, no new study may be needed. If the answer is no, the evaluation names the exact gap a study would have to close.

That ordering protects both people and resources. A study on human participants is a serious undertaking, so one should be run only when reading the existing evidence shows it is genuinely necessary. The evaluation justifies the investigation, and the investigation feeds its result back into the evaluation. The loop does not stop at launch: once a device is in use, real-world data flows back in, and sometimes that data raises a new question that prompts another study. A mature product treats this as a cycle, not a hurdle cleared once.

Why software makes this harder, not easier#

Software tempts teams to skip the slow part. Code feels fast and revisable, so it is tempting to assume the evidence can be revisable too. Regulation does not see it that way, and for good reason.

A machine-learning model sharpens the problem. Its performance is a claim about a number, and a number has to be measured, not asserted. A model can perform beautifully on the data it was built from and then falter on a group it never encountered during training. A clinical evaluation of such a system has to ask which populations the evidence actually covers, not which populations the marketing implies. This is where general clinical judgment matters as much as statistics: a tool that suggests how to manage a chronic condition is only as trustworthy as its performance in the patients who will actually use it.

Software also changes between versions, sometimes without any announcement. That drift is exactly why post-market data matters so much for digital products. The evaluation has to keep asking whether yesterday's evidence still describes today's behavior, and an investigation may be the only honest way to confirm that it does.

The question worth carrying#

For anyone building a medical product, the most useful early question is not how to avoid being regulated. It is what the product is claiming, and what evidence would honestly support that claim. That single question splits cleanly into the two processes above.

Ask first what the existing evidence already shows. That is the clinical evaluation, and it may reveal a claim larger than the evidence can carry. Then ask what new evidence would be needed to close the gap. That is the clinical investigation, and it is how a claim earns the right to be believed.