MDCG 2022-2: Clinical Evidence for IVDs Under the IVDR

Introduction

MDCG 2022-2 is the document that tells you what the IVDR means by clinical evidence. The Regulation itself spreads the concept across Article 56, Annex XIII and Annex I, and never assembles it in one place. The guidance does that assembly, and in doing so it settles several questions that manufacturers had been answering differently from one another.

It is not a long document — thirty-one pages including two appendices — and most of the commentary written about it summarises the purpose section and stops. That is a missed opportunity, because the useful material is further in: the eight elements an intended purpose statement has to contain, the eighteen factors that shape a performance evaluation plan, the chain of events that turns a wrong result into patient harm, and a table in Appendix II that tells you how often each document has to be updated for each device class.

This guide walks the guidance in the order the work actually happens. It assumes you are familiar with the structure of the Regulation; if you are not, our guide to what Regulation (EU) 2017/746 requires covers qualification, classification and the conformity assessment routes first.

Table of Contents

• What MDCG 2022-2 is, and what it deliberately leaves out
• The three pillars that make up clinical evidence
• The intended purpose statement drives everything else
• The performance evaluation process — four phases
• Which documents the process produces, and where each one sits
• Risk management: the hazard is the result, not the malfunction
• The Performance Evaluation Plan and its eighteen inputs
• Scientific validity — and the literature search that supports it
• Analytical performance and its indicators
• Clinical performance, and when it is not required
• Clinical performance studies and ISO 20916
• The Performance Evaluation Report and the appraisal questions
• Continuous update: PMS, PMPF and Appendix II
• Where MDCG 2022-2 stops and the other documents begin
• What Notified Bodies raise most often
• Frequently asked questions
• Conclusions

What MDCG 2022-2 is, and what it deliberately leaves out

MDCG 2022-2 was endorsed by the Medical Device Coordination Group in January 2022. It is guidance: not legally binding, expressly not a Commission document, and only the Court of Justice can give binding interpretations of Union law. In practice this distinction matters less than it sounds, because Notified Bodies assess against it and manufacturers who depart from it are asked to justify the departure.

It applies to every product meeting the IVD definition in Article 2(2), and to accessories, since accessories fall within the scope of the Regulation. Specimen receptacles are included. Four categories are outside it, because Article 1(3) puts them outside the Regulation: general laboratory or research-use-only products, unless the manufacturer specifically intends them for in vitro diagnostic examination; invasive sampling products and products applied directly to the body to obtain a specimen; internationally certified reference materials; and materials used in external quality assessment schemes.

Three exclusions are more important than they look, because manufacturers regularly go to MDCG 2022-2 expecting to find them covered:

The equivalence point is the one that catches people. The guidance permits data from equivalent devices and requires a justification, but it gives no criteria for what makes two IVDs equivalent. That gap is yours to fill in the quality management system, and a reviewer will ask to see the procedure that fills it.

The three pillars that make up clinical evidence

Clinical evidence for an IVD is what comes out of the performance evaluation. The performance evaluation is the assessment and analysis of data establishing three things, and the Regulation treats them as separate demonstrations rather than as one body of proof.

The distinction that gets lost most often is the first one. Scientific validity is a statement about the analyte, not about your product. If the association between the marker and the condition is well established in the literature, the demonstration can rest on published sources — peer-reviewed papers, textbooks, historical data, expert consensus positions. Your device does not enter into it. Analytical and clinical performance are the opposite: they are statements about your device specifically, and published literature about other devices supports them only indirectly.

The intended purpose statement drives everything else

Article 56(1) requires the manufacturer to specify and justify the level of clinical evidence in view of the device characteristics and its intended purpose. MDCG 2022-2 draws the obvious consequence: the intended purpose statement is the key driver of the whole assessment, and a vague one makes the rest of the evaluation unassessable.

Annex I, section 20.4.1(c) lists what the statement has to contain. Eight elements, all of which should appear in the instructions for use:

The guidance adds a point that is easy to miss and expensive to get wrong. Where an analyte is relevant across several conditions rather than diagnostic for one, the intended purpose should be framed at the level the device actually serves — as a marker of inflammation, for instance, rather than as a test for the specific causes of inflammation. The broader clinical contexts then belong in the scientific validity section and in the explanatory parts of the IFU, not in the intended purpose statement. Framing the purpose too narrowly creates evidence obligations you did not intend to take on; framing it too broadly creates obligations you cannot meet.

The performance evaluation process — four phases

Article 56 and Annex XIII Part A set out the process. MDCG 2022-2 summarises it in four phases, and the value of the summary is that it makes the sequence explicit — several of the failures Notified Bodies find are sequencing failures rather than content failures.

Which documents the process produces, and where each one sits

This is the part manufacturers most often get structurally wrong, because the Regulation names the documents in different places and MDCG 2022-2 does not tabulate them. The set is as follows.

Risk management: the hazard is the result, not the malfunction

Section 6.2 of the guidance is the part that manufacturers coming from the medical device side find most unfamiliar, and it is the most useful section in the document.

The observation it starts from is that in the majority of cases a deficiency in an IVD does not directly injure anyone. The device does not deliver energy, does not enter the body, does not fail mechanically in a way that harms the patient. Where harm occurs, it is indirect: it arrives through a clinical decision taken, or not taken, on the strength of an incorrect, delayed, absent or unclear result.

This changes what a hazard is. In an IVD risk file the hazard is not a component failure. It is the wrong answer.

Two practical consequences follow, and both show up in review.

The first is that the risk analysis has to reflect the clinical use and the medical purpose, not only the technical failure modes. A hazard analysis built entirely from component and process failures has skipped the half of the chain where the severity is decided. The same defect — a degraded reagent — carries a different severity depending on whether the result feeds an asymptomatic screening decision or an emergency one.

The second is that the probability of harm has two components, and the second is not in your hands: the probability that the hazardous situation occurs, and the probability that it goes on to cause harm given clinical practice. A result questioned against the patient's signs and symptoms does not cause harm; the same result accepted at face value does. You still have to estimate that second probability and record the basis for the estimate.

The guidance also reminds manufacturers not to forget the direct risks that do exist — chemical, electrical, electromagnetic and mechanical exposure of users and maintenance personnel. These are a smaller part of an IVD risk file, but they are the part most often missing entirely. The process itself is unchanged from the medical device side; our guide to ISO 14971 risk management covers the mechanics that carry across.

The Performance Evaluation Plan and its eighteen inputs

The guidance lists the factors a manufacturer should take into account when designing the PEP. The list is explicitly non-exhaustive, but it is the closest thing available to a checklist, and a PEP that visibly addresses each one is difficult to fault on scope.

Two requirements sit inside the PEP that are frequently placed elsewhere or omitted. The applicable GSPRs have to be identified and specified within the plan itself — not only in the GSPR checklist. And the benefit-risk assessment belongs in the PEP, with pre-determined thresholds and criteria, because the point of setting thresholds in advance is that they were set before the data arrived.

A third is easy to miss: the literature search protocol is part of the PEP. It is not a standalone deliverable, and its results are embedded in the relevant reports rather than in a document of their own.

Scientific validity — and the literature search that supports it

Scientific validity has to be demonstrated and documented for every device. Where the association between analyte and condition is well established, existing sources will carry it: appraised literature, peer-reviewed data, published clinical data including Summaries of Safety and Performance and authority registries, information on the scientific validity of devices measuring the same analyte, proof-of-concept studies, and consensus positions from professional associations.

Where the evidence does not exist, is insufficient, or the device is novel — a new analyte, a new intended purpose — a scientific rationale and new data are required. The guidance asks for a gap analysis to determine what additional evidence is needed, and that gap analysis is a document a reviewer will look for.

On the literature search itself, MDCG 2022-2 sets three expectations, and they are stricter than the practice they replaced:

1. Define the search protocol before retrieval begins. The strategy must be thorough and objective, meaning it has to be capable of surfacing unfavourable data as well as favourable data. A protocol that only finds supportive papers is not an objective protocol.
2. Run several searches with consolidated and relevant criteria, rather than one broad search.
3. Document so the work can be criticised. The methods have to be appraisable, the results verifiable, and the searches reproducible.

Reproducibility is the operative word. A search that cannot be re-run — because the database, the date, the string or the filters were not recorded — fails the third expectation regardless of how good the conclusions are. This is the most common single deficiency in scientific validity documentation.

Two structural points close the section. For devices intended to be used together — a reagent with its calibrators and controls — it may be more appropriate to establish scientific validity for the combination rather than for each component. And an instrument with an independent measuring function that uses no additional reagents still requires scientific validity unless a justification is given.

Analytical performance and its indicators

Analytical performance is the evidence that the device reliably, accurately and consistently measures or detects the analyte. The general rule is that it is demonstrated by analytical performance studies. Published experience from routine diagnostic testing can support the demonstration but does not replace it.

The indicators are largely common across IVDs, and the applicable set depends on the assay technology and the use environment. Annex I section 9.1 and Annex II section 6.1 list them; not all apply to every device, and MDCG 2022-2 is explicit that omissions are acceptable provided they are outlined and justified.

The specimen requirements deserve separate attention because they are where the file and the IFU most often diverge. Every specimen type and every sampling condition named in the intended purpose has to be assessed and demonstrated. Where relevant, that extends to the collection devices, the storage conditions and the transport conditions. The analytical performance report has to state the timeframe between collection, storage and analysis — duration, temperature limits, permitted freeze-thaw cycles — and this matters most for assays with time-critical analysis.

Clinical performance, and when it is not required

Clinical performance is the ability of the device to produce results correlated with a clinical condition or a physiological or pathological state, in the target population and with the intended user. The demonstration has to cover the intended uses, the target populations, the use conditions, the operating and use environments, and all the intended user groups.

The characteristics used to express it depend heavily on the intended purpose and the claims. The guidance lists diagnostic sensitivity and specificity, positive and negative predictive value, positive and negative likelihood ratios, odds ratio, number needed to treat or diagnose, number needed to harm or misdiagnose, and usability or user interface performance. Others may be added where the manufacturer judges them applicable.

Statistical relevance is required, not optional: confidence intervals, and an interpretation of what the robustness of the result means for the intended purpose. A point estimate without an interval is not a demonstration.

There are two carve-outs worth knowing:

• Some devices do not need clinical performance data at all. Article 2(39) makes clinical performance part of device performance only where applicable. Non-sterile specimen receptacles, microscopy glass slides and some general reagents are the examples given. Where the omission is duly justified, a clinical performance report is not expected — but the rest of the performance evaluation still is.
• Multi-indication markers do not need a study per indication. Where a marker is not specific to one condition but is adequately established by scientific validity across several clinical settings, separate clinical studies for each setting are not expected.

Clinical performance studies and ISO 20916

A clinical performance study becomes necessary when compliance with the relevant GSPRs on clinical performance cannot be established from scientific validity, analytical performance studies, literature, routine diagnostic testing experience or other performance studies. It is the last resort in a defined order, not the default.

When one is needed, MDCG 2022-2 points to a specific standard: studies should be conducted in line with well-established international guidance, and it names ISO 20916 on clinical performance studies using specimens from human subjects — regardless of the classification of the device, unless a justification is given under Annex XIII 1.2.3. Manufacturers of Class A and B devices sometimes assume the standard applies only higher up the classification scale. It does not.

Two further points carry practical weight:

The Clinical Performance Study Plan has to define the rationale, objectives, design, proposed analysis, methodology, monitoring, conduct and record-keeping, under Annex XIII 2.3.2. Where personal data is processed, the plan is also where the manufacturer states how the General Data Protection Regulation requirements are met within the data management process.

Studies conducted under the old In Vitro Diagnostic Directive do not meet the requirements of Annex XIII 2.3, and are therefore treated as other sources of clinical performance data under Annex XIII 1.2.3 — not as clinical performance studies. Their quality and completeness have to be assessed for gaps, and they need support from literature or from published routine-testing experience. This is one of the most consequential sentences in the guidance for any manufacturer transitioning a legacy portfolio.

Certain studies carry additional obligations — those requiring surgically invasive sample-taking solely for the study, interventional clinical performance studies, and studies involving additional invasive procedures or other risks for subjects. What those obligations are, and which route each study falls into, is the subject of a separate guidance: see our guide to MDCG 2025-5 and Article 57.

The Performance Evaluation Report and the appraisal questions

The PER is where the manufacturer compiles the evidence, determines the benefit-risk and documents the conclusion. It has to assess all relevant scientific validity, analytical and clinical performance data against the applicable GSPRs of Annex I. Studies other than clinical performance studies — analytical performance studies, PMPF studies — are documented by analogy and reflected in the PER.

Six considerations frame the assessment: the intended users; the state of the art; the nature, severity and evolution of the condition being diagnosed or treated; the adequacy of the risk estimation for each identified hazard; the number and severity of adverse events; and the availability of alternative diagnostic devices and the current standard of care.

Section 6.8 then offers a list of questions to guide the assessment. Treated as a self-check before the file goes to the Notified Body, they are the most useful page in the document. Condensed:

Continuous update: PMS, PMPF and Appendix II

Performance evaluation is a continuous process. The obligation is not to keep the file current in a general sense but to run a defined process that detects when it stops being current: monitoring the state of the art, complaints, PMPF data, direct user feedback, newly published research, new guidance, new harmonised standards.

The guidance is specific about one trigger that manufacturers underestimate. Changes to the intended purpose, the product design, the characteristics or the technology require the clinical evidence to be updated or re-established before the change is implemented, not after it. A change control process that treats the performance evaluation as a downstream document has the sequence wrong.

For assays where the target can shift — the emergence of mutated strains not detected by a specific assay is the example given — PMS planning is what detects the shift. PMPF is required to confirm that safety, performance and clinical evidence hold across the expected lifetime and that the benefit-risk ratio remains acceptable. It can be justified as not required in some cases under Annex III 1(b) and Annex XIII Part B 8, and where it is, the justification belongs in the PER.

Appendix II sets the minimum update frequencies. It is the one page of the guidance that translates directly into a documentation calendar.

Where MDCG 2022-2 stops and the other documents begin

MDCG 2022-2 is the general principles document. It is frequently read as though it were the complete answer on IVD clinical evidence, and it is not. Four documents divide the territory:

For IVD software specifically, MDCG 2022-2 should be read alongside MDCG 2020-1 on performance evaluation of medical device software, which covers how clinical evidence is established for software.

Where all of this lands physically is Annex II Section 6 of the technical file, with the post-market documentation in Annex III alongside — our guide to IVDR technical documentation section by section covers the placement.

What Notified Bodies raise most often

Across IVDR performance evaluation files, a small number of deficiencies account for most of the findings. All of them are visible against MDCG 2022-2 itself.

The intended purpose does not contain all eight elements. Usually the missing ones are the testing population and the explicit statement of whether the device is qualitative, semi-quantitative or quantitative. Because everything downstream is assessed against the intended purpose, this finding tends to bring others with it.

Scientific validity is asserted rather than demonstrated. A handful of references, no search protocol, no appraisal criteria, no gap analysis. The association may well be established; the file has not shown that the manufacturer established it.

The literature search cannot be reproduced. No databases, no dates, no strings, no inclusion and exclusion criteria. This fails the third of the three expectations in section 6.4 directly.

Analytical performance omissions are silent. An indicator that does not apply is fine; an indicator that is simply absent is a finding. The guidance requires omissions to be outlined and justified, which means the file has to say what was considered and why it was set aside.

The specimen types claimed exceed the specimen types tested. The IFU names a specimen type the analytical performance report never assessed. This is one of the fastest findings to raise and one of the most expensive to close, because it usually requires new data.

The risk file has no clinical dimension. Hazards expressed as component failures, severity assigned without reference to what the result is used for, and no assessment of the probability that an incorrect result leads to harm.

The PER collates instead of assessing. The three reports are summarised, the conclusion states that the device conforms, and no weighing of the data against the benefit-risk criteria appears anywhere in between.

Legacy IVDD study data is presented as clinical performance study data. It is other sources of clinical performance data, it requires a gap assessment, and it needs supporting literature or routine-testing experience.

Frequently asked questions

What is MDCG 2022-2?

MDCG 2022-2 is the Medical Device Coordination Group guidance on general principles of clinical evidence for in vitro diagnostic medical devices, endorsed in January 2022. It explains what clinical evidence means under Regulation (EU) 2017/746, how the performance evaluation process works, what the Performance Evaluation Plan and Report have to contain, how risk management connects to the evaluation, and how the evidence is maintained across the device lifecycle.

Is MDCG 2022-2 legally binding?

No. It is not a European Commission document and the views in it are not legally binding — only the Court of Justice of the European Union can give binding interpretations of Union law. In practice Notified Bodies assess performance evaluation files against it, so a departure from the guidance needs a documented justification even though the guidance itself imposes no legal obligation.

What are the three pillars of clinical evidence for an IVD?

Scientific validity, analytical performance and clinical performance. Scientific validity is the association of the analyte with a clinical condition or physiological state, and is a statement about the analyte rather than about the device. Analytical performance is the device's ability to detect or measure the analyte correctly. Clinical performance is its ability to produce results correlated with the condition in the target population and with the intended user.

Does every IVD need a clinical performance report?

No. Article 2(39) makes clinical performance part of device performance only where applicable, and MDCG 2022-2 gives non-sterile specimen receptacles, microscopy glass slides and some general reagents as examples where clinical performance data may not be expected. Where the omission is duly justified, no clinical performance report is required — but the rest of the performance evaluation, including the PER, still is.

How often must the Performance Evaluation Report be updated?

For Class C and D devices, as necessary and at least annually. For Class A and B, continuously on the basis of post-market surveillance and PMPF input, at the frequency defined in the PMS plan rather than on a fixed calendar. The frequencies are set out in Appendix II of the guidance.

Does ISO 20916 apply to low-class IVDs?

Yes. MDCG 2022-2 states that clinical performance studies should follow ISO 20916 regardless of the classification of the device, unless a justification is provided in accordance with Annex XIII 1.2.3. The standard is not reserved for Class C and D.

Can data from clinical performance studies conducted under the IVDD be used?

Yes, but not as clinical performance study data. Studies conducted under the Directive do not meet the requirements of Annex XIII section 2.3, so they count as other sources of clinical performance data under Annex XIII 1.2.3. Their quality and completeness have to be assessed to identify gaps, and they need support from literature or from published experience gained in routine diagnostic testing.

Where does risk management fit into the performance evaluation?

Alongside it, continuously. The risk analysis has to reflect the clinical use and medical purpose of the device, because harm from an IVD is usually indirect — it reaches the patient through a clinical decision based on an incorrect, delayed, absent or unclear result. When new risks are identified or changes are made, the impact on the clinical evidence required has to be considered, and the evidence updated or re-established before the change is implemented.

Conclusions

MDCG 2022-2 does not add requirements to the IVDR. What it does is make the existing ones legible: it collects a concept the Regulation scattered across Article 56, Annex I and Annex XIII, and states the sequence in which the work has to happen.

The sequence is where most files fail. A plan written after the data, a search protocol invented at the end, thresholds set once the results were known, a PER that collects rather than concludes — none of these are content failures, and all of them are visible from the dates on the documents. The guidance is unusually explicit that planning precedes generation and that the evidence is re-established before a change is implemented, not after.

The second thing to take from it is the risk section. For an IVD the hazard is the wrong answer, and the severity is decided by what someone does with it. A risk file built only from technical failure modes has analysed half the chain, and it is the half where nobody gets hurt.

If you are building the performance evaluation documentation from scratch, the EU IVDR Kit covers the Annex II and Annex III set with the performance evaluation documents structured to the Regulation's own order, and MDCG 2022-2 integrated where it applies.

Related articles

• IVDR Explained: What Regulation (EU) 2017/746 Requires
• IVDR Technical Documentation: Annex II Section by Section
• What MDCG 2025-5 Settles That the Regulation Left Open
• MDCG 2025-10: Post-Market Surveillance Under the IVDR
• ISO 20916: Good Study Practice for IVD Performance Studies
• ISO 14971 Risk Management for Medical Devices