Clinical and Performance Evaluation

Clinical evaluation is a lifecycle evidence process

Clinical and performance evaluation asks whether the available evidence supports the safety, performance and benefit claims for the intended population, users and conditions of use. It is not simply a literature review and it does not begin when engineering finishes.

Clinical questions influence requirements, risk controls, verification, usability work, labelling and post-market plans. Engineering evidence also supports the evaluation by showing that the configuration used to generate clinical data represents the product being placed on the market.

Keep the scope aligned with MTL-102 — Intended Purpose, Users and Use Environments.

Translate intended purpose into evidence questions

Every important claim should create a question that can be answered with objective evidence. Broad claims usually require broader populations, comparators, endpoints and follow-up than narrowly defined claims.

• Separate intended purpose from promotional statements and secondary claims.
• Identify which claims are clinical, analytical, technical, usability-related or economic.
• Define claimed benefits in terms meaningful to patients or clinical practice.
• Identify safety outcomes, foreseeable adverse events and residual risks.
• Specify relevant subgroups, contraindications and limitations.
• Trace each question to the evidence source and final conclusion.

Convert the resulting needs into controlled requirements through MTL-103 — User Needs and Design Inputs.

Medical devices and IVDs use related but different models

An IVD performance evaluation integrates scientific validity, analytical performance and clinical performance. These strands answer different questions and should not be collapsed into one generic “validation” statement.

Plan the evaluation before choosing evidence

The plan should define the questions, methods and decision rules before the team knows which answer is easiest to obtain. It should remain proportionate to the device, novelty, claims, risks and maturity of available knowledge.

• Define device scope, variants, accessories, software versions and intended markets.
• List clinical-safety, performance and benefit claims to be evaluated.
• Describe the state of the art, alternatives and relevant clinical practice.
• Define search methods, data sources, inclusion criteria and appraisal rules.
• Identify endpoints, follow-up, subgroups and acceptance criteria.
• Explain how equivalence or similar-device data will be used, if at all.
• Define gaps and the criteria for deciding whether new data are necessary.
• Plan PMCF or PMPF and periodic updating.
• Assign qualified authors, reviewers and medical or scientific expertise.

The evaluation plan should be part of the connected evidence structure described in MTL-104 — Design Controls and Technical Documentation.

Use multiple evidence sources deliberately

Each source has strengths, limitations and possible bias. A database search may miss unpublished adverse information; a company study may be well controlled but narrow; real-world data may be broad but confounded; bench data may explain mechanism without proving clinical benefit.

Appraise relevance and quality separately

A rigorous study may still be irrelevant to the claimed device, population or outcome. A highly relevant observation may be too weak to support a decisive conclusion on its own.

Record both favourable and unfavourable evidence. Excluding inconvenient data without a predefined, scientifically defensible reason undermines the evaluation.

Define the state of the art and clinical context

The state of the art describes current knowledge, accepted clinical practice, available alternatives, relevant outcomes and known risks. It provides the context for judging whether the device’s benefits, performance and residual risks are acceptable.

• Describe the condition, pathway, users and unmet need.
• Identify relevant guidelines, consensus, technologies and treatment alternatives.
• Characterise expected outcomes and complications under current practice.
• Explain whether endpoints and comparators remain clinically relevant.
• Identify changes in practice, epidemiology, technology or diagnostic criteria.
• Use the same context when defining benefits and evaluating residual risks.

The comparison is not necessarily a claim that the device is superior. The purpose is to understand what safe and effective performance means in the current clinical setting.

Treat equivalence as a demanding evidence argument

Data from another device may contribute only when differences are understood and do not invalidate transfer of the clinical conclusions. Similarity can help identify hazards or design questions without establishing equivalence.

A predicate, competitor or previous-generation product is not automatically equivalent. Document why every relevant difference does or does not affect safety, clinical performance or benefit-risk.

Use gap analysis to decide what evidence must be generated

A gap exists when the available evidence cannot answer an important question with sufficient relevance and quality. The response should address the specific uncertainty rather than default to the largest possible study.

• Map each claim, GSPR, risk and residual uncertainty to available evidence.
• Separate absence of evidence from evidence of unacceptable performance.
• Identify whether the gap is technical, analytical, usability, clinical or post-market.
• Consider whether bench, modelling, usability, literature or existing clinical data can answer the question.
• Explain why a new clinical investigation or performance study is necessary when people or specimens are involved.
• Prioritise gaps that affect patient protection, benefit-risk or major claims.
• Agree regulatory strategy early where evidence expectations are uncertain.

Use MTL-105 — Medical-device Risk Management to connect residual uncertainty to risk decisions rather than treating every gap as equally important.

Clinical investigations must be ethical, necessary and scientifically sound

A clinical investigation involving human subjects should answer a question that cannot be resolved adequately by less burdensome evidence. Subject rights, safety and well-being take priority over commercial or project objectives.

• Define the objective, hypothesis or estimand and the decision the result will support.
• Select a population, comparator, endpoints and follow-up appropriate to the intended claim.
• Identify device- and procedure-related risks and establish monitoring and stopping rules.
• Obtain applicable regulatory and ethics approvals before starting.
• Use an investigational configuration that is sufficiently controlled and representative.
• Protect informed consent, privacy, data integrity and vulnerable populations.
• Control investigators, sites, training, deviations, adverse events and device accountability.
• Preserve the protocol, statistical plan, dataset, analysis, report and disclosure obligations.

ISO 14155:2026 provides the current international good-clinical-practice framework for medical-device investigations. National and regional requirements still determine approvals and legal obligations.

IVD performance studies need specimen and clinical-context control

IVD evidence depends on the relationship between specimen, intended user, method, clinical condition, reference standard and statistical analysis. Sample numbers alone do not make a dataset representative.

• Define intended population, specimen type, collection, handling, storage and inclusion criteria.
• Use an appropriate reference method or clinical truth definition.
• Represent relevant disease stages, prevalence, interferents and confounding conditions.
• Control reader, site, instrument, reagent-lot and operator effects where applicable.
• Predefine invalid, indeterminate, missing and discordant-result handling.
• Separate analytical-performance questions from clinical-performance questions.
• Protect subjects and specimen donors and obtain the required approvals and consent.
• Ensure the tested system represents instruments, software, reagents, calibrators and interpretation rules to be marketed.

ISO 20916:2019 remains current for good study practice in IVD clinical performance studies using specimens from human subjects.

Endpoints and analysis must support the intended conclusion

Endpoints should be clinically meaningful, measured consistently and linked to the claim. Surrogate, composite or technical endpoints require justification when they stand in for patient-relevant benefit.

Predefine sample-size assumptions, analysis populations, multiplicity, missing-data handling, sensitivity analyses and subgroup treatment. A statistically significant result may be clinically unimportant, while an underpowered inconclusive result does not demonstrate equivalence or absence of harm.

Identify the device configuration behind every result

Clinical evidence cannot support a product that is materially different from the device investigated unless the impact of the differences is assessed. Hardware, software, algorithms, materials, accessories, workflow and labelling can all affect transferability.

• Record model, variant, serial or lot, software and algorithm versions.
• Identify accessories, consumables, reagents, calibrators and connected systems.
• Control training, instructions, procedure and clinical workflow.
• Document changes during a study and determine whether approval or protocol updates are required.
• Assess whether later design changes affect the mechanism, risk, endpoint or population.
• Use bridging evidence where appropriate and generate new data where uncertainty remains material.

Connect configuration and result traceability to MTL-106 — Verification and Validation.

The evaluation report must show the reasoning

A credible report lets an independent reviewer follow the questions, methods, evidence, limitations and conclusions. It should not hide the difficult parts behind a list of references.

The report should trace to the intended purpose, risk-management file, technical documentation, labelling and post-market plan. Use MTL-307 — EU MDR and IVDR General Safety and Performance Requirements for the wider conformity-evidence structure.

Post-market evidence maintains—not merely confirms—the conclusion

Post-market clinical follow-up for medical devices and post-market performance follow-up for IVDs should address residual questions, verify continued performance and identify emerging risks or changes in the state of the art.

• Define specific objectives rather than “collect more data.”
• Use complaints, vigilance, surveys, registries, literature, studies and real-world data appropriately.
• Track exposure or denominator information so event rates can be interpreted.
• Look for subgroup, use-environment and long-term outcomes not resolved pre-market.
• Compare observed performance with acceptance criteria and benefit-risk assumptions.
• Identify new alternatives, clinical practices, standards and scientific findings.
• Feed conclusions into risk, labelling, design, CAPA and regulatory reporting.
• Update the evaluation after significant changes or new evidence, not only on a calendar.

Clinical and performance evaluation across the lifecycle

Common misconceptions

Practical review checklist

• Are intended purpose, population, users, claims and benefits clearly defined?
• Does every important claim have a specific evidence question and acceptance basis?
• For an IVD, are scientific validity, analytical performance and clinical performance addressed separately?
• Does the evaluation plan define methods before evidence selection?
• Are favourable and unfavourable data identified and appraised transparently?
• Is the state of the art current and relevant to the claimed clinical context?
• Is any equivalence argument supported across technical, biological and clinical characteristics?
• Are gaps connected to risk and resolved with proportionate evidence?
• Are investigations or performance studies ethical, approved and scientifically sound?
• Do endpoints, follow-up and statistical methods support the intended conclusions?
• Is the evaluated product configuration traceable to the marketed device?
• Does the report state limitations, uncertainties and unsupported claims?
• Are PMCF, PMPF and PMS activities tied to specific residual questions?
• Do new evidence and product changes trigger controlled updates?