EU MDR

MEDDEV 2.7/1 rev 4 literature search: a practical, auditable workflow

By Antonio Brundo · 10 June 2026 · Updated 10 June 2026

Direct answer: A MEDDEV 2.7/1 rev 4 literature search needs a written protocol before execution (Appendix A5), at least two literature databases because PubMed alone is not considered sufficient (Appendix A4), documented inclusion and exclusion criteria, a defined appraisal plan, and a search report complete enough that a notified body reviewer could reproduce the search. This article walks through each element and shows how to run the workflow so the documentation produces itself.

Why a 2016 guidance still decides whether your CER passes

MEDDEV 2.7/1 revision 4 was published by the European Commission in June 2016, under the old directives (MDD 93/42/EEC and AIMDD 90/385/EEC). The MDR — Regulation (EU) 2017/745 — superseded the directives, but it never replaced the methodology. Article 61 and Annex XIV of the MDR tell you that a clinical evaluation must rest on a systematic literature review; MEDDEV 2.7/1 rev 4 remains the most detailed published description of how. Notified body reviewers still use its appendices as the de facto checklist when they assess the literature route in a clinical evaluation report (CER).

That makes the guidance worth reading literally. Most literature-search findings raised by notified bodies are not about missed papers. They are about missing process evidence: no protocol dated before the search, no record of what was excluded and why, no way to re-run the strategy and get comparable numbers.

Appendix A5: the search protocol comes first

Appendix A5 of rev 4 ("Literature search and literature review protocol, key elements") expects the search to be planned in a protocol before the search is executed. The key elements it describes are:

• Background and objective of the search, consistent with the scope of the clinical evaluation. The guidance suggests constructing the research question with a structured method such as PICO.
• Sources of data — which databases and registers will be searched, with a justification of the choice (see A4 below).
• Search terms and strategy — the actual query strings, filters, date limits, and language limits you intend to apply.
• Selection criteria — inclusion and exclusion criteria defined up front, not improvised during screening.
• Quality control measures — for example a second reviewer checking screening decisions.
• A data appraisal plan and a data analysis plan — how retrieved studies will be weighted and combined.

After execution, the search is documented in a literature search report, and any deviations from the protocol must be recorded there. The standard rev 4 expectation is that the documentation is complete enough for the search to be critically appraised — and repeated — by a third party.

PICO for devices, not drugs

PICO (Population, Intervention, Comparator, Outcome) was designed for pharmaceutical questions, and it needs adaptation for devices. The intervention is usually the device or device group plus the clinical procedure; the comparator is the current standard of care or an equivalent device; outcomes should include device-specific safety endpoints (revision rates, device deficiencies, use errors) and not only clinical efficacy. A device search protocol often also needs separate strands for the device itself, the equivalent device if equivalence is claimed, and the state of the art for the clinical condition.

Appendix A4: why one database is never enough

Appendix A4 ("Sources of literature") is short but frequently cited in nonconformities. Its central point: searching MEDLINE or PubMed alone is not sufficient. At least one additional database — EMBASE and the Cochrane CENTRAL trials register are the named examples — should be used to compensate for coverage gaps, particularly European journals and conference literature. Trial registries such as ClinicalTrials.gov and the WHO ICTRP, plus safety databases and the manufacturer's own data, round out the source list.

Appraisal: what Appendix A6 is really testing

Identifying papers is the easy half. Rev 4 then requires each retrieved document to be appraised for its suitability (is the data actually about this device, an equivalent device, or the state of the art?) and its contribution (does the study design and quality allow it to support a safety or performance claim?). Appendix A6 lists examples of studies that lack scientific validity — small uncontrolled case series presented as performance evidence, papers with undisclosed conflicts, results that cannot be traced to a method. An automated workflow can pre-sort by study type and flag weak provenance, but the appraisal verdict belongs to a qualified evaluator. Any tool that claims to automate appraisal end-to-end should worry you.

Reproducibility: what a notified body auditor actually checks

A practical test for your search documentation: could a stranger, given only your protocol and search report, re-run the search and arrive at substantially the same set of included studies? Auditors probe exactly this. Typical questions: Which exact query ran against which database, on which date? How many records came back per source? How were duplicates handled? Why was each full-text exclusion made? If the answers live in a reviewer's memory or an untracked spreadsheet, the search is not reproducible in the rev 4 sense — regardless of how good the included papers are.

This is the strongest argument for executing the search through a system that logs itself. AutoSearch records the query variants, source responses, deduplication by DOI, exclusion reasons, and the verification status of every citation (each DOI is checked against Crossref before it can enter the output — the method is described in how we verify DOIs via Crossref). The PRISMA-style flow diagram is generated from those real counts, not drawn by hand, so the numbers in the diagram, the report, and the log cannot drift apart.

A minimal search protocol skeleton

If you are drafting a protocol today, cover at least these sections: (1) device, intended purpose, and scope of the clinical evaluation; (2) objective and research questions (device, equivalent device, state of the art); (3) databases and registers with justification; (4) full search strings, limits, and planned dates; (5) inclusion and exclusion criteria; (6) screening workflow and quality control; (7) appraisal plan with suitability and contribution criteria; (8) analysis plan; (9) protocol version, author, and date. The same structure can be imported into AutoSearch as the run configuration, which closes the gap between what the protocol promises and what the search actually did.

FAQ

Does MEDDEV 2.7/1 rev 4 still apply under the MDR?

Formally it was written for the directives, and parts of it (for example the equivalence criteria) have been tightened by the MDR and MDCG guidance. In practice, its literature-search methodology in Appendices A4 and A5 remains the reference that notified bodies expect manufacturers to meet or exceed.

How often must the literature search be repeated?

The clinical evaluation must be kept up to date across the device lifecycle (MDR Article 61(11)). For most devices that means re-running the search at least annually or per your PMCF plan — see our guide on PMCF literature review cadence.

Can the search cover an equivalent device?

Yes — if equivalence is claimed, the search must cover the equivalent device explicitly, and under the MDR the equivalence demonstration itself faces stricter technical, biological, and clinical criteria than under rev 4 alone.

Disclaimer. This article is general information, not regulatory advice. Responsibility for the clinical evaluation, its literature search, and conformity with Regulation (EU) 2017/745 remains with the manufacturer and its person responsible for regulatory compliance (PRRC, MDR Article 15). Always verify requirements against the current text of the regulation and your notified body's expectations.

Run it instead of reading about it

AutoSearch executes this entire workflow — protocol-first search across 12 source families, logged screening, Crossref-verified citations, auto-generated PRISMA flow — and returns a structured, audit-ready draft. See the EU MDR/IVDR use case, read the methodology, check pricing, or start an MDR clinical evaluation literature search now.