Clinical Literature Search Methodology for Medical Device CERs

The literature search is where most CERs have their most visible methodological weaknesses. It's also the section Notified Bodies scrutinise most closely, because it's the easiest to audit: they can look at your search strings, re-run them, and check whether what you found and what you included are consistent with a systematic approach. Teams that produce informal literature reviews — reading some papers, summarising the good ones — and try to present them as systematic searches don't usually survive that scrutiny.

What "systematic" actually means

A systematic literature search is one that is planned in advance, documented in enough detail to be reproducible, and executed consistently against defined criteria. It's not about searching every database or including every paper ever written on the topic. It's about being able to show, step by step, how you moved from a defined research question to a defined set of included evidence.

The minimum components of a systematic search that will hold up under MDR are:

• A documented search strategy, including the search question (often framed as a PICO: Population, Intervention, Comparator, Outcome)
• Database selection with justification — which databases you searched and why those were appropriate for the evidence type you were looking for
• Search strings — the actual queries you used, including Boolean operators, MeSH terms if applicable, and any field restrictions
• Date range — the period your search covers and a rationale for the cut-off
• Inclusion and exclusion criteria defined before the search results were reviewed, not after
• A PRISMA flow diagram (or equivalent) showing how many records were identified, how many were screened, how many were reviewed in full, and how many were included

The last point is where teams most often cut corners. The PRISMA diagram makes transparent any filtering decisions that happened after the search. Notified Bodies use it to check whether the final included set looks suspiciously aligned with a favourable conclusion.

Database selection

For most medical device literature searches, the minimum database set is PubMed/MEDLINE plus one or two secondary databases relevant to the device type. For implantable devices and Class III devices, Notified Bodies increasingly expect searches to include EMBASE in addition to PubMed. For some device categories, clinical trial registries (ClinicalTrials.gov, ISRCTN, WHO ICTRP) should also be included to capture unpublished or ongoing investigation data.

The choice of databases should be justified in the CER. "We searched PubMed" without further explanation raises questions. "We searched PubMed and EMBASE because they provide the most comprehensive coverage of peer-reviewed clinical and biomedical literature for our device category, supplemented by the relevant manufacturer post-market surveillance data" is a defensible justification.

Building your search strings

Good search strings balance sensitivity (finding everything relevant) with specificity (not drowning in irrelevant results). The two most common failures are strings that are too narrow — missing relevant literature — and strings that are so broad they generate thousands of results that are manually filtered down, making the process non-reproducible.

A practical approach for most device CERs:

1. Start with your device's generic name and key synonyms
2. Add the clinical condition or intended use as a secondary term
3. Use Boolean operators (AND/OR/NOT) to combine terms logically
4. Add any necessary field restrictions (e.g., human subjects only, specific publication types if appropriate)
5. Run the search and review a sample of results to check whether the sensitivity and specificity are appropriate before locking the string

Document the full string exactly as entered. If you run the same search in multiple databases, note any differences in syntax required by each database's interface.

Inclusion and exclusion criteria

Criteria must be defined before reviewing results. Applying criteria after seeing the results is a form of bias — it allows you to exclude inconvenient data post-hoc. Notified Bodies are aware of this, and a CER that can't show pre-defined criteria will be questioned.

Typical inclusion criteria for a device CER: studies involving the device or equivalent device, human subjects, relevant clinical outcomes, peer-reviewed publication or verified clinical data source.

Typical exclusion criteria: animal studies, in vitro studies without clinical relevance, case reports with insufficient data, non-peer-reviewed sources without other validation, studies on clearly different device generations.

If you need to exclude studies that report unfavourable results, you need a methodological reason. "The study used a patient population not representative of our intended use" is a legitimate reason. "The study showed worse outcomes than we wanted to report" is not.

Documenting your search and appraising the results

The search documentation — the strings, databases, dates, and PRISMA diagram — goes into the CER itself or into a documented search protocol that the CER references. Either approach works; what matters is that the documentation is complete and accessible.

After the search is complete, each included paper needs critical appraisal, not just a citation. The appraisal covers study design, risk of bias, comparability of the patient population, relevance of endpoints, and applicability to your device. This is the step that separates a defensible CER from a bibliography with commentary. The MDCG 2020-5 guidance describes the appraisal requirements in detail — treat it as a checklist when reviewing each paper.

Updating the literature search

Your literature search has a date. When you update your CER — whether triggered by PMCF data, a surveillance audit, or a planned review cycle — the literature search must be updated to cover the period since the last search. A CER that references a literature search from three years ago without an update is not compliant, regardless of how thorough the original search was. Build the update cadence into your PMS schedule alongside the CER review.