Planning and running the literature search
The literature search that supplies a device's clinical evaluation with published evidence is run to a written protocol, across more than one database, iterated until it reliably finds the papers it should, and documented well enough to be reproduced at every update.
What the search is for
A literature search for a clinical evaluation report can be designed to identify the device's clinical performance, comparisons of the device with existing devices, and device safety in the form of adverse device effects. Information found in the biomedical literature strengthens every stage of the device lifecycle, from concept and design through clinical trials to commercial release and reimbursement, and then again in post-market surveillance.
What the regulation and guidance require
The MDR prescribes a systematic scientific literature review as part of clinical evaluation, to identify the clinical data relevant to the device and its intended purpose and any gaps in the clinical evidence. The review has to be updated periodically throughout the device's lifetime, and the regulation also includes regular screening of the literature among the methods of post-market clinical follow-up. The adoption of the MDR and the In Vitro Diagnostics Regulation has greatly increased the demand for systematic literature reviews.
MEDDEV 2.7/1 revision 4 places increased emphasis both on a quality assessment of the available evidence from the literature and on the scientific validity of the review itself.
Guidance documents from regulators such as the European Commission and the International Medical Device Regulators Forum advise a highly structured search strategy to ensure traceability and reproducibility. Under MEDDEV 2.7/1 revision 4 and the International Medical Device Regulators Forum's 2019 clinical evaluation document, the search has to be clear, concise, systematic, traceable and reproducible, and the evaluation has to consider unfavourable as well as favourable results. MEDDEV 2.7/1 revision 4 repeatedly asks for an ordered process intended to deliver identical search results, and the Forum's 2019 technical documents on clinical evaluation and on clinical evidence give further key elements and requirements for literature searches. Insufficiently structured searches, with such faults as inappropriate search limits, inadequate use of Boolean operators or unsuitable search terms, possibly played a part in the procedural failures that occasionally let faulty or ineffective devices onto the market by the similarity route, under which a device gains market access on its similarity to a product already marketed.
The protocol and the review questions
A literature review protocol should be consistent with the scope of the clinical evaluation and use objective, unbiased and systematic search and review methods, such as framing the patient characteristics, intervention, control and outcome of a PICO question. The review questions draw on the instructions for use, with the device description, its intended performance and any claims made about clinical performance and safety, and on the risk management process, and they should also address gaps in the clinical evidence, such as how comprehensive the data are and the number and severity of adverse events. Example review questions might include what interventions characterise the state of the art, what comparators can be identified, and what clinical data exist to assess safety and performance and whether that evidence is sufficient for the clinical evaluation.
Revisions 3 and 4 of MEDDEV 2.7/1 both require the review to rest on an objective research question, be conducted systematically, and have a search protocol and a search report, and both speak of the literature review as though one protocol, strategy, report and review were needed. Because each search has to be tailored to its purpose, more than one is required in practice: separate protocols, strategies and search outputs for the current-knowledge and state-of-the-art sections and for the device under evaluation or its equivalent, and further searches may support post-market surveillance.
Search terms and strategy
Choosing the right search terms, developing the search strategy and knowing how to search databases are essential for a successful literature search, and the review questions and previously conducted searches inform the terms.
The search strategy should define which databases will be searched and over what period — ten years is reasonable for initial CE marking, while a renewal searches from the date of the previous search — and limits such as language and article type are then applied to bring the results down to a manageable number of relevant articles.
Searching is iterative: the strategy is adjusted until the researcher is satisfied that as many relevant papers as possible have been retrieved. Testing the strategy is strongly recommended: successive iterations should consistently identify known key papers, and where they do not, the strategy must be modified until they do. The final strategy, the date the search was run and the number of articles identified at each step should be documented in the protocol so that the search can be reproduced if necessary.
Choosing the databases
It is important to search more than one database. MEDLINE or PubMed is free to access and fairly easy to search, but its search features are not sophisticated and its coverage of some European journals is incomplete, so additional databases such as EMBASE and the Cochrane Database of Systematic Reviews should also be used; the Cochrane database is free to access and EMBASE is not.
MEDDEV 2.7/1 revision 4 does not accept PubMed as the sole source of literature: it indicates that MEDLINE or PubMed may not cover all European journals and requests additional databases, such as the Cochrane Central Register of Controlled Trials or Embase. MEDLINE or PubMed is nonetheless the main source of literature. Literature searches should be reproducible, use multiple databases and use verified methodology.
Choosing the right databases for a literature review, including those needed to establish the state of the art, can be tricky. As of 2024, PubMed and MEDLINE hold more than 37 million references, offer multiple selection criteria, export results to Excel and other formats compatible with reference managers such as EndNote, and are free, but may not cover all European journals, especially those reporting user experience, and articles that were not indexed properly may be lost to the search. In 2024 Embase held over 29 million records from 8,500 journals, MEDLINE included; it offers multiple selection criteria but is not free. Embase can be used for all types of systematic literature search, allowing broader as well as narrower searches. Cochrane CENTRAL holds only high-level evidence, such as Cochrane systematic reviews and protocols, clinical trials and clinical answers, and is free but limited to Cochrane publications. Google Scholar covers more references than any of them, but its output cannot be properly documented, varies with every search and includes material that is not peer-reviewed, such as trade and white papers or interviews, and notified bodies do not officially accept it as the primary database.
No database is better than the others and none should be used exclusively, and the choice of a database should be justified as the most appropriate for the search in hand.
For published literature, one practitioner highly recommends PubMed, Embase and the Cochrane Database of Systematic Reviews, with Europe PMC and Google Scholar as viable options.
Trial registers, vigilance databases and safety summaries
Writers are encouraged to search clinical investigation data as well as published literature, since that search can find data not found by other means. ClinicalTrials.gov, the WHO International Clinical Trials Registry Platform and Cochrane Central hold such data, and the EU Clinical Trials Register can also be a good resource. Clinical experience data — suspected device-associated deaths, serious injuries and malfunctions — are held in the FDA's MAUDE, recalls and Total Product Life Cycle databases; Health Canada's incident and recall databases; the UK MHRA database; Germany's BfArM field corrective actions and recommendations; Swissmedic's field safety corrective actions and recalls; and Australia's TGA adverse event and recall databases. Summaries of safety and clinical performance, publicly accessible and up-to-date summaries of a device's clinical data, are another helpful resource, accessible through EUDAMED.
Keeping the review up to date
The MDR dictates how often a device's review is updated, for example annually for the highest-risk class III devices, whereas Cochrane accepts that not every systematic review can be kept up to date and has developed guidance on when an update is appropriate. Each step of an MDR-compliant review has to be documented well enough to be reproduced at future updates. With the conventional review process, made up mostly of manual and repetitive tasks, maintaining a review for each product to meet regulatory requirements quickly becomes untenable for a company with many devices on the European market.
The review and the report
The review is a substantial section that can take as long to write as the rest of the report, and it can sit inside the report or stand as a separate document that the report summarises. A separate review limits the size of the report and makes it more navigable, since a review can easily run to a hundred pages, but the two then have to be kept consistent.