From search to CER – How to make your literature review process smarter

This blog post explains why the literature review process is the often underestimated foundation of every clinical evaluation – and how to set it up systematically, transparently, and in an audit-proof manner. You'll learn how to screen publications without compromising quality, where AI support can be used safely within regulatory frameworks, how structured extraction improves quality, which errors regularly lead to deviations, and how to correctly incorporate evidence into the benefit-risk assessment.

Abbreviations

abbreviation	Meaning
CEP	Clinical evaluation plan
CERIUM	Clinical Evaluation Report
MDR	Medical Device Regulation (EU Ordinance 2017/745)
Pmcf	Post-market clinical follow-up
Sota	State-of-the-art

Underlying regulations, standards and guidelines

EU Regulation 2017/745 (MDR)

MEDDEV 2.7/1 Rev. 4 – Clinical evaluation: A guide for manufacturers and notified bodies

1 Introduction

Literature research is one of the most time-consuming and error-prone tasks in a clinical evaluation according to the MDR. And yet, or perhaps precisely because of this, in practice it is often started in the wrong place, insufficiently documented, or not consistently linked to the clinical evaluation report (CER).

Too often it is understood as a preparatory step – as a means to “collect relevant studies.” But under the MDR, it is far more than that.

The literature forms the basis for key elements of clinical evaluation:

• the State of the Art (SotA)
• the derivation of clinical claims and measurable parameters
• the substantiation of clinical claims via measurable parameters
• the benefit-risk analysis
• the identification of clinical data gaps

If the literature review is unsystematic, this has direct consequences:

Clinical evaluation becomes selective rather than comprehensive, descriptive rather than analytical – and in the worst case, vulnerable to regulatory challenges.

Typical weaknesses repeatedly become apparent, especially here:

• missing or unclear search strategy
• non-reproducible search runs
• no agreed Clinical Evaluation Plan (CEP) as a starting point
• unstructured or inconsistent screening

These problems are rarely the result of a lack of literature. They arise from a lack of structure.

This article shows you how to change exactly that – and turn your literature review from a weakness into a solid foundation for your clinical evaluation.

2. The right start: Why the CEP comes before the search

One of the most frequent – ​​and at the same time most consequential – errors in clinical evaluation lies not in the execution of the literature search, but before its actual start: starting without a defined Clinical Evaluation Plan.

In practice, the research often starts operationally – databases are opened, initial search terms are entered, and results come in. Weeks go by. And then it turns out: the search was not aligned with the scope, purpose, or clinical parameters of the CEP.

What seems intuitive is problematic from a regulatory perspective. Because under the MDR, the following applies:

A literature review must be systematic and reproducible. And that is only possible if the strategy is defined in advance.

An MDR-compliant, efficient literature review process therefore follows a clear sequence:

1. Clinical Evaluation Plan (CEP)

→ First define the scope, purpose, target population, clinical claims, and parameters

2. Search strategy

→ Build databases, search terms, inclusion and exclusion criteria based on the CEP

3. Literature review

→ Perform systematically and fully document each step

4. Screening

→ Check title, abstract and full text against predefined criteria

5. Critical evaluation

→ Assess the methodology, quality, and clinical relevance of included studies

6. CER

→ Synthesize evidence for a report

Each step depends on the previous one. If a phase is skipped or completed too quickly, the entire process becomes difficult to defend.

3. High volume, tight deadlines – how to maintain screening quality

You've developed a solid search strategy and run the search across all relevant databases. And now you're faced with 100, 500, sometimes over 1,000 hits.

To put that into perspective: 1,000 abstracts at only 2 minutes per abstract means over 33 hours of work – before the full-text review has even begun.

Under the MDR, this process must be systematic, transparent, and reproducible. There is no way around this requirement. The question is how quality can be guaranteed at high volumes within realistic project timeframes.

Five principles for robust screening

• Screening protocol first: Inclusion and exclusion criteria must be defined and documented before the first match – they must not be adjusted during the screening.
• Proceed in stages: Title screening → Abstract screening → Full-text review. Each phase reduces the volume before the next one begins.
• Structured screening log: Every decision must be traceable – who reviewed the paper, what decision was made, why a paper was excluded.
• Incorporate quality control: A second person reviews a sample of the exclusion decisions – identify inconsistencies early.
• Using AI in a targeted way: AI can provide support at the title and abstract level – more on this in the next section.

The goal is not just speed, but a screening process that is fully transparent with every decision.

4. AI in literature screening – where it really helps and where it doesn't

AI tools are increasingly becoming part of the workflow of clinical evaluation teams. The crucial question is not whether AI should be used, but where it actually adds value and where its limitations lie.

Where AI brings real benefits

• Title screening: AI can identify potentially relevant titles and add short contextual notes – this significantly speeds up the first viewing round.
• Extraction: Structured extraction of study design, population, endpoints, and results.
• Consistency: AI applies the same logic to hundreds of entries, reducing variability between reviewers.

Where AI has clear limits

• Critical assessment: Assessing methodological quality requires clinical judgment, which AI cannot reliably provide.
• Clinical interpretation: Linking evidence to your specific product and target population is expert work.
• Inclusion decisions: Every final decision must be made and accounted for by a qualified evaluator.
• Risk of hallucinations: AI can generate plausible-sounding content that is not in the source. Critical data points must always be verified.

The core requirement remains unchanged: your screening process must be reproducible, transparent, and fully justifiable. MEDDEV 2.7/1 Rev. 4 explicitly addresses this.

Three questions to assess compliance readiness:

• Can your process be reproduced? Could another evaluator follow your methodology and arrive at the same results?
• Is the role of AI explicitly defined? AI as a supporting tool can be defended. AI as a decision-maker cannot.
• Is human supervision documented at every step?

In short: AI can handle the preparation, selection, and scanning – the time-consuming part. Experts make the decision.

5. From abstract to decision – structured extraction as a quality lever

Abstract screening is the phase in which cognitive fatigue sets in most quickly. You read densely written, inconsistently structured paragraphs – each containing the information you need to make an inclusion decision, but rarely in the same format.

The concentration decreases after 50 abstracts. After 150, inconsistency creeps in.

The solution isn't to screen faster. It's to screen smarter.

Structured extraction means extracting the same key elements from every abstract – regardless of how the original is formulated:

• Study design: RCT, observational study, registry, case series?
• Patient population: Who was examined? How many? What was the indication?
• Product or intervention: Is this truly comparable to your product?
• Primary endpoints: What was the study designed for?
• Key findings: What were the results?

When this information is consistently prepared across all abstracts – whether manually, using a template, or extracted with AI support – reviewers work faster, make better decisions, and create a more defensible screening log.

The principle applies regardless of the method: consistency drives quality.

6. Full-text evaluation in the CER – what matters and how it is documented

The full-text review is where the real clinical work takes place. Here, it's no longer just about relevance – but about the more challenging questions:

What is the design of the study?

Are the results valid?

Is this evidence applicable to my product and my patient population?

It is also the phase in which documentation requirements are highest – and in which incomplete or unclear assessments most frequently lead to deviations.

What a complete full-text review of product-relevant literature must cover:

• Study methodology: Is the design suitable for the research question?
• Patient population: How closely does the study population match your target population? Are there any relevant differences that affect generalizability?
• Endpoints and results: Are the endpoints clinically meaningful? Do they align with the safety and clinical performance claims in your CEP?
• Results and statistical validity: Are the results clearly presented? Is the statistical analysis appropriate?
• Limitations: What does the study itself acknowledge? What additional limitations do you, as a reviewer, identify?
• Clinical relevance to your product: What specific contribution does this evidence make to your clinical evaluation?

7. From literature to benefit-risk assessment – ​​how evidence supports the CER

Conducting a literature review and screening publications is necessary, but not sufficient. The crucial question is: What does this evidence specifically mean for your product?

In practice, many CERs lose their coherence precisely at this point. The literature exists within the document, but it does not drive the document forward.

How evidence should be incorporated into clinical evaluation

• From literature to clinical safety: What does the evidence say about known risks of this type of product? What complication rates are reported? What adverse events appear across studies?
• From literature to clinical performance: Which clinical outcomes does the evidence support?
• From literature to state of the art: What does the literature say about the current treatment standard? How does your product position itself in this regard?
• From all this, let's turn to the benefit-risk ratio: Only when clinical safety, clinical performance, clinical benefit and state of the art are clearly established on an evidence-based basis can a meaningful benefit-risk analysis be carried out.

Literature is not a section of the CER. It is the foundation of the entire document.

If your evidence synthesis and your benefit-risk analysis could have been written independently, that is a sign that the connection and relationship need to be strengthened.

8. The three most common errors that lead to deviations

Project experience repeatedly reveals the same weaknesses, which trigger inquiries from notified bodies. Here are the three most common ones – and what to do instead.

Error 1: The search strategy is insufficiently documented

The search was conducted. The results were reviewed. But the strategy itself – the exact search terms, the databases used, the execution date – was not fully recorded.

An auditor from a notified body cannot assess a search they cannot see. And a search that is not reproducible fails the fundamental MDR requirement of transparency.

What we do instead: Every search is documented with complete search terms per database, execution date, and number of hits. The strategy is included as an appendix (literature search plan and literature search log) in the CER – not just referenced.

Error 2: Inclusion and exclusion criteria were applied inconsistently

The criteria existed – but were interpreted differently by reviewers or informally adjusted during the screening process. This problem is difficult to rectify retrospectively because the inconsistency often only becomes apparent when decisions are compared side by side.

What we do instead: Criteria are defined and documented before the screening begins. The evaluation of the publications is documented in the literature search protocol.

Error 3: The link between literature and clinical conclusions is not explicit

The studies are listed. The benefit-risk analysis has been written. But the path from evidence to conclusion is not clearly outlined.

What we do instead: Every clinical conclusion in the CER is explicitly linked to the evidence that supports it. The literature doesn't simply exist in the document—it drives the document.

9. CER Update vs. New CER – How to Scale Search Correctly

Not every literature review needs to be rebuilt from scratch.

New CER – building a complete evidence base

For a newly created CER, the literature search process must be established from scratch. Its search scope should be broad enough to encompass the entire spectrum of available evidence. The time period typically covers at least the last 10 years for the state of the art and the time since the product's market launch. If it is not yet on the market, similar products serve as a guide.

The goal: a complete, defensible evidence base that supports every clinical conclusion in the CER.

CE Update – Close the gap, don't repeat work

A CER update has a different goal: to identify new evidence that has emerged since the last search and to check whether it changes existing conclusions.

• Search from the date of the last search until today
• The same databases and search terms as the original – consistency is crucial for comparability
• Explicitly document that this is an update search, with reference to the original

When an update is not enough

There are situations where an update search is not enough:

• Significant change in purpose or indication
• New security signals that require broader evidence review
• Significant changes to the product design

10. Conclusion

This is something that is repeatedly demonstrated in practice:

The problem is not the availability of literature, but how it is used.

A reliable literature review is characterized not by the quantity of studies found, but by clear principles:

• Strategy before execution – the search is planned, not exploratory
• Structured documentation – from plan to minutes to report
• Clear separation of evidence levels – state-of-the-art and product-specific data fulfill different functions
• Objective study selection – based on defined criteria and transparent screening
• Critical evaluation instead of description – data is weighted, not just summarized
• Clean linkage with the CER – every statement is evidence-based and traceable
• Continuous updating – as an integral component

Ultimately, this very system determines whether your clinical assessment:

• appears defensive or argues convincingly
• is vulnerable or can withstand an audit

Or to put it another way:

A good literature review does not only answer the question of what is known.

It clearly and comprehensibly demonstrates:

Why you arrive at your conclusions – and why they are sound.

11. How we can help you

An MDR-compliant literature review process is complex – and a critical bottleneck in many projects. We support you in setting up this process in a structured, efficient, and audit-proof manner.

Our focus is on:

• Strategy & Planning: clear definition of search strategy and criteria based on your CEP
• Structure & Documentation: Creation of plan, protocol and report – reproducible and MDR-compliant
• Screening & Evaluation: Systematic study selection and thorough critical appraisal
• Linking to your CER: Clear traceability of claims for evidence and identification of data gaps
• Lifecycle & Updates: Building sustainable processes for continuous updates

The aim is a literature review that is not only complete, but also methodologically sound and regulatory compliant.

Want to know more? Contact us for a free initial consultation!

You can get a free initial consultation here: free initial consultation