How Natural Language Processing Improves Clinical Evaluation Reports for Medical Devices.
The EU Medical Device Regulation (MDR) of 2017 (also known as Regulation (EU) 2017/745) has had a huge impact on the regulation of medical devices in Europe, including more stringent manufacturer oversight and increased post-market surveillance.
Although medical devices approved prior to Regulation (EU) 2017/745 coming into force had been granted a three-year transitionary period, that period ended on May 26 of this year.
That means all manufacturers must now comply with the EU MDR, including creating and maintaining a detailed clinical evaluation report (CER) on any medical devices marketed and sold in Europe. CERs are essential for any device to receive a CE mark, a standardized marking indicating the products conform to European health and safety standards. Manufacturers must also regularly update CERs during the device’s entire lifecycle as part of their post-market surveillance activities.
Because of this increased regulation organizations now face the prospect of having to devote more resources to CER creation than ever before – and continue doing that long after these products hit the market. But CERs can be extremely resource intensive and time-consuming, and often require specialized expertise.
However, AI and machine learning (ML) techniques, including natural language processing (NLP), can make CER creation more efficient, effective, and affordable. NLP allows medical device manufacturers to stay on top of their CER reporting requirements without getting bogged down in the many detailed steps required.
What is a CER?
A CER is a detailed analysis and evaluation of a medical device, including technical specifications, use instructions, risk potential, and safety, that uses clinical data to prove the benefits outweigh any risks associated with that device (or similar devices).
Because of the large amounts of clinical data and considerable rigor required, CERs are very similar to the systematic literature review (SLR) process in that they take substantial time and energy to complete.
All medical device manufacturers must complete and maintain a CER for medical devices in Europe by submitting it to a notified body as part of their European CE technical file.
Elements of every clinical evaluation report
CERs typically contain detailed information on a variety of elements, including descriptions of the device and its purpose, any therapeutic or diagnostic claims made by the manufacturer, clinical data and summaries, and methodologies (for assessing literature accuracy along with device safety and performance).
Every CER involves a similar progression of steps, and must undergo several discrete stages as defined by the European Commission:
- Stage 0: Define the scope and plan the clinical evaluation
- Stage 1: Identify any pertinent data, including data from pre- and post-market clinical investigations, risk management activities, preclinical studies, issues with performance or safety, and post-market surveillance (PMS) reports
- Stage 2: Evaluate all data sets in terms of credibility, scientific validity, relevance, and weighting
- Stage 3: Analyze the data and draw conclusions, specifically around compliance with performance- and safety-based essential requirements
- Stage 4: Finalize the CER
The first step of defining the scope typically involves a clinical evaluation plan (CEP) as per MEDDEV 2.7.1 (Rev 4) outlining scope, methodology, and criteria. This CEP must document any changes in materials, design, or the manufacturing process, along with any clinical concerns surrounding the device.
Any CER inevitably involves an exhaustive search of scientific and medical literature, which can take an outsized amount of time using manual search processes.
Tips for creating the most effective clinical evaluation reports
The European Commission also lays out six crucial elements that must be included in your final CER submission:
-
- Executive summary: Summarizes the benefit/risk profile and demonstrates acceptability
- Scope: Identifies which devices are covered in the CER, including detailed technical information, product names, models, and sizes
- Clinical background and current knowledge: A research-heavy look at the context surrounding the device from a clinical perspective, including a summary of the literature review and descriptions of currently available options
- Device under evaluation: Evidence demonstrating that the device conforms to essential requirements, including scientific literature, the company’s own clinical data, and an analysis of the clinical data.
- Conclusion: Clear statements showing the suitability of the device for intended users and the acceptability of the benefit/risk profile.
- Date of the next clinical evaluation: All CERs must be regularly updated, and the manufacturer must define and justify the frequency of these updates.
While Medical Device News Magazine says “clinical data extracted from the literature on previous studies is a valuable tool in compiling a CER,” it also says it’s “critical that literature surveys are conducted in a systematic process.” This includes preparing a comprehensive protocol governing your literature review, including:
- Defining inputs (databases, search terms, exclusion criteria)
- Defining safety and performance criteria
- Systematically gathering appropriate articles according to the criteria
- Analyzing the data based on an objective framework using multiple reviewers
How NLP can help clinical evaluation reviews
AI and ML techniques can improve the speed, efficiency, and accuracy of the deep literature reviews required for CER production, including drawing on medical databases including Pubmed, Embase, and Cochrane.
NLP can speed up clinical evaluation planning (CEP) by semi-automating some elements of scope, methodology, and inclusion criteria definition. NLP can also identify relevant data from various sources, including unstructured free text, far faster than a manual approach. Once identified, NLP algorithms can automatically re-assemble this data in any format required.
NLP techniques are also well-suited to developing the most efficient search protocol based on relevant data, including performing active learning, applying limits, or ranking search results by relevance based on the criteria. For research organizations facing mountains of clinical and other scientific data, this can eliminate days and even weeks of manual work.
Creating compliant CERs often takes months to complete using manual processes. But along with the complexity and time-consuming nature of the CER process, additional challenges remain. It’s often difficult for manufacturers to find the right people – with the right skills and experience in clinical evaluation regulations, data analysis, and medical writing standards – and keep them on staff. And the recurring nature of CER maintenance means it can distract your team and keep them from focusing on core business activities.
An NLP-aided technology solution along with expert subject matter experts and machine learning engineers can perform literature reviews (and periodic updates) faster and more accurately than a manual approach, reducing the time, cost, and effort required for every CER.
The CapeStart difference
CapeStart’s subject matter expertise in CE regulations, medical writing, literature screening, and data analysis combined with deep ML and NLP knowledge is your key to more efficient, ongoing clinical evaluation reporting that doesn’t tie up your team for months.
Our NLP-aided solution can process large volumes of unstructured data and provide quick, actionable insights to assist with decision-making. The solution runs a precise literature search that extracts valuable and clinically relevant information, saving time and effort in collecting evidence. CapeStart also works closely with multiple groups within a manufacturing company to obtain the right data, ensuring evidence is always available in time for inclusion in a CER (especially during CE mark renewals with specific deadlines).
CapeStart’s team of experienced medical writers – who have years of experience building benefit/risk profiles and providing clinical evidence for conformity assessments – can also build the evaluation report itself after analyzing all relevant data.