Author: Gaugarin Oliver Artificial intelligence. Let’s say you are a medical device manufacturer with any presence in the European Union. You probably already have the EU MDR, a new regulatory system that requires more clinical evidence than the Medical Device Directive (MDD), which was implemented earlier in the year. Bad news! The bad news? The time-consuming process of finding data that conforms to European MEDDEV guidelines (Medical Devices Documents), which regulate the creation and submission of clinical evaluation reports, (CERs), is what causes this resource stress. Medical device manufacturers must maintain compliance and obtain a CE Mark to distribute within the EU. This means that they have to submit a CER, which is an independent, objective, and current clinical device evaluation based on internal and published literature. CERs must conform to the Essential Requirements in MEDDEV.2.7/1 Rev. Annexe 1. This adds up quickly, especially when you consider MDR’s recent requirements regarding device classification, technical documentation and post-market surveillance. Artificial intelligence (AI) and natural language processing technologies (NLP), are both powerful tools that medical device manufacturers can use to create and maintain comprehensive CERs. How exactly do you accomplish this? Let’s see. What is a CER? CERs, at their heart, are very detailed risk/benefit assessments. CERs are designed to ensure that the medical devices have the best possible outcomes. They can also include information about similar devices or specific data. The production of CERs typically involves five stages. Stage 0: Scope and planning Stage 2: Evaluation of relevant data Stage 3: Analyse of clinical data Stage 4. Finalization of the Clinical Evaluation Report. CERs are submitted to designated notified bodies (NBs), who will weigh whether or not they perform as expected. Is it safe? Are there other methods that are better than it? The NB reviews these queries and determines if the device is allowed to be sold in Europe. CERs: Identification of relevant data Companies must submit a complete evaluation of all technical features and testing results of the device as pertaining to both safety and performance, including: Preclinical testing data used for verification and validation (including bench and/or animal testing) Any device failures during testing or otherwise Manufacturer-generated clinical data such as: a) Premarket clinical investigations b) Postmarket clinical data c) Complaint reports d) Explanted device evaluations e) Field safety corrective actions Instructions for use Other crucial data points for inclusion surround the usability of the medical device, determined through human factor studies, and a review of the state of the art (SOTA) of the relevant clinical field. If done incorrectly, data collection for CERs can be confusing at best and dangerous at worst. The results of manual literature search can vary greatly due to both the expertise of search professionals and several common mistakes and challenges. Indani Et Al.’s 2017 study explains. Indani Et Al. – Common data inclusion mistakes Several common errors that could delay or derail scientific literature searches are described in Indani Et al. too much data). Use of general or vague search terms, non-specialized medical databases and flawed boolean logic may lead to overwhelming amounts of information. This can also result in large quantities of duplicate, semi-relevant or irrelevant data. Exclusion errors (i.e. too little data). Search terms that are too narrow or exclude Boolean logic, such as “AND”, can result in the opposite problem: a lack of useful data. Inclusion errors. This includes data selection and keyword biases by those who conduct the literature search for the desired result. Exclusive inclusion errors. This error limits the results because of using very specific search terms. Limited relevance errors due to exclusive exclusions. Bias and excessive specificity can lead to one-sided search results that don’t contain enough information. This can lead to CER production delays or rejections by NBs under MDR. NLP can be your secret weapon in CER literature research. For all these reasons and more, scientists believe that automation via AI/NLP is crucial for increasing the speed, efficiency, cost-effectiveness and accuracy of CER search. Indani Et Al. state that artificial intelligence, natural language processing tools with cognitive abilities and natural language processing provide the perfect solution to literature search. Researchers also claim that NLP-based automation decreases bias and reduces multiple searches by algorithm reuse and customization. It can allow for automatic translation of other sources and speed up literature selection and extraction. This will “significantly decrease” the time required for manual searching and filtering the appropriate content. They state that automation will ultimately reduce costs, time and the resources required for the entire process. The best way to prevent errors in literature searches is to use a literature search tool that’s well-trained and certified. The NLP models are able to learn from reviews and order scientific literature according to relevance. This greatly improves the accuracy and efficiency of literature reviews and speeds up their processing. Medical device manufacturers can now produce compliant, audit-ready CERs in half the time it takes to do a manual review. This is possible because of ready-to use form templates and automated quality assurance. Also, this allows for duplicate identification and detection of common errors like exclusive exclusions. Getting the right data for clinical evaluation reports: an AI-powered approach was published originally in on Medium. People are responding and highlighting this story. Published via
