Author

Marcel Dörkes, Senior Medical Device Consultant & Business Unit Manager - Medical Device Consulting Teams Germany & Benelux, Eurofins Medical Device Services

Article

The increasing availability of Large Language Models (LLMs) and other artificial intelligence (AI)-assisted tools has the potential to significantly enhance the efficiency of toxicological risk assessments and the development of general biocompatibility-related documentation. These technologies can support tasks such as literature screening, data extraction, evidence synthesis, and the identification of potential data gaps, thereby reducing the administrative burden associated with reviewing large volumes of scientific information. Recent publications have demonstrated that LLMs can provide valuable support across multiple stages of toxicological assessment workflows, particularly in the collection, organisation, and synthesis of scientific evidence.

However, it is important to recognise that these technologies are intended to support, rather than replace, the scientific judgement and regulatory responsibility of qualified toxicologists and other competent experts. Current evidence suggests that LLM-generated outputs may be inconsistent, incomplete, or occasionally inaccurate, and therefore require expert validation and contextual interpretation. Critical activities such as study quality assessment, weight-of-evidence evaluation, selection of appropriate points of departure, application of uncertainty factors, and overall risk characterisation continue to rely on expert review and decision-making.

Consistent with the principles outlined in ISO 10993-1, biological evaluations of medical devices should be performed by personnel with demonstrable expertise in the relevant scientific disciplines and a thorough understanding of the device under assessment. Appropriate qualifications, training, and professional experience are essential to ensure scientifically robust and reliable evaluations. 

Within this framework, LLMs may serve as useful tools for information retrieval, literature review, data processing, and evidence summarisation. Nevertheless, their role should be considered supportive rather than determinative. As emphasized by recent publications, the interpretation of scientific evidence, the conduct of weight-of-evidence assessments, and the derivation of regulatory conclusions remain dependent on human expertise and professional judgement.

Consequently, while AI-assisted tools can substantially improve the efficiency of biological evaluations and toxicological risk assessments, they cannot substitute for the competence, experience, and scientific accountability required under ISO 10993-1. Output generated by such systems should therefore be regarded as supportive information that must be critically reviewed, verified, and approved by appropriately qualified professionals.

Eurofins Medical Device Services can provide support at every stage of this process. Leveraging our extensive in-house expertise in toxicology, biological evaluation, chemical characterisation, extractables and leachables, and medical device regulatory sciences, we help clients maximise the benefits of AI-assisted technologies while ensuring scientific rigour and regulatory compliance. 

Our experts can independently review AI-generated outputs, perform comprehensive toxicological and biocompatibility assessments, and develop robust, scientifically justified evaluations tailored to regulatory expectations. 

Through our multidisciplinary team and decades of experience supporting the medical device industry, Eurofins Medical Device Services ensures that all conclusions are based on a thorough assessment of the available evidence and remain fully aligned with the requirements of the ISO 10993 series and applicable global regulatory frameworks.

Get in touch with our experts