Evidence Based Management of medical devices using natural language processing and neural networks to study medical devices failures

The aim of this work is to provide a method to implement Evidence Based Management by developing a framework to support Health Technology Assessment of medical devices through Natural Language Processing (NLP). Electronic Health Records, Computerized Maintenance Management Systems and Spontaneous Reporting Systems are potential resources of massive information and hidden knowledge which can be used to generate Real-World Evidence to assess both effectiveness and safety of a given health technology. Transformers have become the elective deep learning model to solve NLP problems and biomedical causal relation extraction tasks, leading to the development of pre-trained systems such as Google’s BERT (Bidirectional Encoder Representations from Transformers). Specifically pre-trained BERT-based models (SciBERT, BioBERT, ClinicalBERT) have then been developed to more effectively and efficiently process biomedical and clinical text information, outperforming classical deep-learning models. Extracted Real-World Evidence can have a strong significance in fulfilling all the legal obligations regarding Post Market Surveillance and Post Market Clinical Follow-up, promising to be very useful also in analysing faults, planning updates, interventions and avoiding recalls.