Development of a Framework Dealing with Partial Data Unavailability and Unstructuredness to Support Post-Market Surveillance

Under the European Union Medical Device (MD) Regulation 2017/745, expert panel's decision on providing a scientific opinion on the Clinical Evaluation Assessment Report for high-risk MD is required, as part of the conformity assessment procedure. To this aim, the perceived risk of similar MDs already on the market, based on the European Medical Device Nomenclature (EMDN), could help. To generate such information, we propose a generalized framework to automatically collect and display in an aggregated way the publicly available safety notices (SNs), even when characterized by partial unstructuredness and incompleteness. This novel approach was tested on the Dutch data, consisting of 3618 SNs from 2015 to 2022, retrieved from the official government website by Web scraping. After the identification of named entities, the best match MD was searched within the Italian and Portuguese datasets of devices using Natural Language Processing techniques. Algorithm performance was tested on potentially equal SNs (472) published by both the Dutch and Italian authorities: assignment of the same EMDN code at level 1 was present in 454 out of 472 (96.19%) SNs, at level 2 in 447 (94.70%) SNs, at level 3 in 433 (91.74%) SNs. The proposed approach was able to cope with public data unavailability and incompleteness, thus providing structured data with appropriate EMDN usable for aggregation and safety signal detection.

Automated summary

Under the European Union Medical Device Regulation, expert panel's scientific opinions on high-risk medical devices are required. To assist in this process, a framework was proposed to collect and display publicly available safety notices, using named entity recognition and natural language processing techniques.