Development of an IT tool to support post-market surveillance and expert panels in detecting sentinel signals relevant to serious incidents in high-risk medical devices: pilot on italian data

Abstract

Abstract Introduction The EU Medical Device Regulation 2017/745, in force since May 26th 2021, defines new rules for medical device (MD) certification and post-market surveillance (PMS). For high-risk implantable MD, in the certification process notified bodies are obliged to consult clinical Expert Panels (EP) that could decide to proceed for an extensive review of the supporting clinical data provided by the manufacturer, also in view of sentinel signals, such as significantly increased rate of reported serious incidents for a specific MD groups. Purpose 1) To develop an ICT tool to automatically collect and display in an aggregated way the accessible curated regulatory information on MD alerts and recalls to capture possible trends in reported incidents that could be used both for scientific analysis and as information source to EP. 2) To conduct a pilot feasibility study on the Italian data, characterized already by the same European Medical Device Nomenclature (EMDN), organized in a multi-level hierarchical tree code to define a MD, as it will be used in Europe. Methods Web scraping was used to retrieve data of 7622 safety notices (SN) from 2009 to 2021 from the Italian Ministry of Health website. The EMDN code was missing in 68% of cases: to retrieve it, the MD best match was searched within a separate public list of about 1.5M MD on the Italian market containing the EMDN code, using Natural Language Processing techniques and pairwise entity resolution with Cosine similarity to identify similar manufacturers and MD. The performance of this approach was tested on the 2440 SN for which the EMDN code was available as gold standard. A mash up was then performed to integrate data, and to present it to the final user through a graphical interface. Results The implemented entity resolution method was able to correctly assign the correct manufacturer to the MD in each SN in 99% of the cases. Moreover, the correct EMDN code at level 1 (22 categories available) was assigned in 2382 SN (97.62%), at level 2 (146 anatomical or functional groups available) in 2366 SN (96.97%), at level 3 (multiple types available) in 2304 SN (94.45%). The developed interface (Figure 1) allows querying the database by manufactures, devices, type of SN (1a), and selecting the EMDN nomenclature up to the fourth level (1b). As a result, the relevant information is shown, including trends over the selected period and the link to the SN on the original website (2). Conclusions The proposed approach was able to cope with the uncompleteness of the publicly available data in the SN, thus allowing proper matching of MD with its EMDN code up to level 3 with very good performance. In this way, grouping of SN relevant to a specific MD category/group/type could be used as possible sentinel for increased rates in reported serious incidents in high-risk MD. Extension of this approach to aggregate SN from other EU nations could result in an effective support tool in PMS. Funding Acknowledgement Type of funding sources: Public grant(s) – EU funding. Main funding source(s): EU Horizon 2020 - Project CORE-MD