Research Project: Semantic Data Mining
This project is aimed at developing a semantic text mining framework to assist signal detection in post-market as well as pre-market approvals, by analyzing data available to the Center.
Analysts at CDRH study adverse event reports received for identifying signals. However, the adverse event reports themselves are not sufficient to identify and confirm the indication of a signal. Analysts resort to various means to confirm the signal including analyzing the narrative texts in the report, consulting subject matter experts, communicating with the manufacturers, and reviewing pre-market submissions, annual reports and recall databases. Further challenges include handling combination products, imported devices/sub-modules/components, and the use of nascent technologies in medical device design, just to name a few. The number of documents requiring study and analysis is increasing at a rapid rate. Semantic text mining is a technique that enables the analysis of a large number of documents containing text for specific relevancies, summarization, extraction, and retrieval.
From a pre-market perspective, submissions are evaluated by comparing them with similar devices already approved, if these exist. Many of the device documents either claim conformance to or refer to standards and guidance documents. Verification of these claims is again performed by analyzing the narrative text from multiple documents. Semantic text mining can provide the ability to hasten the comparison and analysis.