Title: Developing Evidence for Safety Surveillance from Device Adverse Event Reports Project Summary/Abstract Population level studies have shown that the device-based hysteroscopic sterilization was associated with increased risks of reoperation during follow-up, when compared to traditional laparoscopic sterilization. However, secondary data sources often lack the granularity to understand the nature of patient and device complications related to the device removal and additional surgery. The Manufacturer and User Facility Device Experience (MAUDE) database houses medical device reports submitted to the FDA by mandatory and voluntary reporters. These reports contain detailed information of patient and device adverse events. But due to its narrative structure, research with the reports has been limited, partially due to the restrictions of keyword search and manual review. The proposed study will innovatively apply natural language processing (NLP) to analyze MAUDE reports of device removals. NLP is a powerful tool capable of extracting information efficiently from documents such as medical notes, allowing the summarization of thousands of adverse event reports in a cost-effective way. The primary aim is to develop an NLP program to extract and summarize patient- and device-specific complications associated with device removal and additional surgeries following hysteroscopic sterilization. Secondary objective is to evaluate the impact of regulatory activities on adverse event reporting behavior and structure. The hypotheses are that the majority of reported removals were associated with device-related complications as opposed to persistent symptoms only, and that after the FDA convened a panel discussion in September 2015, adverse event reports were more likely to be submitted by mandatory reporters, with improvement in structured presentation. Adverse event reports related to device removal will be selected from the MAUDE database using keyword search first, and 1,000 reports will be annotated and used to develop and validate the NLP tool. Applying the developed NLP to all reports, extracted information will be used for the analysis, and comparisons will be made before and after September 2015. The significance of the proposed research is that it will develop a method to better utilize adverse event reports to obtain crucial device safety information supplemental to regular population-level studies. By achieving this, the long term goal is to create a useful tool for future medical device safety surveillance to understand the nature of adverse events. The immediate next step will be to use the tool to investigate device safety in other areas. The comprehension of the nature of device adverse events and the elucidation of the crucial role of regulatory activity in facilitating reliable adverse event reporting will help promote patient safety evaluation and monitoring.
The proposed research will focus on device adverse event reports and develop a powerful tool to analyze reports to understand the nature of patient and device complications related to device removals and additional surgeries. The study will also elucidate the impact of regulatory activities on adverse event reporting. Thus, the proposed research is relevant to part of AHRQ's mission to make health care safer.
