A structured method to achieve cognitive depth for medical device use error topic modeling

Abstract

Human error associated with medical device use may lead to devastating consequences for end-users. Identifying post-market associations between instances of use error can inform human factors design decisions and guide regulatory action. The US Food and Drug Administration (FDA) requires medical device manufacturers, importers, and user facilities to track and report instances of adverse events. These reports are available in the Manufacturer and User Facility Experience (MAUDE) database. MAUDE exists to support post-market surveillance and to aid the discovery of adverse event-medical device associations. Each event contained in MAUDE contains an event narrative: a free-text description of the event. These event narratives are coded with a “device problem code” that describes the nature of each event that can aid in identifying trends, how,ever codes related to human factors are limited in detail. In the authors’ prior work, new use error categories for MAUDE entries were proposed tprovideides decomposition based on primary and secondary use error. In this work, these use error categories were used to structure entries based on narrative content. Topic modeling was performed for automatic extraction of narrative themes for use error MAUDE data from 2010 – 2019. Latent Dirichlet Allocation, an unsupervised generative model, was used to provide a descriptive analysis of this textual data and identify thematic topics. Notable outcomes included the categorization of narratives into six distinct topics; the first five primarily involved rule-based errors during the operation of glucose self-management devices, and the sixth involved knowledge-based errors during inpatient surgical procedures. Distinct divides between error narratives for healthcare providers and patients, as well as for different device types were observed, demonstrating an alignment with proposed use error categories. These categories can be used to monitor trends for specific medical device user segments and can inform device manufacturers of usability design requirements that must be addressed.