What question did this study set out to answer?

The aim is to improve signal detection for adverse events using natural language processing on unstructured free-text data.

May 2, 20260 citations

Signal Detection of Adverse Events in Medical Devices Using Natural Language Processing

Signal detection of adverse events in medical devices using natural language processing: a case study in pelvic mesh.

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial demonstrates signal detection for adverse events in pelvic mesh, suggesting implications for safety monitoring.

The aim is to improve signal detection for adverse events using natural language processing on unstructured free-text data.
Implemented a proof-of-concept system integrating natural language processing with disproportionality analysis.
Classified free-text reports using topic modelling from an Australian adverse event report database (2012-2017).
Conducted signal detection quarterly using three disproportionality methods: Proportional Reporting Ratio, Bayesian Confidence Propagation Neural Network, and maximized Sequential Probability Ratio Test.
A safety signal for pelvic mesh was detected compared to hernia and other mesh by all three methods in Q3 2014, three years prior to device withdrawal.
Bayesian Confidence Propagation Neural Network provided the most reliable assessment of uncertainty in classifying pain as an adverse event.

Kelly et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark

May 2, 20260 citations

Signal detection of adverse events in medical devices using natural language processing: a case study in pelvic mesh.

View Full Paper

Ask AI

Bookmark

Authors

Randomized trial demonstrates signal detection for adverse events in pelvic mesh, suggesting implications for safety monitoring.

The aim is to improve signal detection for adverse events using natural language processing on unstructured free-text data.
Implemented a proof-of-concept system integrating natural language processing with disproportionality analysis.
Classified free-text reports using topic modelling from an Australian adverse event report database (2012-2017).
Conducted signal detection quarterly using three disproportionality methods: Proportional Reporting Ratio, Bayesian Confidence Propagation Neural Network, and maximized Sequential Probability Ratio Test.
A safety signal for pelvic mesh was detected compared to hernia and other mesh by all three methods in Q3 2014, three years prior to device withdrawal.
Bayesian Confidence Propagation Neural Network provided the most reliable assessment of uncertainty in classifying pain as an adverse event.

Kelly et al. (2026) studied this question.

View Full Paper

Ask AI

Bookmark