Healthcare investigations produce large volumes of unstructured text containing rich potential for learning over time, but patterns across hundreds of reports are hard to see and manual analysis is resource intensive and costly.

This talk describes I-SIRch, a framework developed with HSIB (now HSSIB) and human factors colleagues at Loughborough University that annotates investigation reports against a SEIPS-based human factors taxonomy. Trained on expert annotations from maternity reports, it was evaluated on unseen reports and tested for consistency across ethnic groups.

Extending to multi-document summarisation, we produced summaries of the factors identified by the classification system while keeping every summary traceable to its source sentence. Communication, teamworking and assessment emerged as the most recurrent factors. Four requirements shaped the design throughout: models trained on healthcare investigation data, offline operation, traceability and human verification.

The session includes a live demonstration of LENS.ai, a new tool that applies the same approach to coroners’ Prevention of Future Deaths reports, classifying contributory factors using the Yorkshire Contributory Factors Framework (YCFF) and supporting thematic analysis across reports.

Meet the speakers:

Georgina Cosma is Professor of Artificial Intelligence at Loughborough University. Her research focuses on AI, natural language processing and data science, with a particular interest in how AI can support healthcare and patient safety. She has worked with Professor Waterson over the last four years to develop AI methods and tools for analysing healthcare investigation reports, including the I-SIRch framework and the LENS.ai tool for coroners’ Prevention of Future Deaths reports.

Patrick Waterson is Professor of Human Factors and Complex Systems at Loughborough University. He has a long-standing interest in safety and accident investigation. He has worked with Professor Cosma in the last few years in the area of applying AI and machine learning to automatically analyse incident reports in healthcare.

Related

[related-posts-thumbnails]