OA2018. Identifying product-related injuries using word embeddings holds potential for injury prevention
Year: 2025
Type: article
Abstract: Background Successful accident prevention requires high quality data and analyses that embrace all incidents. However, identifying all incidents of a given type of accident in hospital records is challenging due to the volume of data and the complicated nature of written human language. This project examines the implication on monitoring and preventing product-related injuries through text mining on hospital records by using keyword lists extended using word embeddings. This is exemplified by mountain biking and coin batteries injuries. Method Word embeddings were used to expand our manually defined keyword lists indicative for injuries involving mountain biking and create a keyword list indicative for coin battery injuries. Incidents were then identified using full-text search on hospital records from Odense University Hospital’s emergency department from 2017 to 2023. The hospital intake area covers 470.000 citizens. Results 1181 mountain biking incidents were identified using word embeddings compared to 506 using keyword extraction. The additional incidents included multiple different spellings of “mountain bike” such as “mountin-bike” or “mointain bike” that were otherwise missed by the keyword extraction. 74 incidents with coin batteries were identified and included both unintentional oral intake and intake related to self-harming behaviour. Discussion Word embeddings keywords is a feasible technique for extended keyword search on specific products in hospital records. More than twice the incidents involving mountain bikes were identified using word embeddings compared to keyword extraction. This is a considerable number of previously unknown incidents, and a detailed audit and review of the incidents regarding demography, circumstances, and severity may reveal a substantial potential for using word embeddings for injury surveillance. Key messages • Successful accident prevention requires high quality data and analyses that embrace all incidents. • Word embeddings keywords is a feasible technique for extended keyword search on specific products in hospital records. Topic Text mining, machine learning, injury surveillance.
Authors: K Kjaergaard, R Bank Lynggaard, L Elkjær Pedersen, P Just Vinholt, J Lauritsen
Institution: Odense University Hospital
Language: English
Field: Medicine
Domain: Health Sciences
Open Access status: gold
Estimated APC paid: $3,181
OpenAlex ID: W4416208836
FWCI:
Cites:
Cited by:
Related to: