Copenhagen ML model beats call-takers at stroke triage
TL;DR
- A machine-learning pipeline hit 63.0% sensitivity for stroke on Copenhagen medical helpline calls, versus 52.7% for human call-takers (p < 0.0001).
- Positive predictive value rose from 17.1% for call-takers to 24.9% for the model on 2021 test calls.
- The system transcribes audio with automatic speech recognition, then classifies the text with an ensemble of five multi-layer perceptrons.
A machine-learning pipeline run on Copenhagen's medical helpline recognised stroke in 63.0% of calls from stroke patients, versus 52.7% for human call-takers, according to a retrospective study in npj Digital Medicine.
The system first transcribes calls with an automatic speech recognition model, then classifies the text as stroke or non-stroke using an ensemble of five multi-layer perceptrons over bag-of-words features. Positive predictive value climbed from 17.1% for call-takers to 24.9% for the model, with p < 0.0001. Calls from 1 January 2015 to 31 December 2020 were used to develop the pipeline; 2021 calls were held out for testing.
The authors frame the output as "a supportive tool for call-takers, aiding in early and accurate stroke recognition," not a replacement. Lead author Jonathan Wenstrup is based at the Departments of Neurology and Emergency Medical Services in Copenhagen.
Shared on Bluesky by 1 AI expert
Originally reported by nature.com
Read the original article →Original headline: A retrospective study on machine learning-assisted stroke recognition for medical helpline calls - npj Digital Medicine