ML beats Copenhagen call-takers at stroke recognition

A machine-learning pipeline scored 63.0% sensitivity on recognising stroke during Copenhagen medical helpline calls, against 52.7% for the human call-takers handling the same lines, according to a retrospective study in npj Digital Medicine. Positive predictive value also rose, from 17.1% to 24.9%, with the authors reporting p < 0.0001 across the metrics.

The system transcribes each call with automatic speech recognition, then runs a text classifier that labels it stroke or non-stroke. Training data came from calls placed between 1 January 2015 and 31 December 2020; 2021 calls were held back for testing.

The authors frame the model as an assistant, not a replacement. Their conclusion: "a machine learning framework for recognising stroke in prehospital medical helpline calls may become a supportive tool for call-takers, aiding in early and accurate stroke recognition."

The abstract does not report total call counts, a split between 1-1-2 and medical helpline 1813 traffic, or any downstream effect on time-to-treatment.