Copenhagen study: ML beats dispatchers at stroke recognition

A machine-learning pipeline trained on six years of Copenhagen medical-helpline calls flagged 63.0% of stroke cases on a held-out 2021 test set, against 52.7% for the human call-takers handling the same calls, according to a retrospective study in npj Digital Medicine.

The researchers used calls from 1 January 2015 to 31 December 2020 to train an automatic speech recognition model paired with a text classifier, then evaluated on 2021 calls. On that test set, call-takers reached a sensitivity of 52.7% (95% CI 49.2–56.4%) with a positive predictive value of 17.1%. The framework, the paper reports, performed "significantly better (p < 0.0001) with a sensitivity of 63.0% (62.0–64.1%) and a PPV of 24.9% (24.3–25.5%)."

The data comes from Copenhagen Emergency Medical Services, with authors drawn from Copenhagen University Hospital, the Technical University of Denmark, and Corti, a Danish company that sells AI tooling to emergency dispatch services. The pipeline classifies calls after transcription, and the study is retrospective: no live deployment alongside call-takers is reported.

The authors frame the system as a "supportive tool for call-takers, aiding in early and accurate stroke recognition," not a replacement. The paper does not report end-to-end latency, nor how the classifier handles non-Danish speakers or calls where the patient cannot speak for themselves.