evidoria

← Back to browse

Good practice Imported

Capital Region of Denmark's AI-Assisted Emergency Call Triage — Corti's Speech-Recognition System for Cardiac Arrest and Stroke Detection

Denmark · Copenhagen · See the Denmark profile · See the Copenhagen profile

Evidence: Observational / pre–post Top 62% 53/100 · Ask Evidence Copilot about this practice

Copenhagen EMS deployed Corti, an AI speech-recognition assistant that flagged cardiac arrest from 161,650 calls with 95% accuracy vs. 73% for human dispatchers; a 2022 peer-reviewed study of 9,049 Capital Region calls projected a similar gain for stroke detection.

95 vs 73 %
Cardiac-arrest identification accuracy, Corti vs human dispatchers
161,650
Calls analysed in early evaluation
16 to 18 %
Modelled thrombolysis-window improvement (stroke) (2016-2018 retrospective)

Details

Maturity
Established
Promoter
Emergency Medical Services Copenhagen, Capital Region of Denmark (with Corti)
Period
2017–2022
Keywords
emergency medical services, public health, artificial intelligence, speech recognition, emergency dispatch

Context

Copenhagen Emergency Medical Services began using Corti, an AI speech-recognition and machine-learning assistant, to listen live to 112 emergency calls and flag likely cardiac arrests to dispatchers in real time.

Activities

Corti analyses cues including keywords and abnormal breathing patterns during live calls; a 2022 peer-reviewed retrospective study separately modelled extending the approach to stroke detection using 9,049 calls from the Capital Region of Denmark (2016-2018).

Results

An early evaluation of 161,650 calls found Corti correctly identified cardiac arrest in 95% of cases versus 73% for human dispatchers alone. The stroke study modelled a rise in patients receiving thrombolysis within the critical window from roughly 16% to 18%.

Conclusions

The 161,650-call cardiac-arrest study had not been peer-reviewed in detail at the time, leaving the false-positive rate undocumented; researchers noted the AI 'cannot explain how it makes decisions', an acknowledged transparency limitation, and the stroke figure is a projection assuming performance parity with less-specific stroke symptoms.

Implementation
Indicative cost
Medium (€50k–€500k) — Not disclosed.
Time to results
Long (> 3 years) — Deployed for cardiac-arrest detection from around 2016-2018; stroke extension modelled in a 2022 peer-reviewed study.
Staffing & skills
Copenhagen EMS dispatchers, Corti (vendor)

Conditions for success

  • AI flags risk for dispatcher review rather than replacing dispatcher judgement

Common failure modes

  • Full cardiac-arrest study not yet peer-reviewed in detail; false-positive rate undocumented
  • AI cannot explain its decisions, an acknowledged transparency limitation
  • Stroke symptoms are less specific (5-14.9% 'stroke mimics' vs 0.6% for cardiac arrest), so the stroke projection may not hold in practice

Commonly funded by

National / regional programmes

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Similar practices you may find useful