Corti model beats dispatchers on sensitivity; RCT shows no gain

AI system

System: Corti machine-learning call-analysis framework for out-of-hospital cardiac arrest recognition (the peer-reviewed papers describe it only as 'a machine learning framework'; the EENA report names Corti)
Model: Not assessed
Release: Not assessed
Application developer: Corti (Copenhagen, Denmark)
Role: Not assessed
Configuration: Not assessed
Inputs and outputs: Not assessed
Autonomy: Not assessed
Workflow: Access to recorded 112 call audio with linked outcome data (7,211 hours of Danish training data per the EENA report). Speech-recognition pipeline running in real time on live calls. A dispatch center whose baseline recognition was already high (90.5% in the control arm).
Human oversight: Not assessed
Relation to AI: The AI listened to live calls and alerted dispatchers. In the randomized trial its alerts were more sensitive than dispatchers alone (85.0% vs 77.5%). They did not significantly change dispatcher recognition (93.1% vs 90.5%, P = .15). The benefit is thus a demonstrated detection capability, not a verified change in any patient's course.
AI contribution: The AI listened to live calls and alerted dispatchers. In the randomized trial its alerts were more sensitive than dispatchers alone (85.0% vs 77.5%). They did not significantly change dispatcher recognition (93.1% vs 90.5%, P = .15). The benefit is thus a demonstrated detection capability, not a verified change in any patient's course.

Benefits

Outcome identifiers: Not assessed
Benefit: higher sensitivity than dispatchers (84.1% versus 72.5%)
Category: Improved Task Accuracy
Status: Not assessed
Beneficiary groups: People. Organizations.
Beneficiaries: Not assessed
Reach: Not assessed
Distribution and access: Not assessed
Duration: Not assessed
Expected?: Yes
Costs and harms: Not assessed

Measurement

Studies: Not assessed
Study design: Not assessed
Measurement: Not assessed
Setting and sample: Not assessed
Measures: higher sensitivity than dispatchers (84.1% versus 72.5%)
Comparison: Not assessed
Results: higher sensitivity than dispatchers (84.1% versus 72.5%)
Additional value from AI: Not assessed
Limitations: Not assessed
Denominators: Not assessed
Materials for replication: Not assessed

Money

Kind of value: Not assessed
Amount: Not assessed
Currency and prices: Not assessed
Period and unit: Not assessed
Realization: Not assessed
Recipients and payers: Not assessed
Gross or net: Not assessed
Included costs: Not assessed
Comparison: Not assessed
Method: Not assessed
Return and payback: Not assessed

People and places

Participants: Developer: Corti (Copenhagen, Denmark). Operator: Copenhagen Emergency Medical Services, Emergency Medical Dispatch Center Copenhagen.
Jurisdictions: Benefit: Copenhagen, Denmark. Developer: Denmark. Operator: Denmark.
Ownership and control: Not assessed
Dependencies: Not assessed
Cross-border connections: Not assessed
Geopolitical implications: Not assessed

Follow-up

Conditions for replication: Dispatchers act on alerts. The trial suggests alerts did not change their behavior enough to shift recognition. Call language and acoustics match the training data (EENA reports weaker performance on smaller French and Italian datasets).
Replication attempts: 1 follow-up entries. Benefit recurred: 1.
Later findings: 4 January 2021. The randomized trial of 169,049 calls confirmed the model's sensitivity advantage over dispatchers (85.0% vs 77.5%) but found no significant improvement in dispatcher recognition when alerts were shown (93.1% vs 90.5%).
Availability: Not assessed
Institutional decision: Not assessed
Grounds: Not assessed
Action and result: Not assessed
Success criterion: Not assessed

Evidence

Source numbers: AIBD-0030-source-0001–0003
Citations: Numbered references below.
Language: Not assessed
Source relationships: Operator, Developer
Original evidence: Not assessed
Access and versions: Not assessed
Supported findings: Not assessed
Evidence links: Not assessed
Evidence strength: Not assessed
Missing evidence: Not assessed

Grades

Assessed benefits: People. Organizations.
Depth: Not assessed
Reasons: People: The evidence is aggregate (sensitivity, time-to-recognition, trial-level recognition rates) with no individually attributable changed outcome. The randomized trial found no significant improvement in dispatcher recognition. Organizations: A capability study and a null randomized trial. No discrete organizational event with an observed outcome is documented.
Observed or intended: Not assessed
Rubric: Not assessed
Assessor and date: Not assessed
Alternative assessments: Not assessed
Fill: The deployer's account

Entry details

Kind

Not assessed

Related entries

Machine listening for emergencies to alert responders: Corti flags cardiac arrest on 112 calls, while ShotSpotter's acoustic classifier missed a Durham drive-by shooting, leaving police unalerted and five victims hospitalized.

Field

Emergency medical dispatch / public safety. 112 emergency call handling.

Purpose

The model was built to recognize out-of-hospital cardiac arrest during emergency calls so dispatchers can start telephone CPR sooner. The measured capability is that intended function, but no discrete changed outcome is documented.

Dates

Date basis unassessed: 2018 (precision unassessed). Updated: 21 October 2026.

Stage

Not assessed

Version

SHA-256: daf5403a07b4…

References

Other identifiers: AI Incident Database: 446.
Category mappings: Not assessed
Brief: Not assessed
Revisions: 6 revisions. Latest: 21 October 2026.

Sources

Peer-reviewed study Deployer

Blomberg SN, Christensen HC, Lippert F, et al. (2021). Effect of Machine Learning on Dispatcher Recognition of Out-of-Hospital Cardiac Arrest During Calls to Emergency Medical Services: A Randomized Clinical Trial. JAMA Network Open 4(1):e2032320.
Peer-reviewed study Deployer

Blomberg SN, Folke F, Ersboll AK, Christensen HC, Torp-Pedersen C, Sayre MR, Counts CR, Lippert FK (2019). Machine learning as a supportive tool to recognize cardiac arrest in emergency calls. Resuscitation 138:322-329. doi:10.1016/j.resuscitation.2019.01.015.
Report Developer

European Emergency Number Association and Corti (2020). Detecting out-of-hospital cardiac arrest using artificial intelligence: project report. EENA, Brussels.