Artificial Intelligence (AI) Support in Medical Emergency Calls (AISMEC)

Artificial Intelligence (AI) Support in Medical Emergency Calls - "The AISMEC-study" -Can Artificial Intelligence Improve the Precision in Identifying Acute Stroke in Emergency Medical Calls?

More than 12.000 patients suffer acute stroke in Norway every year, but less than half of them reach hospital within the current treatment window for thrombolysis. Stroke is the third-highest cause of death and the number one cause of severe disability requiring long time care at institutions. Consequently this has a high impact on society, patients and relatives, in addition to high costs related to care estimated to approximately 10 billion NOK per year. Although there are few studies on emergency medical communication centres (EMCC) in Norway, some have shown that the performance of the emergency medical communication centres can be improved. This project will seek to amend EMCC´s handling of acute stroke inquiries using artificial intelligence (AI), thus contributing to getting the patient to hospital in time for optimal treatments.

Study Overview

• Status: Active, not recruiting
• Conditions: Emergencies; Stroke, Acute; Apoplexy; Brain; Communication, Multidisciplinary
• Intervention / Treatment: Other: Artificial intelligence on emergency calls

Detailed Description

In this project, the investigators will collect data from all stroke patients discharged from Helse Bergen in 2019 (approx. 1000 patients) via the Norwegian Stroke Registry (NSR). For these patients, structured hospital data from Helse Bergen will be retrieved, and based on these and the spoken content of their emergency call regarding the stroke, the investigators will use machine learning to calculate the stroke risk. The connection of historical hospital data to the spoken words in the emergency call, amplifies the analysis of emergency calls in a novel way, in comparison to sound analysis alone.

After retrieving and connecting stroke patient data, the investigators train the deep network using data from 2019. Accordingly, testing will be performed based on patients from the first half of 2020. A separation of the data into training, test, and validation assures that our trained network does not over fit on the training data and can reproduce similar results on previously unseen patients. Finally, the investigators will compare the performance of the AI with the current system through statistical analyses on data from a period of approximately one year of live usage of the AI in AMK Bergen. This will enable us to evaluate to what degree the system is able to improve within the decision process of the EMCC operators in terms of sensitivity and specificity.

Summarized, the primary objective is to build a robust, working prototype of an AI system capable of real-time identification of acute stroke for improved assessment in emergency medical calls.

Our secondary objectives are:

• To implement an AI system capable of providing fast prediction of whether a patient is suffering from acute stroke or not based on audio from emergency call and available data sources within the hospital records
• To prove that AI systems can be used to assist and improve the triage decision procedure of the EMCC operator.

The anticipated result is to deliver fast (i.e. seconds) prediction scores to assist the EMCC operator in recognizing acute stroke patients, which provides an improved sensitivity and specificity compared to manual assessment only.

• Study Type: Observational
• Enrollment (Estimated): 1000

Contacts and Locations

Study Locations

• Norway
  • Bergen
    • Bergen, Bergen, Norway, 5021
      • Haukeland Universitetssykehus, Kirurgisk serviceklinikk, Nasjonalt kompetansesenter for helsetjenestens kommunikasjonsberedskap

Participation Criteria

Eligibility Criteria

• Ages Eligible for Study: 18 years and older (Adult, Older Adult)
• Accepts Healthy Volunteers: No

• Sampling Method: Probability Sample

Study Population

All calls to 113 in Bergen will be subject to the AI. A warning will be given the EMCC-operator if risk of stroke, when combining contents in the call and historical data in the hospital record, exceeds a certain limit.

Description

Inclusion Criteria:

• All callers to medical emergency number 113 in Bergen

Exclusion Criteria:

• Age <18

Study Plan

How is the study designed?

Design Details

• Observational Models: Case-Only
• Time Perspectives: Prospective

What is the study measuring?

Primary Outcome Measures

Outcome Measure	Measure Description	Time Frame
Stroke recognition in medical emergency calls	Survey AI's ability to recognize stroke, compared to the current system	Sept. 22 - Sept. 23

Collaborators and Investigators

• Sponsor: Haukeland University Hospital
• Collaborators: Oslo University Hospital; Helse Vest; Western Norway University of Applied Sciences; The Norwegian Heart and Lung Patient Organization; Helsetjenestens driftsorganisasjon for nødnett HF (HDO); The Norwegian Stroke Register
• Investigators: Study Director: Guttorm Brattebo, Professor II, Haukeland University Hospital

Study record dates

Study Major Dates

• Study Start (Actual): September 1, 2020
• Primary Completion (Estimated): August 31, 2026
• Study Completion (Estimated): August 31, 2026

Study Registration Dates

• First Submitted: November 23, 2020
• First Submitted That Met QC Criteria: November 23, 2020
• First Posted (Actual): December 1, 2020

Study Record Updates

• Last Update Posted (Estimated): September 22, 2025
• Last Update Submitted That Met QC Criteria: September 19, 2025
• Last Verified: March 1, 2025

More Information

Terms related to this study

• Keywords: Stroke; Artificial intelligence; Machine learning; Apoplexy; EMCC; Emergency Medical Service Communication Systems; Dispatch
• Additional Relevant MeSH Terms: Cerebrovascular Disorders; Brain Diseases; Central Nervous System Diseases; Nervous System Diseases; Vascular Diseases; Cardiovascular Diseases; Pathologic Processes; Disease Attributes; Pathological Conditions, Signs and Symptoms; Behavior; Stroke; Emergencies; Communication
• Other Study ID Numbers: 108573

Plan for Individual participant data (IPD)

• Plan to Share Individual Participant Data (IPD)?: NO
• IPD Plan Description: Not relevant at this stage.

Drug and device information, study documents

• Studies a U.S. FDA-regulated drug product: No
• Studies a U.S. FDA-regulated device product: No
• product manufactured in and exported from the U.S.: No