We recommend that you upgrade to the latest version of your browser.
Visual identificator for the AISMEC project

Artificial Intelligence Support in Medical Emergency Calls (AISMEC)

AISMEC is a research project aimed at developing decision support using artificial intelligence (AI) for operators at the medical emergency number (113). By analyzing what is said during emergency calls and combining it with the patient’s hospital history, medication use, and blood test results, the AI model can estimate the risk of stroke. This provides operators with valuable support in critical decision-making.

Background

AISMEC was established in 2020 on the initiative of Guttorm Brattebø, head of KoKom (Norwegian National Advisory Unit on Emergency Medical Communication). Insight into the highly demanding task of answering emergency calls led to the project’s creation, with the goal of strengthening operators’ decision-making in challenging situations. Initially, the focus is on stroke assessment, but the same methodology could be applied to other conditions such as chest pain or breathing difficulties.

The project is a collaboration between several public institutions and has received funding from the Research Council of Norway. All development is rooted in and tested through scientific methods. There are no private interests involved, and profit is not a consideration.

Project Structure

The project is divided into three phases:

  1. Data Collection: Gathering background information and relevant data for AI model development. Analyses of the data will result in two research articles, one of which has been published.
  2. Model Development: Training the AI model on historical patient data.
  3. Model Testing: Evaluating the AI model on a new set of historical emergency calls (a large number of callers to 113 at AMK Bergen).

    If successful, real-time testing (initially in shadow mode) may follow as a separate research project.

Speech-to-Text

Based on the Norwegian National Library’s Whisper model, a speech-to-text system has been trained to handle transcription of emergency calls. This is necessary because part of the AI model involves text analysis of what is said during the call. Emergency calls differ significantly from regular phone conversations, containing uncommon words (e.g., AMK, addresses, names, medical terms), as well as noise, emotional expressions, crying, shouting, etc. A standard speech-to-text model does not perform well enough. The custom-trained model is completed and tested, and performs better than standard models. An article about the model is planned for publication.

Stroke and Thrombectomy

Some stroke patients may be eligible for thrombectomy (“clot fishing”). These patients’ journey through the emergency system is particularly interesting, as routing them directly to a hospital that performs thrombectomy—rather than via a local hospital—could significantly improve outcomes. A subproject has reviewed emergency calls from patients who underwent thrombectomy in 2019 and 2021 to identify common features. The findings are described in an article under review for publication in an international research journal.

AI analysis to investigate whether patients eligible for thrombectomy can be identified during the emergency call

We are in the early stages of exploring whether AI can be used to detect patients eligible for thrombectomy already during the emergency call. Early identification would have major positive consequences for patients, outcomes, the prehospital system, and hospitals.

The project is approved by the Regional Committees for Medical and Health Research Ethics (REK Vest) – approval number: 108573. We maintain ongoing dialogue with REK Vest regarding changes to the project or team. The data protection officers at Helse Bergen and Haraldsplass Diakonale Sykehus have also approved the project (eProtocol number: 1612-1612). All patients whose data has been used have received letters about the project with the opportunity to opt out (passive consent), in accordance with REK Vest’s decision.

  • Emergency call data from Bergen 113-central (AMK Bergen)
  • Hospital records from Haukeland University Hospital and Haraldsplass Diakonale Sykehus (previous illnesses, procedures, and admissions relevant to stroke risk)
  • Prescription data from the Norwegian Prescription Database (relevant to stroke risk)
  • Blood test data from Haukeland University Hospital (relevant to stroke risk)

  • Anders Strand Vestbø – R&D Department, Helse Bergen
  • Lars Myrmel – Emergency Medicine Department, Helse Bergen
  • Alexander Lundervold – Department of Computer Science, Electrical Engineering and Mathematics, Western Norway University of Applied Sciences
  • Christian Autenried – AI Team, Helse Vest IKT
  • Cosimo Damiano Persia – AI Team, Helse Vest IKT
  • Hege Ihle-Hansen – Vestre Viken HF / Oslo University Hospital

  • Merete Landaas – LHL Hjerneslag og Afasi
  • Geir Sverre Braut – Professor / seniorrådgiver, Stavanger Universitetssykehus
  • Bjørn Jamtli – Seniorrådgiver, Helsedirektoratet
  • Øyvind Østerås – avdelingssjef KSK
  • Hanne Klausen (leder) – Klinikkdirektør KSK
  • Regional User Committee, Helse Vest (new representative to be appointed)

AISMEC has received support from the Laerdal Foundation in addition to funding from the Research Council of Norway.

Contact

Emil K. Iversen

PhD Candidate / Physician, Norwegian National Advisory Unit on Emergency Medical Communication

emil.iversen@helse-bergen.no
Last updated 8/6/2025