The AISMEC (Artificial Intelligence Support in Medical Emergency Calls) project aims to develop AI-based decision support for operators handling medical emergency calls. Acute stroke is the condition chosen as proof of concept. We know there is a potential for improvement in the emergency medical communication centres (EMCC, AMK) within today`s system. Further, we assume that there is a potential for improving the local medical emergency call centres (legevaktsentral, LVS) as well, and we aim eventually to introduce the concept there as well.
In Copenhagen EMCC, AI-analysis of sound/voice in emergency calls has shown to out-perform operators in detecting cardiac arrest. We will take this a step further and combine analysis of what the caller says about the patient with the patient's medication and data from previous hospital records, if such exist. Based on these data sources, the AI will calculate the probability for the patient suffering a stroke, and notify the operator.
The project is based on approx. 1000 stroke patients in Helse Bergen during 2018/2019, with an analysis of their contacts with the medical emergency services when the stroke occurred. We will use this combined with 113 calls and the hospital data, to develop the concept.
The project is a collaboration between a number of actors within the disciplines of medical emergency services, AI/machine learning and stroke treatment, in addition to one user organization. The following are involved: Haukeland University Hospital (National Centre on Emergency Communication in Health (KoKom), project management, Department of Neurology, Department of Emergency Medicine, Section for Healthcare Services development), Helse Vest IKT, (Western Norway University of Applied Sciences, MMIV (Mohn Medical Imaging and Visualization Centre), Oslo University Hospital, The Norwegian Stroke Register, LHL Hjerneslag and the Health service organization for emergency network HF.
There is a number of medical conditions such as stroke, acute coronary syndrome, respiratory failure, cardiac arrest, and trauma, also called ‘The First Hour Quintet’, that require effective therapeutic treatment within a certain time limit. Stroke alone makes the third-highest cause of death and the number one cause of severe disability requiring long-time care at institutions.
Every year approx. 12 000 patients in Norway suffer acute stroke, and only half of them reach hospital within the first critical time-frame. The defining role in this process plays an early and accurate symptom recognition by the emergency medical communication centres (EMCC) during a live call. Today, the diagnostic accuracy of the norwegian EMCC operators is limited by a paper-based triage system (the Norwegian Index for Emergency Medical Assistance). As a result, only 60% of the acute stroke patients receive correct diagnosis and subsequent medical treatment. This leads to a vast personal, societal, and economic costs estimated to approximately 10 billion NOK per year.
We believe that urgent change is needed to address this issue. As a measure, we propose an innovative AI-based decision support system for emergency call takers, helping to identify critical medical emergency conditions (e.g. stroke) in real-time. This support system will utilize a research-based approach to combine structured and relevant patient data with speech recognition during the live emergency medical calls. This technology will help improve the quality and reduce variability in the emergency medical services.
We will compare the new system’s ability for disease prediction with the current practice as well as with traditional statistical approaches to risk modelling, for proving effectiveness. The system will first be developed and tested on stroke, as the first case study. A successful solution will then be adapted and implemented to other relevant medical emergencies.