Paranjay Sharma,
Research Member,
Indian Society of Artificial Intelligence and Law
Sameer Samal,
Nodal Advisor,
Indian Society of Artificial Intelligence and Law
Introduction
Cardiac Arrest is a sudden and unexpected loss of blood flow to the heart that results in loss of heart functions, breathing, and consciousness. This emergency requires quick medical attention involving CPR or defibrillators to restore a normal heartbeat, which is easily available at hospitals. However, these life-saving services are difficult to receive outside hospitals in medical emergencies. Therefore, to attend cardiac arrest patients, medical emergency response teams are increasingly developing and relying upon artificial intelligence technology. AI has the potential to dramatically improve the response time and the accuracy of detecting a possible cardiac arrest.
Project AI4EMS:
The European Emergency Number Association (EENA), a non-governmental organization, is working towards the objective of providing care to every European citizen in a medical emergency. In furtherance of the same objective, EENA partnered with Corti, an artificial intelligence company, to deploy AI support in medical emergency service operations. The project consisted of deploying the AI product supplied by Corti in medical emergency calls to detect out-of-hospital cardiac arrests in real-time. The Project was implemented in Italy with AREU emergency medical system, and in France with SAMU-SIDS 74 emergency medical services, to detect out-of-hospital cardiac arrests in real-time via emergency calls.
The artificial intelligence tool developed by Corti was deployed alongside dispatchers to assist them in making quick and accurate decisions. The tool was initially trained with country-specific language datasets to build language understanding for each country and then trained on datasets containing historical data on cardiac arrests, including both positive and negative cases of cardiac arrests. It analyses live medical emergency conversations to convert unstructured data into predictions. The project yielded promising results and by the end of the first training session in France, the number of undetected out-of-hospital cardiac arrest cases decreased by 5.5 percentage points, and in Italy, the number decreased by 3.9 percentage points as compared to an independent human call-taker.
Challenges
Certain challenges faced in the course of the project concerning metadata are:
The caller does not detect certain symptoms and provides incorrect information.
The conversation is chaotic and the dispatcher has difficulty in understanding the situation.
The events in the medical emergency are not in chronological order and therefore the cardiac arrest could take place after the emergency call.
The project in terms of data processing is governed by the EU General Data Protection Regulation and Article 9 establishes a legal basis that benefits this project. However, Article 35 of the EU GDPR mandates that where the Data Controller, having established that new technology is in existence, has to carry out Data Protection Impact Assessment (DPIA) prior to the processing of such personal data to ensure a risk-based approach for data protection in projects involving new technologies.
Conclusions
This artificial intelligence tool has the potential to decrease the response time in emergencies concerning out-of-hospital cardiac arrest cases by assisting human call-takers in decision making. The abovementioned results of the project AI4EMS demonstrate that this tool also has the power to speed up the detection of cardiac arrest cases over the phone. Considering that the accuracy and potential of the AI tool are based upon the quality and quantity of training data, it is expected that the tool will improve with increased training on quality data. It is also pertinent to note that to ensure the highest efficiency of this tool, it should be deployed alongside human dispatchers.