Evaluating the usefulness of a chat bot as an assistant during CPR care by laypersons.

Twenty-one university graduates and university students naive in basic life support participated in this quasi-experimental simulation pilot trial. A version beta chatbot was designed to guide potential bystanders who need help in caring for cardiac arrest victims. Through a Question-Answering (Q&A) flowchart, the chatbot uses Voice Recognition Techniques to transform the user's audio into text. After the transformation, it generates the answer to provide the necessary help through machine and deep learning algorithms. A simulation test with a Laerdal Little Anne manikin was performed. Participants initiated the chatbot, which guided them through the recognition of a cardiac arrest event. After recognizing the cardiac arrest, the chatbot indicated the start of chest compressions for 2 min. Evaluation of the cardiac arrest recognition sequence was done via a checklist and the quality of CPR was collected with the Laerdal Instructor App.

91% of participants were able to perform the entire sequence correctly. All participants checked the safety of the scene and made sure to call 112. 62% place their hands on the correct compression point. A media time of 158 s (IQR: 146–189) was needed for the whole process. 33% of participants achieved high-quality CPR with a median of 60% in QCPR (IQR: 9–86). Compression depth had a median of 42 mm (IQR: 33–53) and compression rate had a median of 100 compressions/min (IQR: 97–100).

The use of a voice assistant could be useful for people with no previous training to perform de out-of-hospital cardiac arrest recognition sequence. Chatbot was able to guide all participants to call 112 and to perform continuous chest compressions. The first version of the chatbot for potential bystanders naive in basic life support needs to be further developed to reduce response times and be more effective in giving feedback on chest compressions.