Aim The chance of surviving Out of Hospital Cardiac Arrest (OHCA) is highly correlated with medical dispatchers’ recognition of the condition during emergency calls. The median sensitivity for OHCA recognition across studies is around 70%.(1) This leaves room for improvement. A novel approach is to improve recognition of OHCA by applying Machine Learning directly on the call-dialogue. The aim of the study is to investigate if recognition can be increased by use of Machine Learning.

Methods Our study used 489 emergency calls regarding OHCA received at the Emergency medical Dispatch Centre Copenhagen (EMDC) in 2013 and a control group of 571 non-OHCA calls. All calls were transcribed and then divided into two datasets, one for training the machine learning model (275 OHCA-calls, 361 non-OHCA calls), and one for testing the model. The Machine Learning model automatically detected patterns and predictive word contexts in relation to OHCA/non-OHCA. The model identified words associated with OHCA, and was able to determine whether a call was regarded as OHCA.

Results The Machine Learning model reached a sensitivity of 95.3% on 214 transcribed OHCA-calls (204 true positive/10 false negative) and a control group of 210 non-OHCA calls. Specificity for the Machine Learning model was 99.0%.

Conclusion These early results show that a Machine Learning model based on neural networks has potential to improve recognition of OHCA. The results are thought-provoking and invites to further research.

Reference

1. Viereck S, Moller TP, Rothman JP, Folke F, Lippert F. Recognition of out-of-hospital cardiac arrest during emergency calls – a systematic review of observational studies. 2017. Ref Type: Submitted article in review

Conflict of interest None declared.

Funding Emergency Medical Services Copenhagen has received an unrestricted research grant from the Laerdal Foundation.