Speech Recognition Technology: Improving Speed and Accuracy of Emergency Medical Services Documentation to Protect Patients

Because hospital errors, such as mistakes in documentation, cause one sixth of the deaths each year in the United States, the accuracy of health records in the emergency medical services (EMS) must be improved. One possible solution is to incorporate speech recognition (SR) software into current tools used by EMS first responders. The purpose of this research was to determine if SR software could increase the efficiency and accuracy of EMS documentation to improve the safety for patients of EMS. An initial review of the literature on the performance of current SR software demonstrated that this software was not 99% accurate and therefore, errors in the medical documentation produced by the software could harm patients. The literature review also identified weaknesses of SR software that could be overcome so that the software would be accurate enough for use in EMS settings. These weaknesses included the inability to differentiate between similar phrases and the inability to filter out background noise. To find a solution, an analysis of natural language processing algorithms showed that the bag-of-words post processing algorithm has the ability to differentiate between similar phrases. This algorithm is the best suited for SR applications because it is simple yet effective compared to machine learning algorithms that required a large amount of training data. The findings suggested that if these weaknesses of current SR software are solved, then the software would potentially increase the efficiency and accuracy of EMS documentation. Further studies should integrate the bag-of-words post processing method into SR software and field test its accuracy in EMS settings.https://scholarscompass.vcu.edu/uresposters/1273/thumbnail.jp