The Department of Emergency Medicine at National Taiwan University Hospital (hereafter 'NTU Hospital') and the University's research unit AINTU have successfully developed 13 AI models for six ED procedures, using nearly ten years and over one million sources of data to update the status of each emergency patient and AI recommendations every 15 minutes. Nearly 10,000 preliminary tests have been completed, and according to the feedback of 206 senior doctors who participated in the tests, the accuracy of AI assistance reached 70%.
The hospital and the research team worked to have established an innovative clinical process for intelligent emergency care. According to Ming-Hsiang Wu, Superintendent of NTU Hospital, in the past, when people went to the hospital and faced doctors with different experience, they received different diagnosis or treatments, but artificial intelligence (AI) can help make medical treatment more efficient, truly improve patient safety and enhance the quality of care.
Accoring to the hospital, using 10 years of retrospective data from NTUH, its Yunlin Branch and Hsinchu Branch, about 1.25 million data sets were collected, and that AI could be used to assist in six key aspects of the emergency medical process, including emergency medical examination and injury, medical history analysis, immediate risk classification and identification, early and appropriate safe exit from the department, prognosis analysis of cardiac arrest events, and early high-risk warning of patient bracelets.
NTUH further indicated that patients need to go through many hurdles after arriving at the emergency department, and each hurdle requires a decision on what to do at the next step; with the help of AI, it can help speed up the correct diagnosis, and it is estimated that each step can be accelerated by 20%, allowing patients who should be discharged to be discharged early and those who should be hospitalised to be hospitalised as soon as possible.
One of the AI models is the "intelligent neurological prognosis assessment for emergency resuscitation". When a patient is brought to the emergency room after a cardiac arrest, numerous studies have pointed out that the neurological prognosis of a cardiac arrest patient is related to the brain's grey-to-white ratio, explained NTUH.
The most important question a patient wants to know in the emergency room is "am I going to be hospitalised or can I go home? In the past, doctors had to rely on their experience to determine whether a patient was in danger, whether he or she needed to be hospitalised, and whether there would be any problems after going home. With AI models in place, the accuracy of level of risk achieves 70% as per 206 senior clinicians doctors.
