Application of Artificial Intelligence in Rapid Response System

초록

In Republic of Korea, the incidence of in-hospital cardiac arrest (IHCA) was reported to be 2.46 cases per 1,000 admissions. The survival rate was under 24%. The need for a rapid response system (RRS) has been highlighted for proactive prevention of IHCA. There is a perspective that most studies have failed to evaluate the effectiveness of RRS because track and triggering system (TTS) of RRS has not been efficient. However, a few studies for AI-based TTSs were demonstrated superior performances. In eCART study by Christopher J et al., the mortality of the patients admitted in general ward was significantly reduced from 13.9% to 8.8%. The eCART classified patients based on their risk level, alerting physicians to consider ICU transfer for high-risk patients and nurse-led protocols for intermediate-risk patients, such as checking vital signs every two hours. These faster ICU transfer and timely reassessment could lead to improved outcomes. The study of Gabriel J et al. showed remote-monitoring nurses with AI-driven system using electronic medical records effectively identified high-risk patients. Among high-risk patients, the remote-monitoring nurses notified to the hospitals RRS was assigned to intervention cohort. The others were assigned control cohort. The hospitals using this system had a significantly lower 30-day mortality (RR 0.84) in intervention cohort. The study by Cho et at. compared the predictive accuracy of AI-based TTS with NEWS and MEWS. The AI-based TTS demonstrated superior performance (AUROC 0.869 vs. 0.756/0.767) in prediction in IHCA and unplanned ICU transfer. Even AI-based TTS performance is demonstrated to be effective, several conditions were considered for AI-based TTS to become more useful. It cannot completely replace clinical judgment by physician. False positive alerts should be reduced further for preventing fatigue. Active participation and additional training of medical staffs is required for effective operation of AI-b