Retrospective study on early prediction of major adverse events in general ward with AI algorithm

Major adverse events (MAEs) include cardiac arrest, intubation, and transfer to ICU directly affect to the outcome of patients and rapid response to those MAEs is crucial. Therefore, AI-based clinical decision support software to predict the onset of MAEs in advance can be helpful. This study conducts externally validate the performance of AITRICS-VC (Mindong Sung. et al. JMIR Med Inform 2021;9(11):e26426) to predict MAEs onset in advance. In this study, we conducted a retrospective, multi-center study at three tertiary hospitals in South Korea, focusing on patients admitted to general wards from 2020 to 2022. The data were extracted from the electronic medical records and input into AITRICS-VC to make alert of MAEs. We evaluated the predictive accuracy of AITRICS-VC within 6 hours before onset of MAEs. Additionally, we compared the accuracy with conventional clinical early warning score NEWS and MEWS. In this study, we included all adult patients in general ward and excluded patients with COVID-19. The total cohort included 250,919 patients (411,397 admissions). The study unveiled significant difference of performance for MAEs prediction between AITRICS-VC and others. Remarkably, AITRICS-VC exhibited exceptional predictive accuracy with an AUROC of 0.8546, surpassing NEWS (0.7396), and MEWS (0.6348). In conclusion, the study affirms the clinical efficacy of an AI-based MAEs prediction software, outperforming traditional clinical assessment tools. Further prospective studies in real-world clinical settings are called for or required to validate its effectiveness and explore the tangible benefits of timely interventions facilitated by this innovative software.