William Trevena ¹ Xiang Zhong ^1* Amos Lal ² Lucrezia Rovati ² Edin Cubro ² Yue Dong ² Phillip Schulte ² Ognjen Gajic ²

• ¹ University of Florida, Gainesville, Florida, United States
• ² Mayo Clinic, Rochester, Minnesota, United States

Digital twins of patients are virtual models that can create a digital patient replica to test clinical interventions in silico without exposing real patients to risk. With the increasing availability of electronic health records and sensor-derived patient data, digital twins offer significant potential for applications in the healthcare sector. This article presents a scalable full-stack architecture for a patient simulation application driven by graph-based models. This patient simulation application enables medical practitioners and trainees to simulate the trajectory of critically ill patients with sepsis. Directed acyclic graphs are utilized to model the complex underlying causal pathways that focus on the physiological interactions and medication effects relevant to the first 6 hours of critical illness. To realize the sepsis patient simulation at scale, we propose an application architecture with three core components, a cross-platform front-end application that clinicians and trainees use to run the simulation, a simulation engine hosted in the cloud on a serverless function that performs all of the computations, and a graph database that hosts the graph model utilized by the simulation engine to determine the progression of each simulation. The proposed patient simulation application could help train future generations of healthcare professionals and could be used to facilitate clinicians' bedside decision-making.