Presenter:
Aaron Scott

Abstract
Sepsis is one of the leading causes of mortality in the world. Currently, the heterogeneity of sepsis makes it challenging to dynamically stratify patients on admission to the hospital. Here, we used a population scale cohort of 1974 individual patients to develop a digital twin-based modelling strategy that circumvents the need for rule-based cutoffs and defined phenotypes for patient stratification. Using only clinical parameters available at time-of-admission the model can predict the development of sepsis, future deterioration, infection, and 30-day mortality with high accuracy. To determine the unique molecular signatures associated with sepsis we analyzed the plasma proteome of all 1974 patients. We augmented the clinical data with the proteome to create multimodal predictive digital twin models that uncover the molecular context associated with the diverse clinical presentations of sepsis. In addition to being highly effective for investigating sepsis, this approach has the potential to generalize to other diseases and to provide clinically actionable results that drive treatment.