A Framework to Tackle Risk Identification and Presentation Challenges in Sepsis

Introduction: Sepsis trajectories, including onset and recovery, can be difficult to assess, but electronic health records (EHRs) can accurately capture sepsis as a dynamic episode. Methods: Retrospective dataset of 276,722 clinical observations (4,726 unique patients) during a two-month period in 2015 were extracted from the EHRs. A Cox proportional hazard model was built to test hazard ratios of risk factors to the first sepsis episode onset within 72 hours for patients with presumed infection. Predisposition, infection, response, and organ failure (PIRO) score-based framework was used in a logistic regression to identify factors associated with in-hospital mortality within the sepsis population. Results: 47.54% of patients with an infection episode experienced at least one sepsis episode (N=1,044 out of 2,196) within 72 hours of admission. The mortality rate was higher for patients with sepsis episodes (7.24%) compared to patient with only organ dysfunction episodes (4.84%) or only with infection episodes (3.96%). Analysis identified factors associated with the first sepsis episode onset and those associated with in-hospital mortality. Discussion: Our study addresses identification of infection, organ dysfunction, and sepsis as dynamic episodes utilizing EHR data and provides a systematic approach to detect risk factors related to sepsis onset and in-hospital mortality.