Learning the magnitude and duration of influence of infections

Abstract

Clostridioides difficile infections (CDIs) impose a substantial burden on the healthcare system leading to poor health out comes, mortality and costs to the heath-care system estimated at greater than $5 billion. One of the reasons why CDIs are hard to control is the contribution of individual infections to the risk of transmission is not well understood. In this paper, we propose modeling incident infections using a Hawkes process, which is a self-exciting stochastic process, encoding the intuition that new infections trigger further infections. Using data from a large urban hospital, we demonstrate that our approach reveals different patterns of infection spread across patient care units. These insights can be used to guide unit-specific interventions aimed at interrupting transmission.