Sepsis is a severe and deadly illness that can occur as a result of an infection. Unfortunately, it can be difficult to recognize sepsis in a patient due to subtle and ambiguous symptoms. Carilion Clinic has decided to adopt a new sepsis risk alert tool that uses predictive analytics to predict if a patient is at risk of having sepsis, and can then alert a clinical care team to improve recognition and treatment. While the decision to adopt the risk alert was made, what was unclear was how the interface and functionality of the tool and alert. In this study, we used focus groups, interviews, and a co-creation session to examine what providers and nurses wanted from a sepsis risk prediction tool.
Sepsis, which is a severe complication secondary to an infection, has approximately 1.7 million annual diagnoses in the United States, with an average mortality rate of 1 in 3 (Teng et al, 2020). Early detection of sepsis is challenging because its early symptoms can be non-specific, such as dyspnea and tachycardia. Due to its high burden in terms of mortality and cost, in addition to its non-specific early presentation, there has been significant interest in developing AI tools to help clinicians identify sepsis earlier. One such tool, the Epic Sepsis Model (ESM), is a proprietary sepsis prediction model that aims to identify patients who are at risk of developing sepsis and has been implemented in hundreds of hospitals in the United States (Epic, 2019; Wong et al, 2021). Carilion Clinic has an internal initiative to implement the ESM in May 2022. The overarching research question of this study is: how does the addition of a sepsis predictive tool affect clinician effectiveness and patient outcomes? Specifically, we hypothesize that implementation of the ESM will decrease the time it takes providers to recognize sepsis and improve patient mortality.
The effects of the ESM will be studied using an quasi-experimental model, in which patients admitted to Carilion prior to May 2022 will be compared with patients admitted after May 2022. With both groups’ data, the complete patient-level dataset will contain up to 5 years of retrospective and prospective data. The primary dependent variable will be the time it takes for clinicians to recognize sepsis. A secondary dependent variable will be patient mortality. To quantify the time it takes for clinicians to recognize sepsis, we will conduct manual chart reviews in order to calculate the initial time (T0) at which a patient develops using both objective and subjective criteria.