Sepsis
Conditions
Keywords
Infection, Emergency Department, Algorithm, Prediction
Brief summary
Sepsis is a severe response to infection resulting in organ dysfunction and often leading to death. More than 1.5 million people get sepsis every year in the U.S., and 270,000 Americans die from sepsis annually. Delays in the diagnosis of sepsis lead to increased mortality. Several clinical decision support algorithms exist for the early identification of sepsis. The research team will compare the performance of three sepsis prediction algorithms to identify the algorithm that is most accurate and clinically actionable. The algorithms will run in the background of the electronic health record (EHR) and the predictions will not be revealed to patients or clinical staff. In this current evaluation study, the algorithms will not affect any part of a patient's care. The algorithms will be deployed across the Emory healthcare system on data from all patients presenting to the emergency department.
Detailed description
The primary goal of this study is to prospectively evaluate three sepsis prediction algorithms that are embedded in the EHR. The models will be deployed in a shadow mode, and the results will not be displayed to the treatment team during this study. Two of the algorithms are proprietary algorithms of the EHR provider (Epic). The third algorithm is an internally developed, open-source algorithm. The algorithms will compute the probability of sepsis at periodic intervals and will continue to run on a patient's data until the patient's discharge, death, or upon initiation of intravenous antibiotics (at which point there is an indirect record of clinical suspicion of an infection).
Interventions
OTHEREmory Sepsis Model
Emory internal algorithm
OTHEREpic Sepsis Model Version - 1
The Epic Sepsis Model (ESM) version 1, a proprietary sepsis prediction model.
OTHEREpic Sepsis Model Version - 2
The Epic Sepsis Model (ESM) version 2, a proprietary sepsis prediction model.
Sponsors
Emory University
Study design
Observational model
COHORT
Time perspective
PROSPECTIVE
Eligibility
Sex/Gender
ALL
Age
18 Years to No maximum
Healthy volunteers
No
Inclusion criteria
* All adult patients admitted through the ED
Exclusion criteria
* None
Design outcomes
Primary
| Measure | Time frame | Description |
|---|---|---|
| Patient hospitalization-level area under curve (AUC) for identification of sepsis, | Duration of hospital stay (until discharge or death), an expected average of 30 days | Definition of Sepsis using the Centers for Disease Control and Prevention (CDC) Adult Sepsis Surveillance. |
Secondary
| Measure | Time frame | Description |
|---|---|---|
| Lead time to antibiotic administration | Duration of hospital stay (until discharge or death), an expected average of 30 days | The time between the initial deployment of the alert in patients confirmed to have sepsis (ture positives) and the physician's ordering of intravenous antibiotic therapy. |
| Percent expected increase in unnecessary antibiotics | Duration of hospital stay (until discharge or death), an expected average of 30 days | Percent of patients who were incorrectly identified as having sepsis (false positives), and received antibiotics. |
| Number needed to screen | Duration of hospital stay (or death), an expected average of 30 days | The number of alerts that would need to be processed to find one true positive sepsis. |
| Sensitivity, specificity, and Positive and Negative Predictive Value of algorithms | Duration of hospital stay (until discharge or death), an expected average of 30 days | Sensitivity, specificity, positive predictive value (PPV), and negative predictive value (NPV). |
| Time-horizon based AUCs | 4 hours, 8 hours, and 24 hours | AUCs will be calculated at 3 pre-specified time horizons. |
| Accuracy and calibration by subgroup | Duration of hospital stay (until discharge or death), an expected average of 30 days | The AUC and calibration curves will be compared by sex and race to ensure predictive accuracy is equal across subgroups. |
| Number of Total and false alert burden | Duration of hospital stay (until discharge or death), an expected average of 30 days | The number of Total and false alert burden cumulative across all study patients over the study period |
Countries
United States
Contacts
Primary ContactSivasubramanium Bhavani, MD
Outcome results
None listed