Sepsis Model Predictions: How Useful Are They Really in the Clinic

 

Sepsis Model Predictions: How Useful Are They Really in the Clinic

On February 26, 2024, researchers at the University of Michigan shed light on the limitations of the Epic Sepsis Model (ESM) in accurately predicting sepsis risk. Their study, featured in the New England Journal of Medicine (NEJM) AI this month, underscores the potential of artificial intelligence (AI) tools for enhancing sepsis treatment and outcomes.

Health systems across the U.S. have widely adopted the ESM, integrated within Epic's electronic health record (EHR) system. However, recent assessments have raised doubts about its accuracy and clinical usefulness. For instance, a 2021 study in JAMA Internal Medicine revealed a lower-than-expected identification rate of patients at risk of sepsis, contrasting with the tool's official information.

While the ESM outperformed other sepsis assessment tools in some aspects, its effectiveness in guiding timely clinical decisions remains uncertain. Clinicians often recognize and treat sepsis before patients meet formal criteria, which questions the practical value of models like the ESM.

To address this, the Michigan researchers analyzed data from over 77,000 adult inpatients at the University of Michigan Health from 2018 to 2020. Their evaluation of the ESM's predictive performance throughout patients' hospitalization stages revealed notable disparities. The model's ability to predict sepsis onset before meeting formal criteria was modest, with a significant drop in accuracy observed when predictions were confined to pre-criteria data.

These findings suggest that the ESM may inadvertently incorporate clinicians' suspicions about sepsis, potentially limiting its clinical utility. Understanding when the model is most effective within the clinical workflow is crucial for assessing its true value to healthcare providers.

While challenges persist with models like the ESM, ongoing research aims to refine sepsis prediction tools. For instance, a recent study from the University of California, San Diego, introduced an AI model named COMPOSER, which significantly reduced sepsis mortality rates in emergency departments. By leveraging real-time patient data, such as lab results and demographics, COMPOSER notifies care teams promptly, leading to improved patient outcomes since its implementation.

Read More:-