Research Validates AI-Based Tool to Revolutionize Sepsis Care.

A study by the Department of Emergency Medicine at MedStar Washington Hospital Center, and others validates an AI-based tool that analyzes an mRNA signature to improve the specificity of sepsis diagnoses.

Sepsis, a life-threatening response to an infection, is a medical emergency that requires immediate treatment. Untreated sepsis can quickly cause severe organ damage and death. Yet diagnosis of sepsis is complicated—its symptoms overlap with other serious diseases, and it can arise from a range of different infections. 

MedStar Washington Hospital Center participated in a study of an AI-based blood test called TriVerity, which is designed to improve specificity and sensitivity in diagnosing acute infections. Our research, published in Nature Medicine, shows it can help identify sepsis, pinpoint its source, and even reveal the severity of the infection much sooner than conventional methods. 

At least 1.7 million adults in the U.S. develop sepsis each year. Sepsis can move fast, and current tools for diagnosis can be slow or imprecise. The breakthrough technology in our study uses AI and the body’s own genetic alarm signals to offer a fundamental shift in how blood infections are treated.

Why is sepsis difficult to diagnose?

Sepsis is the body’s extreme response to an illness—usually a bacterial, viral, or fungal infection. There isn’t a single germ doctors can look for, and symptoms such as shortness of breath, fever, and exhaustion could be signs of many different problems. 

Traditionally, when a patient arrives at the hospital with symptoms that could indicate sepsis, we incorporate vital signs along with tests such as white blood cell counts and C-reactive protein (CRP)  to assess for signs of inflammation that could signal infection. 

But these tests aren’t always very specific. Someone could have a high white blood cell count because they were just in a car accident, for example, or because they’re under extreme physical stress, or taking medications such as steroids. CRP is a protein produced by the liver that rises in response to bacterial sepsis, or arthritis, a major injury, or a severe cold.

Related: Read “Knowing the Signs and Symptoms of Sepsis Could Save Your Life.”

Listening to the body’s alarm.

A patient with bacterial sepsis (the most common cause) needs antibiotics immediately. But those drugs won’t help if a viral infection causes sepsis. Often, we need to wait for laboratory results to know for sure. In the meantime, we must make the best guess we can. 

The research validated in the study takes a different approach. Instead of spending time looking for a specific germ, TriVerity examines the patient’s immune response.

When the body fights an infection, it sends specific signals called messenger RNA (mRNA) to relay instructions  to the immune system. The test measures 29 mRNA markers in a single blood sample. By understanding how the body signals itself to fight, we can learn what it’s fighting and gain a sense of the opponent’s toughness.

Using an AI-driven algorithm to read the mRNA signals, TriVerity can distinguish between infections in about 30 minutes. 

The SEPSIS-SHIELD study.

To assess the test’s validity, our prospective study enrolled more than 1,200 adult patients from  diverse backgrounds. These patients, all suspected of having sepsis, were seen at 22 different emergency departments between 2020 and 2024. 

Researchers took a single blood sample from each patient to analyze their immune response using the TriVerity system. To examine the test’s accuracy, these results were compared with a “consensus adjudication,” a panel of doctors who reviewed every detail of each patient’s case.

The results were definitive, with the test outperforming other methods. Specifically:
​

• The test achieved high accuracy scores for identifying bacterial and viral infections. 

• It correctly identifies more than 95% of patients without a bacterial infection.

• The test provided a severity score that predicted which patients would require admission to the intensive care unit (ICU) or a ventilator within seven days.

This research can help us bridge the gap between general treatment protocols and the next generation of data-driven care.  

More personalized, equitable sepsis treatment.

No two people are alike, but many current medical standards use a one-size-fits-all approach to determine who is sick. This type of AI-powered technology enables us to move past universal standards and understand what’s normal for each unique patient based on their personal medical history and risk factors. 

Because this test can deliver results more quickly, we can start more specific treatment while the patient is still in the emergency department. This innovation enables us to replace guesswork with molecular certainty, sparing patients from unnecessary suffering.

For example, instead of prescribing antibiotics that might not be helpful for virus-related sepsis, we could use TriVerity to identify the optimal treatment for the underlying cause. Using antibiotics when they’re not needed can cause side effects such as diarrhea and the destruction of beneficial gut bacteria, and it can cause population-level problems such as drug-resistant “super-bugs.” This technology could help us reduce unnecessary antibiotic use by 60% to 70%, helping us protect both the patient and the community.

Related: Read “Study Demonstrates Costs of Antibiotic-Resistant UTIs and the Power of Partnership.”

Improving sepsis care for everyone.

While this study is an important step forward, it’s just one part of our work to improve sepsis care throughout the MedStar Health system. 

Soon, we’ll launch a pilot program to evaluate a new AI-based decision tool for sepsis. This could enable doctors to track data in real time, allowing us to spot changes in a patient’s condition up to six hours sooner than with older methods.

Sepsis can impact anyone. By using new tools to handle complex data and interpret warning signals, we can offer our patients better, more personalized care.