Genomics

Triaging sepsis effectively: Gene expression meets machine learning 

A surgeon's frustration with diagnostic limitations inspired a new technology to harness the information from the immune system to combat a public health emergency – sepsis.

When Tim Sweeney, MD, PhD, was a medical resident in 2011, he was excited about beginning his career as a doctor. He loved the adrenaline rush of working in the emergency department. But he quickly became frustrated by a common problem doctors face worldwide – the inability to differentiate patients who were recovering from those whose health was deteriorating.

He often encountered patients with a high fever. But he had no reliable tools at hand to determine whether the fever was caused by an infection, or was simply an immune response to a medical or surgical procedure.  

If he suspected an infection, hours would pass, waiting for the results of blood cultures and other tests, Sweeney recalls. In some cases, by the time a diagnosis was confirmed, it was too late – the patient had already progressed to sepsis.  

Sepsis – when the immune system reacts abnormally to a pathogen or injury – is one of the most frequent causes of death worldwide, according to the World Health Organization – representing 20% of all deaths, half of which are children under the age of five. In hospitals, it is a leading cause of mortality, with the risk of death increasing by up to 8% for every hour treatment is delayed.  

The condition can mimic other illnesses, and its symptoms are often nonspecific, such as fever, rapid heart rate, and confusion. Traditional diagnostic methods are time-consuming and often inconclusive.

Sepsis, a life-threatening condition caused by the body’s extreme response to infection, is a leading cause of death worldwide and the most expensive diagnosis covered by Medicare. Tim Sweeney, MD, PhD, co-founder and CEO of Inflammatix, introduces a new device that detects inflammation - both viral and bacterial - and assesses a patient’s risk of developing sepsis. What does this mean for future treatment plans?
Sepsis is a public health emergency. It precedes half of all inpatient deaths. It's also the number one source of malpractice claims and injuries to patients and misdiagnosis.
Tim Sweeney, MD, PhD, co-founder and CEO of Inflammatix

Differentiating between non-infectious inflammation and sepsis

When patients arrive in the emergency department, physicians first evaluate their symptoms. The triage processes involve measuring a patient’s vital signs, such as temperature, heart rate, blood pressure. Clinicians then carry out a detailed physical exam to identify potential sources of infection, such as wounds, or surgical sites. They try to assess whether their condition will improve or deteriorate and the risk of sepsis.

In addition, blood tests can detect abnormal white blood cell counts, which can indicate an infection. If an infection by a pathogen is suspected, physicians can order a culture of blood, urine, sputum or a wound to identify the specific bacteria or fungi causing an infection, although it can take up to 72 hours to obtain results.

Other tests, such as measuring lactate levels, can indicate that tissues aren’t getting enough oxygen, a hallmark of sepsis, while tests that measure C-reactive protein and procalcitonin can help to identify an inflammatory response.

But such tests are not very accurate, says Sweeney. “Many of the tests that hospitals use are decades old.” The lack of accuracy does not only put patients at risk, but leads to a lot of wasted resources in the health care system, Sweeney adds.
Tim Sweeney, M.D., Ph.D.
The TriVerity test measures mRNA expression of 29 genes from a blood sample. The test then applies machine learning algorithms to also analyze their expression patterns and to generate three results: (1) the likelihood of a bacterial infection, (2) the likelihood of a viral infection, and (3) the likelihood of the patient progressing to requiring ICU-level care in the next 7 days.
The TriVerity test is a blood test that looks at 29 mRNAs from the patient's immune system in blood… And from that we have algorithms on the device that estimate the risk of a bacterial infection.
Tim Sweeney, MD, PhD, co-founder and CEO of Inflammatix

Connecting surgery, genes and big data

During his surgical residency, Sweeney took a break and worked in the bioinformatics lab of Dr. Purvesh Khatri at Stanford University. Khatri’s research had focused on knitting together multiple data sets to find genes responsible for organ transplant rejection. Inspired by the research, Sweeney decided to use a similar approach for a similar project: He wanted to analyze gene expression data from patients suspected of having sepsis to find patterns that could separate those with an infection from those with non-infectious inflammation.  

“Surgery was good background for this because we knew that at a cellular level, the immune system knows the difference between inflammation caused by a scalpel and that caused by a pathogen,” Sweeney explains. “I thought that we could use big data to find a better way to diagnose postoperative infections.” 

His research project led to his first published paper in 2015. Analyzing blood samples from over 1,000 patients, he and his colleagues identified a signature of gene expression from seven genes that could discriminate inflammation caused by a pathogen with a sensitivity of 94% (1). The approach presented a departure from most existing tests that identify infection by looking for proteins or nucleic acids from a pathogen.  
 
Instead, the technique analyzes the expression of genes of the immune system’s response to the invading pathogen. The researchers then combined the gene expression data with an algorithm that gives patients a score based on the severity of their symptoms. They found that the model they developed could predict which patients might have the most severe outcomes, and hence should be treated aggressively.
The 'gray zone' in sepsis diagnosis represents patients with unclear symptoms - neither obviously septic nor entirely well - posing a challenge for doctors to predict who might worsen. Tools like the TriVerity test aim to decode this uncertainty with precision, which would help manage cases.
We've relied on QIAGEN Genomic Services to do the RNA sequencing work on many samples collected for our research...  [while] PAXgene stabilizes the RNA expression levels in the tube of blood. And we're reading those RNA expression levels.
Tim Sweeney, MD, PhD, co-founder and CEO of Inflammatix

The beginning of Inflammatix

“The innovation was not the idea of using gene expression per se, but showing how to build something that could be applied to incoming patients with high accuracy,” says Sweeney. Soon after, the research team began exploring if they could use the same concept to separate bacterial from viral infections and to predict 30-day mortality from a blood sample taken from patients upon being admitted to the hospital (2,3).  

These advancements laid the groundwork for Inflammatix, which Sweeney co-founded in 2016 to revolutionize diagnostics in emergency and intensive care settings by leveraging the body’s immune response. 

Their flagship product, TriVerity, rapidly analyzes gene expression profiles of specific immune system genes from a blood sample. Applying advanced machine learning algorithms to the expression levels of the genes measured, TriVerity can differentiate between non-infectious conditions, bacterial and viral infections and can determine the likelihood of severe illness in about 30 minutes.   

TriVerity’s utility at triage is potentially broad. The bacterial and viral scores provide information to help physicians decide whether antibiotics or antivirals are warranted. The separate severity score “can help doctors decide what level of care a patient really needs,” Sweeney says. Another important advantage over traditional tests like blood cultures is that TriVerity can also identify infections in other parts of the body that can’t be detected by a blood culture, Sweeney adds.   
Tim Sweeney, M.D., Ph.D.
The US Food and Drug Administration designated the TriVerity test as a breakthrough device in 2023 and is currently reviewing the company’s application to market the device. Breakthrough designation was given to the test because it was able to pick out those patients that progressed to severe illness and death, despite appearing only mildly ill at initial presentation. Identifying these “occult sepsis” cases could make the test truly lifesaving, says Sweeney.
The fact that there's this separate severity score really allows us to identify the risk of progression and help make better decisions beyond just is there an infection.
Tim Sweeney, MD, PhD, co-founder and CEO of Inflammatix

Research collaboration with QIAGEN

In the fast-paced world of research, collaboration is the key to success. “Over the years, we've relied on QIAGEN Genomic Services to do the RNA seq work on the patient samples we bring in. And I think we've done several thousand samples at this point with QIAGEN,” says Sweeney.

“We started as a bunch of data scientists, not producers of hardware like cartridges and instruments. We've relied on QIAGEN Genomic Services to do the RNA sequencing work on many samples collected for our research, which forms the core of our signature technology.”

The Inflammatix team also worked extensively with QIAGEN QIAcube RNA extraction system to develop their test. “The QIAcube is an excellent system,” says Sweeney. “We used it as the gold standard for doing RNA extractions before we had fully developed the TriVerity test system.”

The final TriVerity platform is intended to be used with QIAGEN PAXgene Blood RNA tubes for blood collection, which stabilizes RNA for further processing and analysis. “The PAXgene tubes have been critical to our research and development efforts,” says Sweeney. “We found them to be the most reliable way to capture and store RNA without it degrading. We often collected samples, stored them in the freezer and ran them years later with great reliability.”

“Sepsis is a public health emergency,” Sweeney adds. “It’s the single most expensive diagnosis of patients who are insured by Medicare and it is the number one source of malpractice claims because of misdiagnosis. If we could get this right, we could save a lot of lives and spare the health care system unnecessary costs by giving patients the right treatment at the right time.”
Tim Sweeney, M.D., Ph.D.
Tim Sweeney, MD, PhD, is co-founder and CEO of Inflammatix. Sweeney has experience across clinical practice (general surgery & critical care), bench research, and bioinformatics and machine learning. While training at Stanford University, he helped invent the core technology on which Inflammatix is based, and is named on over a dozen patents related to medical diagnostics.
References:

1. Sweeney TE, et. al. A comprehensive time-course–based multicohort analysis of sepsis and sterile inflammation reveals a robust diagnostic gene set. Sci. Transl.Med. 2015;7: 287ra71-287ra71. DOI:10.1126/scitranslmed.aaa5993

2. Sweeney TE, et. al. Robust classification of bacterial and viral infections via integrated host gene expression diagnostics. Sci. Transl. Med. 2016; 8: 346ra91 DOI: 10.1126/scitranslmed.aaf7165 

3. Sweeney TE, Perumal T, Henao R, et. al. A community approach to mortality prediction in sepsis via gene expression analysis. Nat Comms 2018; 9: 694 DOI:10.1038/s41467-018-03078-2 

February 2025

Related content