UF Health sepsis study uses AI technology, urinary cellular gene expression for early detection

A groundbreaking study conducted by a multidisciplinary team of researchers at the University of Florida Sepsis and Critical Illness Research Center explores how urine analysis during early onset of sepsis may help determine the likelihood of chronic kidney problems in the long term.

Sepsis is a life-threatening condition that occurs when the body’s response to infection injures its own tissue or organs. Each year, according to the Centers for Disease Control and Prevention, more than 1.5 million American adults develop sepsis. Many suffer long-term, chronic effects from the serious condition, which also can cause organ failure.

The study, “Discovery and Validation of Urinary Molecular Signature of Early Sepsis,” was led by Azra Bihorac, M.D., the R. Glenn Davis Professor of Medicine, Surgery and Anesthesiology in the UF College of Medicine’s division of nephrology, hypertension and renal transplantation.

Sepsis impacts a wide range of organs, but for this study, Bihorac and her fellow researchers focused on the kidneys, which play a crucial role in many vital bodily functions by filtering blood and clearing our bodies of toxins.

“My research has been around acute kidney injury and sepsis, as well as informatics and AI, and linking all these together,” Bihorac said. “So, our hypothesis was that among all organ dysfunction, kidney dysfunction is particularly important.”

The research stems from a $12 million National Institute of General Medical Sciences grant that established UF’s Sepsis and Critical Illness Research Center in 2017.

Working from their hypothesis, the group sought to identify which genes in urinary cells were markers for early sepsis. They turned to a process of artificial intelligence called machine learning, which creates algorithms from data and then builds on the acquired knowledge without human programming or direction.

In a parallel clinical study, “Clinical Trajectories of Acute Kidney Injury in Surgical Sepsis,” Bihorac revealed that half of patients who don’t have recovery after kidney dysfunction die. The study further showed that of the patients who survive, one out of two die within a year of being discharged, and those who do survive often face chronic kidney problems for the rest of their lives.

Azra Bihorac, M.D. 

“We proved that in cases of sepsis with persistent acute kidney injury and a lack of recovery of acute kidney dysfunction mark this malignant clinical phenotype of the patients who don’t do well,” said Bihorac. “So, the idea is if you can recognize in the first few hours of sepsis who might be at risk for developing more malignant phenotypes, you can focus on kidney recovery or at least not aggravating the initial injury.”

The recent sepsis study builds off this work and homes in on cellular RNA isolated from the urine samples of septic patients.

Today’s technology allows researchers to use cellular RNA to look at an entire genome, but the process is costly. The team narrowed its scope to focus on 1,000 specific genes to find the best subset of genes that really differentiate sepsis from people who don’t have sepsis.

“For that we used ensemble machine learning algorithms. I think machine learning was very appropriate in this instance to find that combination,” she said.

Bihorac said the researchers analyzed urine for this study because it is a readily available, but sometimes overlooked, biofluid that can convey important information.

“A lot of proteins are filtered in the urine, but also a lot of RNA and DNA from the cells are shed in the urine,” said Bihorac. “These cells can come from different sources. Obviously, there are DNA and RNA coming from bacteria and viruses in the urine. People think urine is sterile, but it’s not completely. Even white blood cells from the blood can leak through the cells into the urine.”

To the researchers’ knowledge, this is the first study using urine-based gene expression in sepsis.

“In a single-center prospective cohort of patients with sepsis, an ensemble of four machine learning algorithms identified 239 gene expressions unique to systemic immune and kidney-specific processes using whole-genome transcriptomic analysis of cellular RNA isolated from urine samples within 12 hours of sepsis onset,” according to the study.

Bihorac said future investigations can use this study as a jumping-off point to confirm whether this approach can complement a blood transcriptomic approach for sepsis diagnosis and prognostication.

“The whole-genome transcriptomic analysis of cellular RNA isolated from the urine samples of septic patients reveals changes in gene expressions unique to systemic immune and kidney-specific processes as early as within 12 hours of sepsis onset,” she said.