Nature: A very accurate diagnostic tool for sepsis

Sepsis is an overreaction of the immune system to infection, responsible for 20 percent of deaths worldwide each year and 20 to 50 percent
of hospital deaths in the United States.
However, despite its prevalence and severity, this condition is difficult to diagnose and effectively treat.

The disease causes reduced blood flow to vital organs, inflammation throughout the body, and abnormal blood clotting
.
Therefore, if sepsis is not quickly recognized and treated, it can lead to shock, organ failure, and death
.
But it's hard to determine which pathogen is causing the sepsis, and whether the infection is in the blood or elsewhere in the body
.
In many patients with symptoms similar to sepsis, it can be difficult to determine whether they are actually infected
.

Now, researchers at the CZ Biohub, the Chan Zuckerberg Initiative (CZI), and the University of California, San Francisco (UCSF) have developed a new diagnostic approach that applies machine learning to advanced genomic data from microbes and hosts to identify and predict cases of
sepsis.
As reported in the journal Nature Microbiology on October 20, 2022, this approach is surprisingly accurate and has the potential to far exceed current diagnostic capabilities
.

"Sepsis is one of the top ten public health problems facing humanity," said
senior author Chaz Langelier, MD, associate professor of medicine in the UCSF Division of Infectious Diseases and researcher at the CZ Center for Biology.
"One of the key challenges in sepsis is diagnosis
.
Existing diagnostic tests fail to capture the two-sided nature of the disease — the infection itself and the host's immune response
to the infection.
”

Current sepsis diagnosis focuses on detecting bacteria by culturing them, a process that "is critical to proper antibiotic treatment, which is essential for the survival of sepsis," said
the researchers on the new method.
But culturing these pathogens is time-consuming, and the bacteria
causing the infection are not always correctly identified.
Similar to viruses, PCR tests can detect viruses that infect patients, but they do not always identify the specific virus
that causes sepsis.

"This results in clinicians not being able to determine the cause of sepsis in an estimated 30 to 50 percent of cases," Langelier said
.
"It also leads to a mismatch
between antibiotic treatment and the pathogen causing the problem.
"

In the absence of a definitive diagnosis, doctors often prescribe a cocktail of antibiotics in an attempt to stop the infection, but the overuse of antibiotics has led to an increase
in antibiotic resistance worldwide.
"As doctors, we don't want to miss a single case of infection," said
M.
A.
S Carolyn Calfee, M.
A.
S.
, M.
S.
, professor of medicine and anesthesiology at the University of California, San Francisco and co-senior author of the new study.
"But if we have a test that can help us determine exactly who isn't if they're infected, that can help us limit antibiotic use in these situations, which is good for
all of us.
"

Disambiguation

The researchers analyzed whole blood and plasma samples from more than 350 critically ill patients who were admitted to UCSF Medical Center or Zuckerberg San Francisco General Hospital between 2010 and 2018
.

But instead of relying on culture to identify pathogens in these samples, the team, led by scientists at CZ Biocenter, PhD, and Angela Pisco used metagenomic next-generation sequencing (mNGS).

This method identifies all nucleic acids or genetic data present in the sample and then compares this data to a reference genome to identify the presence of microbial organisms
.
The technique allows scientists to identify genetic material from completely different biological kingdoms — whether bacteria, viruses or fungi — that all appear in the
same sample.

However, detecting and identifying the presence of pathogens alone is not enough to accurately diagnose sepsis, so researchers at the Biocenter also conducted transcription profiling — quantifying gene expression — to capture a patient's response
to infection.

Next, they applied machine learning to mNGS and transcribed data to distinguish between sepsis and other serious diseases to confirm the diagnosis
.
Dr.
Katrina Kalantar, CZI's lead computational biologist and co-first author of the study, created an integrated host-microbe model that trained data from patients, whether sepsis or non-infectious systemic inflammatory diseases, makes the diagnosis of sepsis highly accurate
.

Kalantar explains: "We developed this model
by looking at a range of metagenomics data and the results of traditional clinical tests.
" First, the researchers identified gene expression changes between patients with confirmed sepsis and those with clinically similar non-infectious systemic inflammation, and then used machine learning to determine the genes
that best predicted those changes.

The researchers found that when traditional bacterial cultures identified a pathogen that causes sepsis, there was often too much genetic material
from that pathogen in the corresponding plasma samples analyzed by mNGS.
With this in mind, Kalantar programmed the model to identify organisms that were too numerous compared to other microbes in the sample, and then compared
them to a reference index of microorganisms known to cause sepsis.

"In addition to this, we also notice any viruses detected, even if they are at low levels, because they really shouldn't be there
," Kalantar explained.
"With this relatively simple set of rules, we can do just fine
.
"

Near-perfect performance

The researchers found that the mNGS methods and their corresponding models worked better than expected: they were able to identify 99 percent of confirmed cases of bacterial sepsis, 92 percent of confirmed cases of viral sepsis, and were able to predict sepsis
in 74 percent of undiagnosed clinically suspected cases.

Dr.
Lucile Neyton, a postdoctoral researcher in the Caltech lab and co-first author of the study, said: "We were expecting good performance, even excellent performance, but it was almost perfect
.
" "By using this approach, we can get a good idea of what causes the disease, and we can know with relative confidence whether a patient has sepsis
.
"

The team is also excited to find that they can use this method that combines host response and microbial testing to diagnose sepsis using plasma samples, which are routinely collected from most patients as part of standard clinical care
.
"The fact that you can identify patients with sepsis from this widely available, easy-to-collect sample type makes a lot of sense in terms of practical use," Rangelier said
.

The idea for this work stemmed from previous research by Langelier, Kalantar, Calfee, UCSF researcher and CZ Biohub president Joe DeRisi, PhD, and colleagues, who used mNGS to effectively diagnose lower respiratory tract infections
in critically ill patients.
Because this approach works so well, Kalantar says, "we wanted to see if the same approach could work for sepsis
.
" ”

Broader impact

The team hopes to build on this successful diagnostic technique to develop a model that can also predict antibiotic resistance in pathogens detected with this method
.
"We've had some success treating respiratory infections, but no one has come up with a good treatment for sepsis," Langelier said
.

In addition, the researchers hope to eventually predict outcomes for patients with sepsis, "such as mortality or length of hospital stay, which will provide critical information that enables clinicians to better care for their patients and match resources to those who need them most," Rangelier said
.

Calfee added, "Novel sequencing methods like this have great potential to help us more precisely determine the cause of
severe disease in patients.
" "If we can do that, it's the first step towards precision medicine and understanding the individual patient
.
"