To satisfy this major healthcare need, 24 scientists from across the world collaborated on a community approach to derive gene-based models that can accurately predict 30-day mortality in patients with sepsis at the time of patient enrollment. The project was led by scientists from Sage Bionetworks, the University of South Alabama, Duke University and Stanford University, and the results were published today in Nature Communications.
“Sepsis is a heterogeneous disease contributing to half of all in-hospital deaths in the United States and is a leading cost for our health care system,” said Dr. Raymond Langley, Assistant Professor at the University of South Alabama and one of the senior authors on the publication.
“Despite dozens of clinical trials, there are no pharmacologic treatments specific for sepsis that have been successfully utilized in clinical practice,” he said. “Patient treatment still focuses on general management strategies including source control, antibiotics and supportive care. With improved accuracy in sepsis prognosis we can hopefully improve critical care through appropriate matching of patients with resources.”
For this study, the team identified a large collection of both public and privately-held gene expression data from clinical sepsis studies at the time of diagnosis to study molecular changes in immune cells, also known as leukocytes. The patients were enrolled in the emergency department or the intensive care unit from hospitals and universities around the world; such as the United States, Spain, England and Sydney, Australia.
These highly valuable datasets have been published publicly so others can perform independent analyses. Three scientific groups were then invited to build algorithmic models to predict 30-day mortality. Four different prognostic models, which had widely different predictive features (genes), were then evaluated utilizing independent external validation cohorts composed of patients with either community-acquired sepsis or hospital-acquired infections.
The research comprehensively revealed that patients with sepsis can be risk-stratified based on their gene expression profiles at the time of diagnosis.
“One the major strengths of this paper is that the data were subjected to analysis by several different groups of scientists using different methods. This allows us to suggest that the accuracy of the models we derived is unlikely to be beaten by other groups looking at similar data” said Timothy Sweeney, MD, PhD, lead author of the study and a researcher at Stanford at the time the work was performed. “Our focus here was explicitly to derive a highly predictive gene expression ‘fingerprint’ that could someday become a clinical tool. For that reason, our focus was on improving accuracy more than understanding biology.”
Still, said Dr. Langley, “These gene expression models reflect a patient’s underlying biological response and could potentially serve as a valuable clinical assay for prognosis and for defining the host dysfunction responsible for sepsis. These results serve as a benchmark for future prognostic model development and as a rich source of information that can be mined for additional insights. Our community approach identified a large number of genes associated with sepsis mortality that may point to underlying biology.”
Dr. Langley said the data from this study holds far-reaching potential to improve critical care medicine on a global scale.
“A very sick patient could be diverted to ICU for maximal intervention, while patients predicted to have a better outcome may be safely monitored in the hospital or even discharged early,” he said. “More precise estimates of prognosis would also allow for better discussions regarding patient preferences and the utility of aggressive interventions, while better molecular phenotyping of sepsis patients has the potential to improve clinical trials.”
The study was funded by the Defense Advanced Research Projects Agency (DARPA) as a DREAM Challenge initially tasked to determine resiliency. DREAM Challenges, powered by Sage Bionetworks, allow for a community approach to determine solutions to complex health problems to improve translational medicine.
To view the data published in Nature Communications, visit https://www.nature.com/articles/s41467-018-03078-2.
