News: EHR data reveals four sepsis subtypes, study shows

A study published in the Journal of the American Medical Association (JAMA) used machine learning to analyze data in EHRs and found four subtypes of sepsis.

Researchers from the University of Pittsburgh School of Medicine used statistical, machine learning, and simulation tools to analyze 29 clinical variables over 20,000 EHR records of patients recognized to meet sepsis-3 criteria within six hours of arriving at the hospital from 2010 to 2012.

The algorithm found four distinct sepsis types: alpha, beta, gamma, and delta. The importance of the different phenotypes is that sepsis patients might benefit from different types of treatment instead of a one-size-fits-all approach, researchers suggest.

Alpha was found to be the most common subtype, with 33% of patients falling into this category. Subtype alpha patients had the fewest abnormal lab test results, the least organ dysfunction, and lowest in-hospital death rates (2%).

Beta phenotype patients were found to be generally older and had the most chronic illnesses and kidney dysfunction. The beta sepsis subtype was found in 27% of the patients.

Gamma subtype sepsis was also present in 27% of the studied patients. These patients were found to have more inflammation and pulmonary dysfunction.

Delta phenotypes accounted for 13% of patients. Subtype delta patients suffered more liver dysfunction and septic shock, making it the deadliest of the sepsis phenotypes with an in-hospital death rate of 32%.

Researchers used the same algorithm for an additional 43,000 sepsis patient EHR records from the years 2013 to 2014. Results found a similar frequency of each sepsis type and clinical characteristics as the initial analysis. Similar results show the algorithm can be used for multiple patient populations and yield the same result.

The research team also applied the algorithm and study tools to data from recent international clinical trials, texting different promising sepsis treatments ending with average results. When the trial participants were sorted into the four sepsis subtypes, researchers found that some clinical trials saw better results for each phenotype.

“These proof-of-concept clinical phenotypes could be incorporated prospectively in future study designs that test new biologically active therapeutics. Novel designs could enrich for a priori phenotypes as well as confirm the boundaries around predictive phenotypes during the trial,” the researchers said.

Editor’s note: The JAMA published study can be found here. To read about the evolution of sepsis definitions, click here.