top of page
Writer's pictureanalisakvmed
Oct 203 min read

AI and Sepsis

Sepsis: a Dangerous Condition

Sepsis is a life-threatening condition that accounts for 1 in 3 hospital fatalities in the United States.  It occurs due to an untreated infection that causes the body’s immune system to attack itself.  This often results in tissue damage, organ failure, and death.  With the timely administration of antibiotics, this condition is treatable.  However, due to the variety in its presentation from patient to patient, sepsis is often misdiagnosed or overlooked, leading to increased mortality. 


A New Approach to the Detection of Sepsis

A new diagnostic tool has been developed to improve the detection of sepsis using artificial intelligence (AI) and machine learning(ML).  This tool, Sepsis Immunoscore, is the first AI diagnostic tool for sepsis authorized for marketing by the FDA De Novo.  Immunoscore uses 22 parameters, including the patient’s demographics, vital signs, and biomarkers from lab tests to generate a “risk score.”  Based on the score, the patient’s risk for sepsis is categorized into one of the following “risk levels.”


  • Low 

  • Medium 

  • High

  • Very high


In addition, this score can be used to predict adverse clinical outcomes (within the 24 hours following the test) such as:


  • In-hospital mortality 

  • Length of hospital stay

  • Admission to the ICU 

  • Mechanical ventilation 

  • Vasopressor administration


The Development of Sepsis Immunoscore

This tool uses AI and ML, which require a dataset to “learn” from.  The Carle Foundation Hospital was the first healthcare system to participate in the development of Immunoscore.  They aided in developing a “biobank” by providing data from the blood tests of 2000 patients.  From this data, the algorithm was trained to predict the probability of a given patient meeting the criteria for sepsis within the next 24-hour period.  This dataset has since grown and continues to improve the algorithm’s accuracy.  


The Reliability of Sepsis Immunoscore

This algorithm has a 95% confidence score (meaning that the algorithm is “95% sure” that it is correct).  This was estimated by calculating the standard error with binomial approximation.  Immunoscore is not the first diagnostic test for sepsis.  However, its accuracy has proven to be comparable or superior to existing sepsis tests including:


  • Procalcitonin 

  • IntelliSep Test

  • Coulter Cellular Analysis System’s Early Sepsis Indicator

  • FebriDx 

  • MeMed BVTM


This is due to Immunoscore’s unique ability to create a more comprehensive risk score for a given individual.  While the above tests can only detect certain biomarkers, the Immunoscore algorithm is multidimensional and uses that patient’s specific demographics, vital signs, lab tests, and other information in addition to the biomarkers the other tests detect, to generate a risk score.  Because sepsis varies so much between individuals, the use of an algorithm designed to analyze each individual’s risk could lead to more timely diagnoses of sepsis, and therefore, a decrease in mortality. 



Potential Limitations

While this new technology is promising, there are a few factors that remain to be considered before its more widespread distribution.  Cost is a limiting factor, as sepsis is one of the most expensive conditions to treat.  Additionally, there are many technical limitations, as this is an ML algorithm, which requires a lot of data and a robust technological infrastructure, neither of which exists in many healthcare facilities.  Finally, as with every new medical tool, clinician acceptance is essential.  A lack of this acceptance could hinder the implementation of this innovation and could result in misuse.  


However, despite the current limitations, medical innovations such as Sepsis Immunoscore show promising potential for the improved detection of sepsis, and perhaps for other applications in the future. 


References:

Authors, et al. “Artificial Intelligence for the Prediction of Sepsis in Adults.” Nih.gov, Canadian


Bhargava, Akhil, et al. Development and Validation of the First FDA Authorized Artificial

Intelligence/Machine Learning Diagnostic Tool for the Prediction of Sepsis Risk. 7 May 2024, www.medrxiv.org/content/10.1101/2024.05.06.24306954v1.full.pdf, https://doi.org/10.1101/2024.05.06.24306954. Accessed 8 Oct. 2024.


Carle. “CI MED Faculty Member’s Research Fuels New AI Tool for Sepsis Diagnosis,


Prenosis. “PRENOSIS ANNOUNCES COMMERCIAL DISTRIBUTION COLLABORATION

with ROCHE for SEPSIS IMMUNOSCORETM, the FIRST FDA de NOVO AUTHORIZED ARTIFICIAL INTELLIGENCE TOOL for SEPSIS DETECTION.” Prnewswire.com, 10 Apr. 2024, www.prnewswire.com/news-releases/prenosis-announces-commercial-distribution-collaboration-with-roche-for-sepsis-immunoscore-the-first-fda-de-novo-authorized-artificial-intelligence-tool-for-sepsis-detection-302112509.html. Accessed 8 Oct. 2024.


Varma, Deepa. “FDA Approves AI Diagnostic Tool for Early Sepsis Detection.”

Medscape, 15


“What Is Sepsis? - End Sepsis.” End Sepsis, 28 Sept. 2023, www.endsepsis.org/what-is-


Cover image from: https://encrypted-tbn0.gstatic.com/images? q=tbn:ANd9GcQjR3cHUJBzmA-ZnnTuv2qZyjzk8OwZ-41zvA&s Assessed and Endorsed by the MedReport Medical Review Board









bottom of page