Bloodstream infection (BSI) such as sepsis can be difficult to detect at an early stage. The laboratory tests to determine BSI can take two to three days, whereas treatment for sepsis must begin in hours. Data in electronic health records (EHR), including rapidly available laboratory data and medical history, can be used to generate alerts of possible sepsis.
Treatment of sepsis typically calls for administration of antibiotics within a few hours of diagnosis, long before a specific infection has been identified, and leading to the use of broad spectrum antibiotics. Widespread use of broad spectrum antibiotics is one of the factors to have led to the emergence of antimicrobial resistance (AMR) as a major healthcare concern in recent years. AMR has been identified as a major concern in the treatment of sepsis. Early detection of AMR is essential for clinicians to be able to determine an appropriate antimicrobial therapy.
In a recent paper, researchers at King's College London, reported the use of machine learning (ML) to predict BSI and AMR using EHR data. They report that an ML approach based on Multi-Objective Symbolic Regression (MOSR) trained on a dataset, including EHR, BSI, and AMR data, provided an ability to predict BSI and AMR from readily available patient EHR data without relying on previous patient microbiological data relating to infections and resistance that have been used in other attempts to use ML to predict BSI and AMR.
While the described ML approach only provides a binary outcome for each indication, it illustrates how AI has the potential to achieve a tailored antimicrobial therapy more quickly that improves patient outcome and limits the use of broad spectrum antibiotics that feeds into the problem of AMR.
'We anticipate that 50% of sepsis cases are due to resistant bacteria,'
The content of this article is intended to provide a general guide to the subject matter. Specialist advice should be sought about your specific circumstances.
