Photo Credit: Zerbor

The following is a summary of “Machine learning for predicting mortality in adult critically ill patients with Sepsis: A systematic review,” published in the December 2024 issue of Critical Care by Nikravangolsefid et al. 

A variety of Machine Learning (ML) models have been employed to predict sepsis-associated mortality.  

Researchers conducted a retrospective study to examine the methodologies used in studies predicting mortality among individuals with sepsis.  

They followed a pre-registered protocol at the International Prospective Register of Systematic Reviews to conduct a thorough database search from inception to February 2024. Peer-reviewed articles reporting mortality prediction in adult critical illness along with sepsis were included.  

The results showed that out of 1,822 articles, 31 were selected, encompassing 1,477,200 individuals with sepsis, 19 studies exhibited a high risk of bias. Among the ML models, logistic regression and extreme Gradient Boosting were most used, appearing in 22 and 16 studies, respectively, 9 studies included both internal and external validation. When compared to traditional scoring systems like Sequential Organ Failure Assessment (SOFA), the ML models demonstrated slightly improved performance in predicting mortality (AUROC range: 0.62–0.90 vs 0.47–0.86).  

Investigators concluded the ML models showed a modest improvement in predicting sepsis-associated mortality, the certainty of these findings was limited by high bias risk, significant heterogeneity, and methodological shortcomings. 

Source: sciencedirect.com/science/article/abs/pii/S0883944124003769