A recent study from Sichuan University, the University of A Coruña, and their colleagues tested the reliability of a machine learning model for analyzing patients’ sepsis mortality risk.
The researchers utilized data from the eICU Collaborative Research Database. According to a Chinese Academy of Sciences release, the two-phase transformer-based model “dynamically processes both hourly and daily health indicators.” The model scored an operating characteristic curve (AUC) of 0.92.
Additionally, “the inclusion of SHAP (SHapley Additive exPlanations) visualizations ensures interpretability, allowing clinicians to understand which factors drive predictions.”
The researchers hope the model can lead to more personalized care and less sepsis mortality.
The study is published in Precision Clinical Medicine.