New model enhances sepsis risk assessment in ICUs
Researchers have developed a new two-stage Transformer model to improve sepsis risk assessment in ICU patients. This model uses hourly and daily data from over 13,000 patients, achieving a predictive performance score of 0.92 by the fifth day of admission. The model incorporates SHAP-derived temporal heatmaps to highlight key biomarkers linked to patient outcomes, such as lactate and chloride levels. This feature helps clinicians understand the model's predictions better. External validation showed the model's effectiveness across different populations, with accuracy rates of 81.8% on Chinese data and 76.56% on the MIMIC-IV database. This confirms its adaptability in various healthcare settings.