Sepsislab: Early sepsis prediction with uncertain quantification and active sensing

Abstract

Early prediction of sepsis is critical for improving patient outcomes, but remains challenging due to the complex and heterogeneous nature of the condition. This paper introduces Sepsislab, a novel approach for early sepsis prediction that incorporates uncertainty quantification and active sensing. Our method addresses two key challenges in sepsis prediction: the uncertainty in predictions and the need for timely and relevant clinical measurements. By quantifying prediction uncertainty, Sepsislab provides clinicians with confidence levels for risk assessments, enabling more informed decision-making. The active sensing component strategically recommends which clinical tests should be prioritized to maximize prediction accuracy while minimizing unnecessary testing. Experiments on real-world electronic health record data demonstrate that Sepsislab outperforms existing methods in prediction accuracy, calibration, and clinical utility. Our approach represents a significant step toward more reliable and resource-efficient sepsis prediction in clinical settings.

Dakuo Wang

Associate Professor

My research interests include human-computer interaction, human-AI collaboration, natural language processing, health informatics, data science, and computer-supported cooperative work.