SepsisLab: Early Sepsis Prediction with Uncertainty Quantification and Active Sensing

Area of Science:

• Medical Informatics
• Clinical Decision Support
• Machine Learning in Healthcare

Background:

• Sepsis is a leading cause of in-hospital mortality, necessitating early prediction and diagnosis.
• Real-world clinical data often contain missing values, degrading the performance of existing sepsis prediction models.
• The uncertainty introduced by imputation methods in predictive models has not been adequately addressed.

Purpose of the Study:

• To quantify the uncertainty propagated from data imputation to sepsis prediction outputs.
• To develop a robust active sensing algorithm for high-risk patients with limited observations.
• To implement a practical system (SepsisLab) for early sepsis prediction and active sensing.

Main Methods:

• Defined propagated uncertainty as the variance of prediction output.
• Introduced uncertainty propagation methods for quantification.
• Developed and validated a novel active sensing algorithm using MIMIC-III, AmsterdamUMCdb, and OSUWMC data.

Main Results:

• Propagated uncertainty significantly impacts sepsis prediction, especially early in hospital admission.
• The proposed active sensing algorithm demonstrated superior performance compared to existing methods.
• Validation across multiple datasets confirmed the robustness and effectiveness of the developed models.

Conclusions:

• Addressing imputation uncertainty is crucial for reliable sepsis prediction.
• The active sensing approach enhances diagnostic confidence in critical early stages.
• The SepsisLab system offers a valuable tool for clinicians and patients in early sepsis detection.