Bayesian Gaussian processes for identifying the deteriorating patient.

Area of Science:

• Biomedical Engineering
• Clinical Informatics
• Data Science in Healthcare

Background:

• Step-down units (SDUs) manage high-acuity patients post-intensive care unit (ICU) discharge.
• A significant minority of SDU patients (approx. 14%) deteriorate, requiring ICU readmission.
• ICU readmission is linked to increased mortality and prolonged hospital stays.

Purpose of the Study:

• To develop and evaluate Gaussian process regression (GPR) models for continuous patient vital sign monitoring in SDUs.
• To focus on robust heart rate time-series forecasting for early patient deterioration detection.
• To address the clinical need for improved SDU patient surveillance.

Main Methods:

• Application of Gaussian process regression (GPR) models for time-series analysis.
• Forecasting patient heart rate trends using GPR.
• Utilizing a University of Pittsburgh Medical Center SDU dataset (333 patients, 59 with clinical events).

Main Results:

• GPR-based heart rate monitoring demonstrated superior early warning capabilities for patient deterioration.
• The proposed GPR methods outperformed current rules-based thresholding practices.
• GPR performance slightly surpassed the state-of-the-art kernel density method, which requires additional vital signs.

Conclusions:

• GPR models offer a flexible, probabilistic approach for SDU patient monitoring.
• Early detection of deterioration via GPR can potentially mitigate risks of ICU readmission.
• GPR-based heart rate monitoring represents a promising advancement in SDU clinical surveillance.