Forecasting Hypotension during Vasopressor Infusion via Time Series Analysis

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Data Science in Healthcare

Background:

• Hypotension during vasopressor infusion requires precise management in intensive care units (ICUs).
• Predictive modeling can improve the early detection of sustained hypotension episodes (SHEs).

Purpose of the Study:

• To evaluate the efficacy of logistic regression (LR) and auto-regressive (AR) models in predicting SHEs.
• To compare these predictive models against a standard hypotension threshold detection method.

Main Methods:

• Utilized datasets from 207 patients across two hospitals.
• Developed and tested LR and AR forecasting models to predict SHEs.
• Compared model performance against a simple blood pressure (BP) threshold detector.

Main Results:

• The LR model predicted SHEs 7.0 min (Hospital 1) and 2.5 min (Hospital 2) prior to occurrence (at 60 mmHg threshold).
• The AR model predicted SHEs 10.5 min (Hospital 1) and 2.0 min (Hospital 2) prior to occurrence (at 60 mmHg threshold).
• Both models demonstrated significantly better prediction than the threshold method with comparable false alarm rates; AR model showed flexibility for various hypotension thresholds.

Conclusions:

• Predictive models, particularly the AR model, offer superior early detection of SHEs compared to traditional threshold methods.
• These models can aid in optimizing hypotension management during continuous vasopressor infusion in ICUs.
• The AR model's adaptability to different hypotension thresholds enhances its clinical utility.