Vital Signs Prediction for COVID-19 Patients in ICU

Area of Science:

• Critical Care Medicine
• Biomedical Engineering
• Data Science

Background:

• COVID-19 intensive care unit (ICU) patients require continuous monitoring of vital signs.
• Predictive modeling can potentially forecast patient health status, enabling timely interventions.

Purpose of the Study:

• To develop and evaluate machine learning models for predicting future vital signs in COVID-19 ICU patients.
• To assess the impact of the number of predictors and observation frequency on model performance.

Main Methods:

• Developed five machine learning models using varying combinations of vital signs (heart rate, respiration rate, oxygen saturation, blood pressure) and observation rates (hourly to bi-hourly, and five-minute intervals).
• Utilized leave-one-subject-out cross-validation for robust performance assessment.
• Predicted three upcoming vital sign observations, averaging three hours ahead.

Main Results:

• Models using three vital signs (heart rate, respiration rate, oxygen saturation) achieved acceptable prediction accuracy (MAPE: 12%, 21.4%, 5% respectively).
• Increasing observation rate to five-minute intervals significantly improved prediction performance (MAPE: 8%, 17.8%, 4.8% respectively).

Conclusions:

• Machine learning models can effectively predict vital signs for COVID-19 ICU patients.
• Optimizing predictor selection and increasing observation frequency enhances predictive accuracy for real-time health monitoring.