Missing data imputation techniques for wireless continuous vital signs monitoring

Area of Science:

• Biomedical Engineering
• Health Informatics
• Remote Patient Monitoring

Background:

• Wireless vital signs sensors are crucial for remote patient monitoring.
• Missing data periods pose significant challenges for data analysis in remote patient monitoring.

Purpose of the Study:

• To evaluate the performance of various imputation techniques for continuous vital signs measurements.
• To assess the impact of imputation on signal features and early warning scores.

Main Methods:

• Simulated missing data gaps (5-60 min) in wireless vital signs measurements (heart rate, respiratory rate, oxygen saturation, temperature).
• Imputed gaps using linear interpolation, spline interpolation, last observation/mean carried forward, and cluster-based prognosis.
• Evaluated imputation performance using Mean Absolute Error (MAE) and analyzed effects on signal features and Early Warning Scores (EWS).

Main Results:

• Linear interpolation demonstrated the lowest median MAE across all vital signs and gap lengths.
• Other imputation techniques showed significantly higher MAE compared to linear interpolation.
• Imputation techniques, particularly non-linear ones, introduced bias in signal features and led to EWS misclassification (1-8% of simulations).

Conclusions:

• Linear interpolation is the most effective technique for imputing missing vital signs data from wireless sensors.
• The choice of imputation technique impacts signal feature accuracy and patient risk assessment via EWS.
• Careful selection and implementation of imputation methods are essential for reliable remote patient monitoring.