Missing data imputation techniques for wireless continuous vital signs monitoring
Authors' conclusion
Imputation error ranges vary between imputation techniques and increase with gap length. Imputation may result in larger signal feature bias compared to performing no imputation, and can affect patient risk assessment as illustrated by the EWS.
Study context
- Population
- Surgical ward patients with wireless continuous vital signs monitoring (heart rate, respiratory rate, SpO2, axillary temperature); simulated gaps of 5–60 min in two-hour windows1,743 data windows
- Participants
- 52 patients
- Comparison
- Linear interpolation vs spline interpolation, last observation and mean carried forward, cluster-based prognosis, and no imputation
- Outcome
- Imputation performance: mean absolute error (MAE) between original and imputed gap samples
Methodological detailsrepeated random simulation of missing data periods
- Repeated random simulation of missing data periods
Main finding
As reported in the abstract
Median MAE for 5-60 min gaps
| Group | Heart rate mean absolute error (MAE) | Respiratory rate MAE | Temperature MAE | Oxygen saturation MAE |
|---|---|---|---|---|
| Linear interpolation | 0.9-2.6 beats/min | 0.8-1.8 breaths/min | 0.04-0.17 °C | 0.3-0.7% |
Linear interpolation
Heart rate mean absolute error (MAE)0.9-2.6 beats/min
Respiratory rate MAE0.8-1.8 breaths/min
Temperature MAE0.04-0.17 °C
Oxygen saturation MAE0.3-0.7%
Further findings
Signal feature bias (window's slope, mean)
- Three techniques resulted in larger ranges of signal feature bias vs no imputation
Early warning scores (EWS) misclassification
- Imputation led to EWS misclassification in 1-8% of all simulations
Original publication
Missing data imputation techniques for wireless continuous vital signs monitoring. J Clin Monit Comput. Published 2 February 2023. doi:10.
Composed by AI from the published abstract and checked against it. How Briefs are made →