Handling missing data in rest-activity time series measured by actimetry.

Area of Science:

• Chronobiology
• Biomedical Engineering
• Data Science

Background:

• Actigraphy is widely used to assess motor activity and sleep-wake patterns.
• Off-wrist episodes, or missing data, can significantly impact the accuracy of actigraphy analysis.
• Simulating and handling these missing data episodes is crucial for reliable results.

Purpose of the Study:

• To evaluate different methods for handling missing data (off-wrist episodes) in actimetry time series.
• To compare the impact of various missing data imputation strategies on key actigraphy parameters.
• To provide recommendations for handling missing data in motor activity analysis.

Main Methods:

• Generated simulated missing data intervals (1-24 hours and random) in 1-minute epoch actigraphy data.
• Imputed missing data using 'zeroes' (immobility), mean, or median imputation methods.
• Analyzed two types of records: regular and irregular sleep-wake cycles over 14 days.

Main Results:

• Single missing episodes up to 12 hours caused less than 5% variation in original values.
• Irregular sleep-wake cycles exhibited higher variability in parameters like acrophase and MESOR.
• Imputing missing data with zeroes significantly increased variance; mean/median imputation yielded patterns similar to original data.

Conclusions:

• Replacing missing data with mean or median is recommended over using zeroes to avoid inflating variance.
• Zeroes should be replaced with missing value (NA) indicators whenever possible in actigraphy data.
• If parameters cannot be computed with missing data, using the weekly mean of corresponding time points is advised.