Multiple imputation approaches for epoch-level accelerometer data in trials.

Area of Science:

• Biomedical Engineering
• Clinical Research Methodology
• Digital Health

Background:

• Accelerometers are crucial for measuring physical activity in clinical trials, collecting data in fine-grained epochs.
• Missing data from non-wear time is a common challenge, often handled by day-level aggregation, losing temporal information.
• Existing methods for missing data analysis may not fully capture the nuances of epoch-level missingness.

Purpose of the Study:

• To develop and evaluate novel methods for identifying and classifying missing accelerometer data at the epoch-level.
• To compare parametric and non-parametric multiple imputation techniques for handling missing epoch-level data.
• To assess the performance of different imputation strategies in estimating treatment effects in physical activity interventions.

Main Methods:

• Proposed an epoch-level approach to identify and classify missing accelerometer data, moving beyond day-level definitions.
• Implemented two multiple imputation strategies: a parametric approach considering missing epochs per day and a non-parametric approach using donor data.
• Conducted simulation studies to compare the bias and precision of the imputation methods.
• Applied the methods to real-world data from the 2017 PACE-UP trial.

Main Results:

• The non-parametric multiple imputation approach demonstrated the least bias in treatment effect estimates.
• This method also maintained small standard errors, indicating good precision.
• The proposed framework effectively handles missing data in wearable device studies.

Conclusions:

• Epoch-level analysis of missing accelerometer data provides a more nuanced understanding than day-level aggregation.
• Non-parametric multiple imputation is a robust method for addressing missing data in physical activity research.
• The developed framework is adaptable for various digital health outcomes and wearable sensors.