A framework for handling missing accelerometer outcome data in trials.

Area of Science:

• Clinical Trials Methodology
• Biostatistics
• Wearable Technology in Health

Background:

• Wearable devices (e.g., accelerometers) are crucial for objective physical activity measurement in clinical trials.
• Missing data from these devices is common, arising from non-adherence or technical failures, complicating interpretation.
• Current literature lacks consensus on defining and handling missing data, impacting analysis consistency.

Purpose of the Study:

• To propose a novel analysis framework for handling missing physical activity data from wearable devices in clinical trials.
• To establish a standardized method for defining missingness based on device wear time.
• To introduce a multiple imputation approach for robust data analysis and sensitivity assessments.

Main Methods:

• Developed a framework defining missing data at epoch and day levels using device wear time.
• Proposed a day-level multiple imputation method treating partially observed step counts as right-censored.
• The approach incorporates auxiliary variables and is compatible with existing primary analysis models.

Main Results:

• The proposed framework provides a flexible and consistent method for defining and imputing missing data.
• The multiple imputation technique allows for the inclusion of auxiliary variables, enhancing analytical power.
• Sensitivity analyses can be readily performed to assess the impact of the missing at random assumption.

Conclusions:

• The presented framework offers a standardized approach to managing missing physical activity data in clinical trials.
• This method improves the reliability of objective physical activity assessments from wearable devices.
• The approach was successfully illustrated using data from the 2019 MOVE-IT trial.