Individual information-centered approach for handling physical activity missing data.

Area of Science:

• Biomedical Informatics
• Data Science
• Gerontology

Background:

• Missing data is a common challenge in physical activity research.
• Accurate imputation methods are crucial for valid analysis of accelerometer and pedometer data.
• Existing methods for handling missing data may not be optimal for specific demographic groups.

Purpose of the Study:

• To validate individual information (II)-centered methods for missing data imputation.
• To compare the performance of II-centered methods against group information (GI)-centered methods.
• To identify effective missing data recovery strategies for physical activity data in middle-aged and older adults.

Main Methods:

• A semisimulation approach was used to create six datasets with missing values.
• Three physical activity measures (step counts, activity counts, minutes of moderate to vigorous physical activity) were analyzed.
• Missing values were imputed using two II-centered and two GI-centered methods.
• Effectiveness was assessed using Root Mean Square Difference (RMSD), mean signed difference, paired t tests, and Pearson correlations.

Main Results:

• II-centered methods demonstrated smaller RMSDs compared to GI-centered methods across all datasets and groups.
• No significant mean differences were observed between known and imputed values under any condition.
• The average of remaining days was identified as an accurate missing data recovery method.

Conclusions:

• II-centered methods are superior to GI-centered methods for imputing missing physical activity data.
• Imputing missing data using the average of remaining days is a valid and accurate approach for middle-aged and older adults.
• These findings support the use of II-centered imputation for enhancing the reliability of physical activity monitoring studies.