Identifying reprioritization response shift in a stroke caregiver population: a comparison of missing data methods

Area of Science:

• Health Services Research
• Psychometrics
• Longitudinal Data Analysis

Background:

• Response shift (RS) can significantly impact longitudinal health-related quality of life (HRQOL) assessments.
• Missing data, common in longitudinal HRQOL studies, can hinder the detection of subtle RS effects.
• Effective strategies for managing missing data are crucial for accurate RS detection.

Purpose of the Study:

• To compare the efficacy of different imputation methods in detecting reprioritization response shift (RS) in the HRQOL of stroke survivor caregivers.
• To evaluate the statistical power of various imputation techniques for identifying RS in incomplete longitudinal HRQOL data.

Main Methods:

• Longitudinal data from 409 Canadian caregivers of stroke survivors over one year.
• Comparison of mean imputation, expectation-maximization (EM) imputation, and multiple imputation methods.
• Monte Carlo simulations assessed the power of relative importance tests for detecting RS under different missing data scenarios.

Main Results:

• Complete-case analysis failed to detect statistically significant changes in relative importance weights.
• Mean imputation and EM imputation detected significant changes in physical functioning and/or vitality domains.
• Multiple imputation identified significant changes in physical functioning, mental health, and vitality domains; it demonstrated the highest statistical power.

Conclusions:

• The detection of RS using relative importance measures is sensitive to missing data handling.
• Mean imputation and multiple imputation methods are recommended for robust RS detection in longitudinal HRQOL studies with missing data.
• These imputation methods enhance the power to detect RS effects compared to complete-case analysis.