Specification of covariance structure in longitudinal data analysis for randomized clinical trials

Area of Science:

• Biostatistics
• Longitudinal Data Analysis
• Statistical Modeling

Background:

• Covariance structure misspecification in longitudinal analysis can bias regression parameter estimates and standard errors, particularly with missing data.
• The sandwich estimator corrects variance but not point estimate bias.
• Unstructured (UN) covariance eliminates bias but may lead to convergence problems with extensive missing data.

Purpose of the Study:

• To examine the theoretical and simulated existence and magnitude of biases in longitudinal data analysis due to covariance misspecification.
• To evaluate the effectiveness of unstructured (UN) covariance in mitigating these biases.
• To provide guidance on the use of UN covariance in randomized clinical trials.

Main Methods:

• Theoretical examination of bias in longitudinal data analysis.
• Simulation studies to assess bias and convergence properties.
• Evaluation of the unstructured (UN) covariance approach.
• Development of an algorithm to aid convergence with UN covariance.

Main Results:

• Covariance misspecification leads to biased regression estimates and inaccurate standard errors when data are missing.
• Unstructured (UN) covariance effectively removes these biases.
• Convergence issues with UN covariance can occur but are manageable with appropriate algorithms, especially in trials with moderate subjects and time points.

Conclusions:

• Unstructured (UN) covariance is recommended as a default strategy for analyzing longitudinal data in randomized clinical trials with moderate to large sample sizes and limited time points.
• The proposed algorithm facilitates the use of UN covariance, improving the reliability of statistical inferences.
• Addressing covariance misspecification is crucial for accurate longitudinal data analysis.