An application of a pattern-mixture model with multiple imputation for the analysis of longitudinal trials with...

Area of Science:

• Clinical Medicine
• Biostatistics
• Infectious Diseases

Background:

• Randomized clinical trials are essential for evaluating treatment efficacy.
• Protocol deviations in trials can lead to missing data, compromising results.
• Sensitivity analysis is crucial for assessing the impact of missing data assumptions.

Purpose of the Study:

• To evaluate the effect of prednisolone on CD4 count changes in tuberculosis pericarditis patients.
• To address missing data challenges using robust statistical methods.
• To assess the sensitivity of inferences to different missing data assumptions.

Main Methods:

• Analysis of CD4 count changes using a pattern-mixture model with multiple imputation (PM-MI).
• Adjustment for baseline and time-dependent covariates in the statistical model.
• Conducting simulation experiments to evaluate imputation method performance.

Main Results:

• Prednisolone treatment showed no significant effect on CD4 count changes over time.
• CD4 counts increased significantly over the study period, with higher levels in patients on anti-retroviral therapy (ART).
• Older patients exhibited lower CD4 counts; parameter estimates were robust to missing at random (MAR) and missing not at random (NMAR) assumptions.

Conclusions:

• Missing data in CD4 counts were determined to be missing at random (MAR).
• Statistical inferences derived from MAR analyses are robust to NMAR assumptions.
• Valid inferences can be achieved using likelihood-based methods or multiple imputation.