Presentation (CTI)

OBJECTIVES: Missing data pose significant challenges in the statistical analysis of clinical trials, particularly in crossover designs where missing observations may arise for several reasons. Despite the growing interest in missing data methodologies, limited research has focused on three-arm, three-period crossover trials. This study aims to develop and compare missing data handling techniques for such crossover trials using microarray gene expression data.
METHODS: A three-arm, three-period crossover trial dataset comprising microarray gene expression values was analyzed, where gene expression levels were considered as therapeutic outcomes. Missing data were addressed using two approaches: Multiple Imputation (MI) and Bayesian imputation. Under the Bayesian framework, missing data imputation and causal effect estimation were performed jointly. In contrast, due to methodological limitations, the mixed-effects model could not jointly accommodate imputation and causal inference. Therefore, Multiple Imputation was first applied to generate complete datasets, followed by mixed-effects model analysis to assess the causal effects of therapeutic interventions on gene expression.
RESULTS: Both Multiple Imputation and Bayesian approaches effectively handled missing observations and enabled valid statistical inference. The Bayesian framework provided the additional advantage of jointly estimating missing values and treatment effects, whereas the Multiple Imputation approach required a two-stage analysis involving imputation followed by mixed-effects modeling. A comparative evaluation of both methods demonstrated their applicability for analyzing missing data in crossover trial settings.
CONCLUSIONS: The proposed methodologies provide effective solutions for handling missing data in three-arm, three-period crossover trials. While both approaches yield reliable inference, the Bayesian framework offers a more integrated strategy by jointly modeling missing data and treatment effects. These findings contribute to improving the analysis of crossover clinical trials with incomplete observations and facilitate more robust inference for gene expression studies.