Primary analysis method for incomplete CD4 count data from IMPI trial and other trials with similar setting

TL;DR

This paper recommends using maximum likelihood or multiple imputation methods as primary analysis techniques for handling missing CD4 count data in clinical trials, ensuring unbiased and consistent treatment effect estimates under missing at random assumptions.

Contribution

It provides a comparative analysis of different methods for handling missing CD4 data, advocating for maximum likelihood or multiple imputation as primary approaches.

Findings

Simple imputation methods produce biased estimates.

Maximum likelihood and multiple imputation yield unbiased, consistent estimates.

Recommended methods are suitable under missing at random assumptions.

Abstract

The National Research Council panel on prevention and treatment of missing data in clinical trials recommends that primary analysis methods are carefully selected before appropriate sensitivity analysis methods can be chosen. In this paper, we recommend an appropriate primary analysis method for handling CD4 count data from the IMPI trial and trials with similar settings. The estimand of interest in the IMPI trial is the effectiveness estimand hypothesis. We discussed, compared, and contrasted results from complete case analysis and simple imputation methods, with the direct-likelihood and multiple imputation methods. The simple imputation methods produced biased estimates of treatment effect. However, the maximum likelihood and the multiple imputation methods produced consistent estimates of treatment effect. The maximum likelihood or the multiple imputation approaches produced unbiased and consistent estimates. Therefore, either the maximum likelihood or the multiple imputation methods, under the assumption that the data are missing at random can be considered as primary analysis method when one is considering sensitivity analysis to dropout using the CD4 count data from the IMPI trial and other trials with similar settings.