Implications of missing data in tuberculosis non-inferiority clinical trials

Non-inferiority designs have been increasingly used in randomised clinical trials in recent years. However, there remain several key issues with this design that can have important implications for the primary analysis and its interpretation. Specifically, choosing the population for inclusion in the primary analysis and how to deal with missing values, remains unclear. This thesis tackles three related methodological issues in tuberculosis (TB) clinical trials: (i) a lack of clear guidance on design and reporting; (ii) the need for a valid approach to missing data and (iii) how to perform sensitivity analysis. First, widely available guidance documents on non-inferiority trials are critiqued, highlighting differences in recommendations between them on fundamental issues. These differences are reflected in inconsistent reporting from a systematic review we conducted, and make suggestions for improvements. Second, using data from two recent TB non-inferiority trials, we compare and contrast (i) different imputation approaches, (ii) inverse probability weighting with marginal models, and (iii) multi-state Markov models, for handling missing outcome data under the missing at random assumption. We find a form ... (continues)