Methods for deterministic treatment effect estimates for clinical trials with missing data

点击翻译按钮获取中文摘要

The statistical analysis of clinical trials is often complicated by missing data. Several methods have been developed for obtaining estimates from trials which accommodate such missingness in a principled way, by making an assumption about how the missing data relate to the observed data. However, these untestable assumptions cannot be verified from the observed data. Multiple imputation is applied to handle missing data, however, a drawback of this methodology is that the imputations are drawn randomly, and hence the treatment effect estimates obtained are random. This results in slightly different answers depending on the computer's random number seed. For the primary analysis of a clinical trial, this is quite undesirable and deterministic methods would be much preferable. This project aims to investigate deterministic single imputation methods for handling missing data in clinical trials, as an alternative to multiple imputation. This project will allow me to explore, expand and develop novel methods and models used for analysis of missing data. Application of these methods will improve the precision of treatment estimates in clinical trials and hence help to generate clinically useful results. Through this project I will develop skills and experience in the development and application of novel statistical methodologies for application in clinical trial analyses, using both analytical and simulation-based approaches. This research will also enable me to gain statistical programming skills using modern software. I will gain experience in working collaboratively within the environment of an internationally leading biotech company. I will gain Transferrable skills, including research ethics, scientific writing, and presentation skills through the Doctoral Transferable Skills Programme at LSHTM and the UBEL DTP Core training program. Overall the studentship will result in the development of my talent and skills in addition to the development of the novel statistical methods which will be applicable in research.