Statistical Methods for Longitudinal Data and Meta-Analysis

One aspect of my work is to develop statistical methods to analyse data from the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS). Data from this trial consists of up to six annual sets of questionnaires from each participant. The questionnaires assess anxiety, sexual function and acceptability of the screening process. I will model how these three processes change over time, and how they depend on one another. There are several approaches to analysing this type of data, and I will explore the pros and cons of three different methods. The challenge in this work lies in the complexity of the UKCTOCS data-set. I will also use some of these methods to analyse data from the MRC Cognitive Function and Ageing Study.

A second aspect of my work is to develop methods for use in meta-analysis. Meta-analysis refers to the pooling of data from different studies, and is used to provide an overview of the available evidence. I will investigate methods for the meta-analysis of data which consists of times to events. I will also explore the meta-analysis of treatment `networks', investigating the relative effects of a set of treatments for a given condition.

Objective: to carry out a diverse program of original research in biostatistics, incorporating two broad research areas, longitudinal data analysis and meta-analysis.
Longitudinal data analysis: My research will be motivated by two studies, the MRC Cognitive Function and Ageing Study (CFAS) and the UK Collaborative Trial of Ovarian Cancer Screening (UKCTOCS). Regarding MRC CFAS, our aim is to model simultaneously two processes - cognitive impairment and death. From this study we have interval-censored data with possibly informative loss-to-followup. We will use a five-state model, with loss-to-followup as an explicit state in the model. We will fit the model using maximum likelihood, imposing constraints to make all parameters identifiable. For the UKCTOCS trial we again have longitudinal data in the form of annual questionnaires. Our aim is to model simultaneously three outcomes of the trial, anxiety, sexual function, and the acceptability of two screening procedures. We will compare three different methods of analysis, random effects, dynamic variables and multi-state models. We will explore the pros and cons of these three methods in addressing the objectives of the UKCTOCS study, such as the effect of treatment allocation on outcomes, and identifying causal relations between the outcomes. Additional complexities which we will address are the inclusion of baseline covariates, adjustments for individual/treatment interactions, the possibility of informative loss-to-followup and multiplicity.
Meta-analysis: My research will focus on the individual patient data meta-analysis of time-to-event outcomes and on multiple treatment meta-analysis. For the meta-analysis of time-to-event outcomes we will explore alternatives to the use of the proportional hazards assumption, for example the use of accelerated failure time models. We will also develop non-parametric methods for constructing a summary survival curve. In multiple treatment meta-analysis we will investigate methods of allowing for inconsistency between direct and indirect evidence. We will model the inconsistency as a treatment by study `design' interaction, where the `design' of a study here means the set of treatments used in the study.
Scientific opportunities: The applications of my research are very topical, and results will be of interest to the wider medical community. In addition, methodological developments will be of use to the statistical community in a variety of other applications.