Project Funding Details

Aim To develop longitudinal statistical methods for disease prediction in individuals, which are accessible to applied statisticians and epidemiologists. Objectives 1. To develop, evaluate and apply statistical methods that allow for complex within individual correlation structures, non-ignorable dropout, left censoring of measurements, non-linear trends and temporal changes in correlation between multiple biomarker trends. 2. To generate guidelines for conducting analyses, and produce accessible software implementing the developed methods that can be used by applied statisticians and epidemiologists. Methodology The proposal concerns the development of statistical methods to model one or more biomarkers whilst accounting for the issues stated in objective 1 that can complicate longitudinal analysis. The inclusion of non-stationary stochastic random terms, such as the integrated Ornstein-Uhlenbeck process will be used to model the within-individual correlation structure. Joint modelling of the longitudinal data and one or more dropout processes will account for informative dropout. Maximum likelihood methods will be used to analyse linear mixed models with left censoring. Non-linear trends in one or more biomarkers will be modelled by fractional polynomials. All developed methods will be applied to datasets collected on patients with progressive diseases: HIV, prostate cancer and multiple sclerosis. Developed methods will be implemented as freely available routines in widely used statistical packages Stata and R. Scientific/medical opportunities The developed methods will be used to answer important clinical research questions of disease prediction in individuals in three medical research areas. The provision of methods, guidelines and software for use by applied statisticians and epidemiologists will also result in clinical advancement in other areas too.