Statistical modelling of cognitive decline to improve estimation, identifying risk factors and population benefits.

Age related cognitive decline is a major public health concern. Most statistical methods used to measure it model the average change over time. However, scientists have reached a consensus that cognitive decline is too complex to be described as an overall mean, and rather person specific models are needed to fully understand it. I will investigate the effect of missing data due to death and ill health on estimates of cognitive decline, an investigation that will produce more realistic estimates of true cognitive decline and its potential causes. Core hypotheses of two large population based studies of older individuals will be investigated. These studies are the CC75C and the MRC CFA study.
The work proposed on the models of cost effectiveness of drugs for Alzheimer's disease will combine epidemiological data and clinical trials data within a sound framework that introduces realism into policy decisions. Models of individuals' trajectories combined with cost information and outcomes from clinical trials will provide a robust framework of the true cost of treatment for dementia.

Background: Longitudinal studies provide the framework to investigate age related cognitive change. This modelling is complex due to factors such as the shape of the change, study design and missing data; factors that are often ignored. Preventing and slowing cognitive change is a target for drug interventions including those aimed at individuals with Alzheimer's disease. However, these investigations also suffer from similar modelling complexities. Investigating cost-effectiveness must consider these difficulties and also difficulties in modelling treatment effect at the individual and carer level.
Aims: To explore parametric random effects and growth mixture models that will allow for the identification of potential risk factors for cognitive decline, either for all participants or class specific groups including the modelling of missing data. To improve health economic models for Alzheimer's disease drugs by modelling the epidemiological setting as well as the clinical effectiveness.
Methodology:
" Random change point models will be fitted to investigate terminal decline with missing data using selection model approach (Bayesian analysis in WinBugs)
" Growth mixture models will investigate the analysis of class specific risk factors including the potential for test re-test effects (Using Mplus).
" Extension of investigation of the time matrix for analysis (such as time in study, age and time before death).
" Complex study design issues will be investigated via the incorporation of sampling weights into the models.
" Investigation of analysis modelling of missing data (including non ignorable data) such as inverse probability weighting, multiple imputation and pattern mixture models.
" The Cambridge City over 75 Cohort study, MRC Cognitive Function and Ageing Study and simulation studies will be used to develop the methodology and produce the applied papers.
" A literature review will initially be conducted to produce all necessary evidence regarding economic models for cholinesterase inhibitors.
" A Bayesian evidence synthesis will be undertaken to include the effect of Alzheimer's disease on mortality, the severity of the disease, the effect of transfer to institutional care as well as carers and the individuals' response to treatments. Initially the cholinesterase inhibitors will be modelled, but the framework will be flexible should new treatment strategies arise.
Opportunities. Proper models will describe potential risk factors for cognitive decline that could have population impact. Cost-effectiveness models using Bayesian synthesis for Alzheimer's disease will provide results of direct clinical usage.