Statistical methods for analyzing risk of Alzheimer's Disease and biomarker measurements

PROJECT SUMMARY The objective of this project is to develop and apply statistical methods to improve scientific inferences in Alzheimer's disease (AD) research when cross-sectional or longitudinal study designs are employed. The proposed approaches are largely motivated by three large-scale studies (BIOCARD, ABC-DS, PAC) to study AD, but can also be applicable to study other chronic diseases. The study team plans to de- velop efficient and statistically proper methods to better model, estimate, and predict the risk of disease incidence and longitudinal biomarker outcomes. The proposed approaches will help identify individuals who are at a high risk of experiencing incidence of disease (MCI or AD) so that these individuals can be targeted for more intensive monitoring or treatment. The proposed research includes the following aims: Aim 1: to develop statistical methods to handle age-specific prevalent cases in case-control stud- ies. Prevalent AD cases identified from cross-sectional populations are frequently sampled and included in case-control data. When adopting such a sampling design, survivor bias could form a serious problem to lead to biased analysis results. Existing approaches typically treated cases and controls in binary form without specifying age at incidence of disease, and the prevalent cases are sampled with length bias sam- pling in stationary models. Instead of binary disease outcome, we consider age-specific risk outcome and propose a composite likelihood approach which handles survival bias in either stationary or non-stationary models. Aim 2: to develop regression methods for analyzing age at biomarker positivity and the remaining time to onset of clinical symptoms. In some longitudinal studies of AD, MCI-free subjects are recruited for follow-up of the subsequent development of AD-related clinical symptoms. Meanwhile, age at biomarker positivity, a crucial characterization of the stage of biomarker deterioration, is of interest and retrospectively identified. This type of recruitment creates a special sampling structure that we refer to as partial left truncation. Under Aim 2, we will develop estimating equation methods and computa- tional techniques that correct the sampling bias in the presence of competing risks due to death without disease, and study a bivariate failure time model to analyze age at biomarker positivity and the remain- ing time to onset of clinical symptoms of MCI. Aim 3: to develop statistical methods for biomarker trajectories prior to onset of clinical symptoms. AD biomarkers often undergo changes years before the emergence of clinical symptoms. Existing studies often model biomarker trajectories using a forward time scale such as age and time since study baseline. Trajectory models are proposed to examine the changes and trend of biomarkers prior to the onset of clinical symptoms. We propose semiparametric procedures to estimate the marker trajectories stratified by the age at onset of clinical symptoms and the type of failure events, in the presence or absence of knowledge on the shape of the trajectory.