Sunday, May 18, 2025
5/18/2025
Statistical Methods for Accurate Estimation and Prediction in Alzheimer's Disease - Project Summary Longitudinal cohort studies are a rich resource for estimation and modeling of Alzheimer's disease (AD) progression. Datasets extracted from these studies often feature complex truncation (selection) and censoring. Estimates based on these datasets are used for (1) clinical trial design; (2) improved understanding of AD progression, risk and prevention; and (3) individual prediction, but do not fully account for the complex truncation. This proposal develops methods that make proper adjustments and thereby enhance each of these essential needs in AD. Use of time-to-event endpoints in clinical trials for AD is supported by regulatory authorities when the time origin and the event are clinically meaningful, e.g., onset of cognitive decline and mild cognitive impairment (MCI). Time-to-event estimation and modeling are needed to support trial design (e.g., numbers of events and participants, length of follow-up, eligibility). These estimates are commonly based on subcohorts from large longitudinal cohort studies, which feature complex truncation and censoring. Current estimates use a time origin of convenience, such as study entry, rather than a clinically meaningful time origin, as required by clinical trials. Aim 1 develops methods for estimation and regression that enable use of clinically relevant time origins through proper adjustment for fully and partially sequential truncation in longitudinal cohort studies. Unbiased estimates of associations between measures of interest in studies of AD such as blood biomarker levels or amyloid imaging levels and factors such as maternal age at dementia onset or other biomarker values are important for the evolving understanding of AD progression, which informs drug development and treatment and personalized risk assessment. Many published regression analyses ignore the censoring and “cure” (i.e., unobserved latent classes) of these factors and obtain biased associations. Aim 2 develops methods that adjust for covariates that are subject to censoring or cure and yields accurate estimates of association. Accurate predictions of clinically meaningful times-to-event that adjust for subject-specific features are essential to provide prognostic guidance to patients. Conformal prediction provides a flexible framework for quantifying uncertainty based on the use of arbitrary prediction algorithms. Conformal prediction targets the actual event time itself, which is the most meaningful measure for patients. However, it has not been developed for settings of truncation (Aim 1) or incompletely observed covariates (Aim 2). Aim 3 fills these gaps and will enable use of conformal prediction in the setting of times-to-events that are clinically meaningful in AD progression based on longitudinal cohort studies.
HHS’ Tracking Accountability in Government Grants System (TAGGS) website provides access to detailed descriptions of grants, loans, aggregated direct payments and other types of financial assistance awarded by the HHS Operating Divisions (OPDIVs) and the Office of the Secretary Staff Divisions (STAFFDIVs).