P3–301: Clinical dementia rating modeling and simulation: Modeling and simulation of CDR progression in the ADNI cohort based on item response theory

Background: Future clinical trials in prodromal AD are expected to rely on CDR as an important efficacy endpoint. Sensitivity of the CDR in the prodromal population is favorable compared to other common clinical endpoints. Statistical models for the longitudinal progression of CDR scores provide a basis for more insightful analysis of clinical trial data, as well as a basis for better prospective understanding of the operating characteristics of candidate trial designs through simulating from the model. Models that describe CDR progression as a function of demographic and biomarker covariates may be used to evaluate the likely impact of various trial enrichment strategies through simulation. Simulation-based trial design has already been successfully applied in the context of mild-to-moderate AD, and our current effort seeks to enable similar approaches for prodromal AD. Methods: The CDR is comprised of 6 subscores, but analysis of the CDR often focuses on an aggregate score, the sum of boxes, that weights each item equally. Recent approaches improve upon this using item response theory (IRT), and we extended the IRT approach to include longitudinal and covariate data. Patient subscores were modeled using a graded-response IRT model, where progression was captured as change in the latent disease status over time, controlled by baseline covariates. Results: Modeling the joint progression of individual item responses allowed for better understanding of the item sensitivity in different subpopulations, at different stages of the disease. To illustrate, Figure 1 shows the progression of the probability of a subscore being greater than zero for a patient with baseline MMSE of 29 with (in order of increasing sensitivity): personal care, community affairs, home and hobbies, orientation, judgement and problem solving, and memory. The longitudinal aspect of the model facilitated quantification of progression rates to later stages of dementia, and the covariate model allowed further quantification within subpopulations. Conclusions: Identification and enrichment of populations identified by this approach in trials will lead to better signal detection, smaller trials, and encourages innovation in development. This work demonstrates a modeling approach that utilizes longitudinal data with covariates, greatly expanding on the existing CDR IRTmodeling methodology. P3-302 RECRUITMENT OF PEOPLE WITH AMNESTIC MILD COGNITIVE IMPAIRMENT FOR AN EXERCISE STUDY Candace Hill, Niki Mirshams, Kyle Armstrong, Kristin Martin-Cook, Rong Zhang, Institute for Exercise and Environmental Medicine, Dallas, Texas, United States; Institute for Exercise and Environmental Medicine, Texas Health Presbyterian Hospital Dallas, Dallas, Texas, United States; Alzheimer’s Disease Center, University of Texas Southwestern Medical Center, Dallas, Texas, United States. Contact e-mail: candacehill@ texashealth.org