Design and Data Analytics PIA: Bayesian multi-timescale modeling for disentangling cognitive decline and learning effects in repeated mobile cognitive assessments

Recorded: February 22, 2021

In the preclinical phase of Alzheimer’s disease and related dementias, early detection of subtle cognitive decline over time is crucial for facilitating the development of interventions. This endeavor, however, is complicated by the fact that multiple, multi-timescale factors influence changes in cognitive performance measures over time, and requires careful statistical modeling. With the double exponential learning model, we can account for learning effects and stochastic fluctuations in longitudinal studies with high-frequency assessments (i.e., measurement burst studies). This webinar demonstrates how a Bayesian multilevel implementation of the double exponential model allows for the inclusion of person-level predictors on learning model parameters and drawing intuitive inferences on cognitive change with Bayesian posterior probabilities. Mobile cognitive task performance data is used to show how individual differences in peak performance and change therein can be related to predictors such as age and MCI status.

Moderator: Micheal Donohue

Panelists: Zita Oravecz