|
Abstract:
|
Recently, we have have begun to use HMM models as a tool to help understand cogn itive aging processes within the MCSA, a long term population based study. A HM M can be visualized as a model with two layers. The first layer is an ordinary multi-state hazard model based on a set of true underlying states, which exists in continuous time. The second layer is the set of observations we make on the subjects, episodic in time, containing items that are related to the true states but are an imprecise measurement of them. A primary advantage of the HMM is tha t it can accomodate a large number of such markers, which can differ from patien t to patient, with variable intervals between patients. Such data is a reality in our studies: subsets have received different batteries of cognitive tests and /or biomarkers, and planned visit intervals quickly become irregular. Disadvanta ges include computational challenges induced by the unequal spacing, and how to present and explain these more complex models. This talk will focus on practica l application to the MCSA, and on the insights gained from this approach , touching only briefly on underlying formulas and computational issues.
|
