Models of progression in AD

Three articles in this issue of Neurology use AD patient databases to develop models that predict patient movement among the stages of AD.1-3⇓⇓ These articles illustrate how models can be built based on many short observation intervals (months) to predict the behavior of a disease such as AD which takes years to run its course. Costs and progression are associated with either the group or the individuals in each stage to determine how much would be saved if a treatment alters the dynamics of disease progression. The concept of model-building is not new to the AD literature,4-6⇓⇓ in which earlier models explored cognitive changes using different types of regression analyses instead of survival approaches. However, caution must be exercised when interpreting results from any model that is built from a particular dataset because the characteristics of the underlying data can limit their clinical application.

Neumann et al.1 used the Consortium to Establish a Registry in AD (CERAD)7 database to estimate the probability of moving among the mild, moderate, and severe stages of AD with transition to death possible from any of these stages. They modeled the holding time of an individual in each AD stage using the Cox proportional hazard statistical model that adjusts for patient heterogeneity defined by patient …