Subgroups of Alzheimer’s Disease determined by longitudinal change across three domains of cognition

Background There is evidence that Alzheimer’s Disease (AD) may have distinct subtypes. Most approaches for identifying subgroups use cross‐sectional data. Here we use longitudinal cognitive test data from two large cohort studies to determine if there are distinct patterns of cognitive decline among those with AD and if those patterns differ across three domains of cognition. Method We used data from the Rush Memory and Aging Project and Religious Orders Study (Table 1). Previously, cognitive test items were assigned to a single domain of memory, language, or executive function and scores for each domain were estimated using confirmatory factor analysis. We used Growth Mixture Models to estimate trajectories of decline starting at AD diagnosis and identified subgroups with differing trajectories. Each domain was modeled individually and in a combined model that included distinct latent processes for each domain. Models with one to five classes were estimated and the number of classes was selected using Bayesian Information Criterion (BIC) and entropy. Result The two class models fit best for each of the individual domains (Figure 1). For memory and language one class was characterized by a higher intercept and steeper slope while the other had a lower intercept and flatter slope. For executive function the two classes had similar intercepts, but different slopes. Five classes best fit the data in the combined model. The pattern of decline within a class was similar across the three domains (Figure 2). For each domain, four of the classes were differentiated from each other by lower intercepts and increasingly steep slopes, while a fifth class had the lowest intercept, but a flatter trajectory over time. Conclusion We did not see different patterns across domains of cognition; knowing the trajectory of one cognitive domain explains the most likely trajectory of the other domains. Still, differences identified in patterns of overall decline may be important for AD treatment. Future work will focus on determining biological relevance of subgroups through association with genetic and other biomarkers as well as addressing methodological issues such as incorporating survival time and accounting for the presence of prevalent cases and floor scores.