Abstract
OBJECTIVE: Memory and cognitive problems are central to the diagnosis of Alzheimer's disease (AD). Psychometric approaches to defining phenotypes can aid in identify genetic variants associated with AD. However, these approaches have mostly been limited to affected individuals. Defining phenotypes of both affected and unaffected individuals may help identify genetic variants associated with both AD and healthy aging. This study compares psychometric methods for developing cognitive phenotypes that are more granular than clinical classifications. METHODS: 682 older Old Order Amish individuals were included in the analysis. Adjusted Z-scores of cognitive tests were used to create four models including (1) global threshold scores or (2) memory threshold scores, and (3) global clusters and (4) memory clusters. An ordinal regression examined the coherence of the models with clinical classifications (cognitively impaired [CI], mildly impaired [MI], cognitively unimpaired), APOE-e4, sex, and age. An ANOVA examined the best model phenotypes for differences in clinical classification, APOE-e4, domain Z-scores (memory, language, executive function, and processing speed), sex, and age. RESULTS: The memory cluster identified four phenotypes and had the best fit (χ(2) = 491.66). Individuals in the worse performing phenotypes were more likely to be classified as CI or MI and to have APOE-e4. Additionally, all four phenotypes performed significantly differently from one another on the domains of memory, language, and executive functioning. CONCLUSIONS: Memory cluster stratification identified the cognitive phenotypes that best aligned with clinical classifications, APOE-e4, and cognitive performance We predict these phenotypes will prove useful in searching for protective genetic variants.