Validation of a latent dementia index in the Aging, Demographics, and Memory Study

Abstract

AbstractBackgroundLongitudinal studies of aging often lack clinical gold standard measures of dementia and instead rely on nondiagnostic approaches to categorizing study participants as demented or cognitively normal. Recently, latent variable models have been used to estimate latent dementia indexes (LDI) using cognitive ability, memory, and functional ability measures to distinguish dementia risk from general cognitive functioning. The LDI exhibits clinical utility and puts dementia risk on a continuous scale, to approximate the process of becoming demented. The purpose of the current study is to validate an LDI in the Aging, Demographics, and Memory Study (ADAMS) using participants’ neuropsychological assessment and functional ability scores.MethodWe evaluated the 2001‐2003 ADAMS including 856 adults age 65+ who had participated in the United States Health and Retirement Study (HRS). ADAMS respondents were given an extensive neuropsychological battery of tests spanning various cognitive domains and provided information on functional abilities. We analyzed cognitive tasks (trails A&B, digits forward/backward, Shipley VocabularyTest, delayed 10‐word recall, recognition memory, and logical memory) and functional tasks (difficulty with household tasks, using familiar appliances, shopping independently, ability with complicated tasks, and handling small sums of money). Using confirmatory factor analysis, we modeled the LDI by performance on cognitive, memory, and functional ability reports. Validation involved comparing the LDI to the diagnosis of dementia made by a consensus panel including neuropsychologists, neurologists, geropsychiatrists, and internists.ResultsA single factor LDI model fit the data well (Comparative Fit Index = .96; Root Mean Square Error of Approximation = .05, .90CI: [.03, .06]). The LDI significantly predicted clinical gold standard diagnosis (OR = .55, .95CI = [.50, .60]), with LDI scores +1 SD above the LDI mean predicting .36 times lower odds of dementia and LDI scores ‐1 SD below the mean predicting 2.78 times greater odds of dementia.ConclusionAn LDI that does not rely on clinical gold standard dementia diagnoses has good clinical utility in the ADAMS 2001‐2003 sample. Future research includes validating LDI scores longitudinally in ADAMS and extending the LDI approach to the HRS core sample.