Alternative statistical approaches to identifying dementia in a community-dwelling sample

Abstract

Little attention has been paid to examining the extent to which alternative statistical models may facilitate identification of persons with dementia. Using a sub-sample of the Duke Established Populations for Epidemiologic Studies of the Elderly, two analytical approaches were compared: logistic regression (which focuses on identifying specific characteristics predictive here of dementia), and recursive partitioning methods using tree-based models (which permit identification of the characteristics of those groups with high dementing disorder). In the stepwise multiple logistic regression model which included as potential predictors, gender, age, history of chronic health conditions, scales of basic and instrumental activities of daily living (IADL), and cognitive status, only IADL and cognitive status were significant predictors, with cognitive status the single most important factor. The classification tree approach, which permits identification of the characteristics of those groups with particularly high dementia rates, identified cognitive status as the most important criterion for dementia (as did logistic regression analysis). Among those without cognitive impairment, older age was a risk factor, confirming findings consistently reported in the literature. Among the cognitively impaired, IADL was an important risk factor. Those with five or more IADL problems were further classified into two risk groups, based on number of ADL problems. While classification tree analysis encourages identification of groups at risk, logistic regression encourages targeting of specific characteristics.