A semiparametric method for estimating the progression of cognitive decline in dementia

Abstract

ABSTRACTAlthough many patients develop cognitive decline, their trajectories of cognitive decline are diverse and incompletely understood. Accurate prediction of the cognitive decline process is critical for early treatment and management of dementia. We used the Clinical Dementia Rating Score Sum of Boxes (CDR SUM) score as a cognitive decline proxy, and investigated factors that are potentially associated with cognitive decline, including duration, age at onset, sex, education, health-related symptoms, and neuropsychiatric symptoms. We analyzed data from an established 10-year longitudinal patient registry of patients diagnosed with non-normal cognition. We compared a multi-level polynomial regression model and two semiparametric mixed-effects models, and applied Nakagawa and Schielzeth's R2GLMM and correlation coefficient as model selection criteria. The semiparametric method was selected to describe and predict the cognitive decline trajectory. Neuropsychiatric symptoms were indicators of a higher CDR SUM score. History of stroke, presence of disinhibition, and nighttime behavior disturbances were also associated with higher CDR SUM score. Older age of onset (> 86 years), educational level higher than high school education (≥ 12 education years), and the presence of irritation were indicators of slower cognitive decline. The semiparametric model can assist in estimating cognitive decline in terms of CDR SUM score, given individual characteristics.