An algorithmic approach to the differential diagnosis of dementia.

Abstract

The careful definition of cases is fundamental to diagnosis and to any study of cognitive, behavioural and functional problems in dementia. This paper presents an algorithmic approach which mimics a crucial component of diagnostic decision-making; symptoms and signs do not occur independently, but are conditioned on each other. First, we examine whether the conditioned items can be assembled to yield a differential diagnosis of dementia which corresponds to clinical diagnoses, and second, we explore whether subjects whose algorithmic profiles do not fit the clinical diagnoses form new discernable patterns. Such a technique offers two advantages: it allows for the development of validation protocols which are crucial to epidemiological studies, and it allows for the analysis of new patterns of signs and symptoms for emerging criteria of dementia subtypes. This approach has the potential to refine and enhance criteria for the differential diagnosis of dementia and to have an impact on case identification and assessment, particularly in large epidemiologic studies.