Diagnosing Alzheimer's disease--non-clinicians and computerised algorithms together are as accurate as the best clinical practice.

BACKGROUND: An accurate diagnosis of Alzheimer's disease and an exclusion of other dementias is important in many clinical studies. Obtaining such a clinical diagnosis in epidemiological studies and clinical trials that recruit large numbers of patients is time consuming.
OBJECTIVES: To construct an algorithm using a limited number of data points to generate a diagnosis of the commonest forms of dementia using information collected by non clinicians.
METHODS: We constructed a computer algorithm to generate a diagnosis of Alzheimer's disease (AD), Dementia with Lewy Bodies (DLB), frontotemporal dementia (FTD), vascular dementia or to flag the case as needing a clinical review based on a limited number of data points taken from a largely structured interview using widely used scales. The diagnosis generated in life by the algorithm in a prospective, longitudinal study was compared to definitive diagnosis at post mortem.
RESULTS: Post mortem diagnosis was available for 43 cases. The positive predictive value of the algorithm was greater than 95%. AD was diagnosed by the algorithm and at post mortem in 36 of the cases. Two cases with FTD were wrongly diagnosed as having AD by the algorithm, five cases were flagged as needing a clinical review due to concomitant medical conditions of whom four had AD and one, who had been diagnosed clinically as having AD, was diagnosed on post mortem with corticobasal degeneration.
CONCLUSIONS: A combination of non-clinical researchers, a structured interview and a computerised algorithm is as effective at identifying AD as highly trained and skilled clinicians. Copyright 2007 John Wiley & Sons, Ltd.