Lay person-based screening for early detection of Alzheimer's disease: development and validation of an instrument.

Symptoms of cognitive impairment reported to telephone interviewers by caregivers of 272 patients were analyzed with respect to research diagnoses of dementia. All patients received neuropsychological evaluation for establishing the research diagnoses. A data mining program that used machine learning algorithms produced an optimized binary decision tree for differentiating patient groups according to all available information. The results of this analysis were used to help four dementia experts create a dementia screening instrument amenable to application and scoring by nonclinical personnel. The validity of the resulting instrument was then evaluated in an independent sample of 103 patients administered neuropsychological testing within the previous 60 days. The psychometric properties of the empirically derived scale and its performance for discriminating control from probable or possible Alzheimer's patients indicate strong potential for use as a dementia screener for the general population.