Diagnosis of dementia from the history and aetiology schedule

Abstract

Objective: To provide a computerised method of diagnosing organic brain syndrome from history data without the use of mental state data.

Methods: Interview dataset from participants in a community study of the incidence of dementia was used to form a training sample and validation sample. The algorithm was developed on the training sample and tested on the validation sample.

Results: Performance in the training and validation samples was very similar. The algorithm shows monotonically increasing probability of being diagnosed with dementia as a function of the proposed level of diagnostic confidence. At the proposed cut point it has sensitivity 94% and specificity 84% for detecting concurrent psychiatrist's diagnosis of dementia.

Conclusions: The method provides a good agreement with psychiatrist's diagnosis, and the results in the validation sample show little shrinkage. The method will prove useful in studies where it has proved impossible to collect mental state information on all the study participants.