INTRODUCTIONThe aim of this study was to assess the usefulness of 4 clinical prediction rules, the neuroimaging guidelines from the Canadian Consensus Conference on Dementia (CCCAD) and the modified Hachinski's Ischaemic Score (HIS) in identifying patients with suspected dementia who will benefit from neuroimaging.

MATERIALS AND METHODSTwo hundred and ten consecutive patients were referred to the memory clinic in a geriatric unit for the evaluation of possible dementia. Sensitivity, specificity and likelihood ratios (LR) were calculated for each of the prediction rules and the CCCAD guidelines, in terms of their ability to identify patients with significant lesions [defined firstly as space-occupying lesions (SOL) alone and secondly as SOL or strokes] on neuroimaging. Similar analyses were applied for the HIS in the detection of strokes.

RESULTSWhen considering SOL alone, sensitivities ranged from 28.6% to 100% and specificities ranged from 21.7% to 88.4%. However, when strokes were included in the definition of significant lesions, sensitivities ranged from 16.2% to 79.0% and specificities ranged from 20.9% to 92.4%. The modified HIS had a similarly poor sensitivity and specificity (43.3% and 78.9% respectively). The LR for the clinical decision tools did not support the use of any particular instrument.

CONCLUSIONSClinical decision tools do not give satisfactory guidance for determining the need for neuroimaging patients with suspected dementia, when the detection of strokes, in addition to SOL, is regarded as important. We recommend therefore that neuroimaging be considered for all patients with suspected mild or moderate dementia in whom the potential benefits of any treatment outweigh the potential risks.