Background and objectives: To avoid costly evaluation of healthy individuals, efficient methods of screening for incident dementia must combine adequate sensitivity and high specificity. Two-stage screening may offer improvements over single-stage methods. We therefore investigated a two-stage screening protocol for incident dementia among 3,308 elderly.

Methods: We administered the Modified Mini-Mental-State (3MS) or, rarely, Jorm's IQCODE, to a validation sample of 441 high-risk respondents. Informants then completed the Dementia Questionnaire (DQ). Finally, all 441 sample members underwent physical, neurologic, and neuropsychologic assessment. We studied the sensitivity and specificity of the 3MS/IQCODE and DQ using Receiver-Operating Characteristic analyses.

Results: A 3MS cut point of 82/83 (of 100) yielded sensitivity and specificity of 91.5 and 90.1%. With 3MS scores of ⩽82, a DQ cut point of 2/3 (of five) yielded conditional sensitivity and specificity of 90.2 and 55.3%. Combining these instruments yielded sensitivity and specificity of 82.5 and 95.6%. Age stratification and use of longitudinal decline score criteria did not materially improve these figures.

Conclusions: The improved specificity of the two-stage approach offers economies that are attractive, particularly if sensitivity can be enhanced, for example, by examination of a high-risk validation sample.