Background: Few norms exist for the elderly on the cognitive tests commonly used to screen for dementia; conventional cutpoints used in clinical settings may be of limited value in population screening. A particular problem is posed by elderly populations with low educational levels, as performance on most cognitive tests is affected by education. Thus, a healthy but poorly educated population may obtain test scores in the range considered impaired in the clinical setting.
Methods: A random sample of 1,367 subjects aged 65+ years was screened for dementia in a rural community in Southwestern Pennsylvania. Two sets of cognitive measures were used: a global cognitive scale (the MMSE) and a brief battery of tests tapping a variety of cognitive domains. Rather than using a priori cutoff scores, we examined the specificity and sensitivity for dementia of two operationally defined levels of cognitive impairment, at the 5th and 10th percentiles of the study sample on each set of measures.
Results: Results suggest that the screening of multiple cognitive domains at the 10th percentile had significantly greater sensitivity but not lower specificity for definite dementia than did the use of the single global scale.
Conclusion: Our data support the use of population-based cutpoints over standard cutoff scores, in that the global scale at the conventional cutoff was less sensitive than the battery at the same percentile, and because adequate norms do not exist for tests such as those in the battery.