Abstract
Background
The Memory Binding Test (MBT) Version 1 demonstrated significant predictive validity for incident amnestic MCI (aMCI) and incident dementia. Version 2 of the MBT was developed to be briefer because brevity is essential for screening. Herein we aimed to compare the two versions in terms of predictive validity for aMCI and dementia, separately.
Method
As sub‐studies of the Einstein Aging Study (EAS), the two MBT versions were administered to two independent study samples between May 2003 and December 2007 at baseline, and these participants were followed up to Jan 2017 (For differences in versions see Table 1). The EAS enrolls a systematically recruited community sample of adults age over 70. The two versions were evaluated in parallel using the same methodologies. We evaluated a range of cut‐scores on the MBT score of choice, the Total Items in the Paired condition (TIP), including the empirical optimal cut‐score which maximized the sum of sensitivity and specificity from the cross‐sectional discriminative analyses. The predictive validity was assessed by Kaplan‐Meier curves, log‐rank tests and Cox regressions.
Result
Predicting incident aMCI: Both versions were strong predictors of incident aMCI (Table 2): For the empirical optimal scores for TIP, hazard ratio (HR)=2.33, 95% CI: (1.26, 4.29), p=.007 for Version 1, HR=3.38, 95% CI: (1.09, 10.5), p=0.04 for Version 2; for a range of TIP scores (17−22): HR range: 2.27−6.07, p ≤ .01, for Version 1, and 2.67−5.91, p≤.04 for Version 2. Predicting incident dementia: Both versions were strong predictors of incident dementia (Table 3): For the empirical optimal scores for TIP, HR=8.32, 95% CI: (3.51, 19.7), p<.0001 for Version 1, HR=9.30, 95% CI: (3.43, 25.2), p<.0001 for Version 2; for a range of TIP scores (17−23): HR range: 3.09−8.32, p ≤ .003, for Version 1, and 6.40−9.30, p<.0001 for Version 2.
Conclusion
The predictive validity for incident aMCI and incident dementia was strong and similar for the two versions of the MBT based on the overlapping confidence intervals. These results strongly support the hypothesis that poor performance on memory binding is an important marker for the early detection of aMCI and dementia.