Herman Buschke, Wenzhu B. Mowrey, Wendy S. Ramratan, Molly E. Zimmerman, David A. Loewenstein, Mindy J. Katz, Richard B. Lipton

We aimed to assess reliability and cross-sectional discriminative validity of the Memory Binding Test (MBT) to distinguish persons with amnestic cognitive impairment (aMCI) and dementia from cognitively normal elderly controls. The MBT was administered to 20 participants with dementia, 31 with aMCI and 246 controls, who received the first administration of the MBT from May 2003 to December 2007, as a substudy of the community-based Einstein Aging Study (age range: 70+). The optimal index resulted from comparing the partial area under the receiver operating characteristic curves (ROC AUC) of four major MBT indices for specificities ≥0.70. Optimal cut-score of the optimal index was selected by maximizing the sum of sensitivity and specificity. Age and education effects were assessed using stratified cut-scores and adjusted logistic regression. Reliability was computed as intraclass correlation between scores at baseline and 1-year follow-up for participants who remained cognitively normal. Total number of Items recalled in the Paired condition (TIP) was elected the optimal index. TIP cut-score was ≤22 for differentiating aMCI alone (sensitivity = 0.74, specificity = 0.73) and aMCI and dementia combined (sensitivity = 0.84, specificity = 0.73) from controls. It was ≤17 for differentiating dementia from aMCI and controls (sensitivity = 0.95, specificity = 0.87). Age and education adjustments did not materially improve discriminative validity. The reliability of TIP was 0.77. MBT achieved moderate to good reliability. TIP had superior cross-sectional discriminative validity than the other MBT indices. We recommend using the empirical cut-score of TIP ≤22 for discriminating aMCI and dementia and ≤17 for discriminating dementia alone.