Development and validation of the Memory Performance Index: Reducing measurement error in recall tests

Abstract 

Background

The Memory Performance Index (MPI) quantifies the pattern of recalled and nonrecalled words of the Consortium to Establish a Registry for Alzheimer's Disease Wordlist (CWL) onto a 0 to 100 scale and distinguishes normal from mild cognitive impairment with 96% to 97% accuracy.

Methods

In group A, 121,481 independently living individuals, 18 to 106 years old, were assessed with the CWL and classified as cognitively impaired (N = 5,971) or normal (N = 115,510). The MPI and CWL immediate free recall (IFR), delayed free recall (DFR), and total free recall (TFR) scores (the outcome measures) were each regressed against predictors of age, gender, race, education, test administration method (in-person or telephone), and wordlist used. Predictor effect sizes (Cohen's f²) were computed for each outcome. In addition, CWL plus Functional Assessment Staging Tests (FAST) were administered to 441 normal to moderately severely demented (FAST stages 1 to 6) patients (group B). Median MPI scores were tested for significant differences across FAST stage.

Results

For group A, the variance explained by all predictors combined was MPI = 55.0%, IFR = 24.9%, DFR = 23.4%, and TFR = 26.9%. The age effect size on MPI score was large, but it was small on IFR, DFR, and TFR. The other predictors all had negligible (<0.02) or small effect sizes (0.02 to 0.15). For group B, median MPI scores progressively declined across all FAST stages (P < .0002).

Conclusions

MPI score progressively declines with increasing dementia severity. Also, MPI score explains 2 to 3 times more variance than total scores, which improves ability to detect treatment effects.

Keywords: Alzheimer's disease, Mild cognitive impairment, Dementia, Accuracy, Normal aging, Episodic memory, Declarative memory, Short-term memory, Hippocampus, Working memory, Correspondence analysis, Receiver operating characteristic, Logistic regression