Classification of amyloid‐positivity using memory scores in subjects with early‐onset Alzheimer’s disease

Background The Rey Auditory Verbal Learning Test (RAVLT) is a common neuropsychological test for characterizing episodic memory deficits. Performance of the RAVLT has been described in numerous presentations of cognitive impairment. However, there is limited research for analyzing the predictive val...

Full description

Saved in:
Bibliographic Details
Published in:Alzheimer's & dementia 2023-12, Vol.19 (S24), p.n/a
Main Authors: Bushnell, Justin, Hammers, Dustin B., Apostolova, Liana G., Kramer, Joel H., Carrillo, Maria C., Dickerson, Brad C., Rabinovici, Gil D., Clark, David G.
Format: Article
Language:English
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Background The Rey Auditory Verbal Learning Test (RAVLT) is a common neuropsychological test for characterizing episodic memory deficits. Performance of the RAVLT has been described in numerous presentations of cognitive impairment. However, there is limited research for analyzing the predictive value of RAVLT scores. We examine the ability of the RAVLT to classify individuals with early‐onset cognitive impairment in terms of amyloid positivity. Method We transcribed RAVLT recordings from 214 cognitively impaired subjects in the Longitudinal Early‐Onset Alzheimer’s Disease Study (LEADS). Each RAVLT test consists of eight tasks: five learning trials, an interference list, short delay free recall, and long delay free recall. Subjects were stratified by amyloid positivity (EOAD vs. EOnonAD). We calculated traditional and novel scores based on counts (raw score, primacy, recency) and timings (stopping time and speed). To split the data into training and test sets, we selected subjects such that Cohen’s D was below 0.05 between nuisance covariates (age and education) and included similar proportions of each sex. Due to the imbalance of classes in the data set, we generated 1,000 equal‐sized bootstrap samples of cases and controls (n = 200) and fit a random forest classifier for each sample. Subjects in the test set were classified by majority vote. This classification process was applied to three different sets of RAVLT variables for comparing performance. Result With raw scores and demographics alone, we achieved AUC of 0.839. We obtained an AUC of 0.841 using variables with possible utility identified in previous work. The best model used only speed scores and demographic variables, with an AUC of 0.858. Conclusion We marginally improved the classification accuracy of amyloid‐positive subjects over raw scores alone. Further work is needed to determine the utility of novel scores in predicting biomarker positivity.
ISSN:1552-5260
1552-5279
DOI:10.1002/alz.083119