Concurrent validity of Dubois' five-word-test compared with CERAD memory subtests

Abstract

Background: We have recently demonstrated the use of an innovative and sensitive behavioral assay, the Visual Paired Comparison (VPC) task, to successfully predict the transition from healthy aging to mild cognitive impairment (MCI), and fromMCI to Alzheimer’s disease (AD) (Zola, et al., 2013). For a subset of subjects, however, predictability is less clear (gray zone). We report here that the application of Machine Learning algorithms and eye movement characteristics can significantly improve the predictive ability of the task for this subset of individuals. Methods: The VPC task measures how individuals view side-by-side novel and familiar images. Normal individuals typically look at the novel image about two-thirds of the viewing time. Previous work has relied on a single parameter, i.e., percent time viewing the novel image (novelty preference, NP). In the present study we used a Machine Learning-based approach that uses additional variables, namely, distribution of NP across multiple trials, and latent factor decompositionbased analysis of eye movement trajectories, to develop amore accurate prediction algorithm. Results: For subjects in the gray zone, NP alone is not sufficient to distinguish between the subjects who will convert to MCI/AD and those who will not, resulting in ROC AUC of 0.70. Our new approach achieves AUC of 0.87, which represents a 24% relative improvement in predictive ability for the subjects in the gray zone. Conclusions: The VPC task can offer early detection of impending cognitive decline as much as 3 years prior to clinical diagnosis. To our knowledge, this is the first behavioral task that can identify healthy subjects who will soon develop MCI and possibly progress to AD within the next few years. For individuals in the gray zone, Machine Learning approaches can enhance still further the predictive capabilities of the VPC task. Finally, when combinedwith biomarker assays, this approachmay provide a newway of interrogating the underlying dynamics of AD, as well as other illnesses in which impaired cognition is a leading symptom.