Optimizing the Color Shapes Task for Ambulatory Assessment and Drift Diffusion Modeling: A Factorial Experiment.

backgroundRecent advances in cognitive digital assessment methodology, including high-frequency, ambulatory assessments, promise to improve the detection of subtle cognitive changes. Computational modeling approaches may further improve the sensitivity of digital cognitive assessments to detect subtle cognitive changes by capturing features that map onto core cognitive processes.

objectiveWe explored the validity of a brief smartphone-based adaptation of a visual working memory task that has shown sensitivity for detecting preclinical Alzheimer disease risk. We aimed to optimize properties of the task for computational cognitive feature extraction with drift diffusion modeling.

methodsWe analyzed data from 68 participants (n=47, 69% women; n=55, 81% White; mean age 49, SD 14; range 24-80 years) who completed 60 trials for each of 16 variations of a visual working memory binding task (the Color Shapes task) on smartphones, over an 8-day period. A drift diffusion model was fit to the response time and accuracy data from the task. We experimentally manipulated 3 properties of the Color Shapes task (study time, probability of change, and choice urgency) to test how they yielded differences in key drift diffusion model parameters (drift rate, initial bias toward a response option, and caution in decision-making). We also evaluated how an additional task property, the test array size, impacted responses across all conditions. For array size, we tested a whole display of 3 shapes against a single probe of 1 shape only.

resultsThe 3 task property manipulations yielded the following results: (1) increasing the ratio of different responses was credibly associated with higher initial bias toward the different response (mean 0.06, SD 0.02 for the whole display; mean 0.15, SD 0.02, for the single probe condition); (2) increasing the choice urgency during the test phase was credibly associated with decreased caution in decision-making in the single probe condition (mean -0.04, SD 0.02) but not in the whole display (mean -0.01, SD 0.02); and (3) contrary to expectation, longer study times did not yield a credibly faster drift rate but produced credibly slower ones for the whole display condition (mean -0.28, SD 0.05) and a null effect for the single probe condition (mean 0.01, SD 0.05). In addition, as expected, we found that individual differences in drift rate were associated with age in both array sizes (r=-0.45 with Bayes factor=191), with older participants having a slower drift rate. Older participants also showed higher caution (r=0.42 with Bayes factor=80.76) in the single probe condition.

conclusionsWe identified a version of the Color Shapes task optimized for smartphone-based cognitive assessments in real-world settings, with data designed for analysis through computational cognitive modeling. Our proposed approach can advance the development of tools for efficient and effective early detection and monitoring of risk for Alzheimer disease.