Novel Virtual Reality Assessment of Functional Cognition

 Lilla Alexandra Porffy ¹  ;  Mitul A Mehta ¹  ;  Joel Patchitt ^{1, 2}  ;  Celia Boussebaa ¹  ;  Jack Brett ³  ;  Teresa D’Oliveira ¹  ;  Elias Mouchlianitis ⁴  ;  Sukhi S Shergill ^{1, 5, 6} 

doi: 10.2196/27641

PDF: https://www.jmir.org/2022/1/e27641/PDF

Abstract

Background:Cognitive deficits are present in several neuropsychiatric disorders, including Alzheimer disease, schizophrenia, and depression. Assessments used to measure cognition in these disorders are time-consuming, burdensome, and have low ecological validity. To address these limitations, we developed a novel virtual reality shopping task—VStore.

Objective:This study aims to establish the construct validity of VStore in relation to the established computerized cognitive battery, Cogstate, and explore its sensitivity to age-related cognitive decline.

Methods:A total of 142 healthy volunteers aged 20-79 years participated in the study. The main VStore outcomes included verbal recall of 12 grocery items, time to collect items, time to select items on a self-checkout machine, time to make the payment, time to order coffee, and total completion time. Construct validity was examined through a series of backward elimination regression models to establish which Cogstate tasks, measuring attention, processing speed, verbal and visual learning, working memory, executive function, and paired associate learning, in addition to age and technological familiarity, best predicted VStore performance. In addition, 2 ridge regression and 2 logistic regression models supplemented with receiver operating characteristic curves were built, with VStore outcomes in the first model and Cogstate outcomes in the second model entered as predictors of age and age cohorts, respectively.

Results:Overall VStore performance, as indexed by the total time spent completing the task, was best explained by Cogstate tasks measuring attention, working memory, paired associate learning, and age and technological familiarity, accounting for 47% of the variance. In addition, with λ=5.16, the ridge regression model selected 5 parameters for VStore when predicting age (mean squared error 185.80, SE 19.34), and with λ=9.49 for Cogstate, the model selected all 8 tasks (mean squared error 226.80, SE 23.48). Finally, VStore was found to be highly sensitive (87%) and specific (91.7%) to age cohorts, with 94.6% of the area under the receiver operating characteristic curve.

Conclusions:Our findings suggest that VStore is a promising assessment that engages standard cognitive domains and is sensitive to age-related cognitive decline.

https://www.jmir.org/2022/1/e27641