Abstract
The Sustained Attention to Response Task (SART) has been used to measure neurocognitive functions in older adults. However, simplified average features of this complex dataset may result in loss of primary information and fail to express associations between test performance and clinically meaningful outcomes. Here, we describe a new method to visualise individual trial (raw) information obtained from the SART test, vis-à-vis age, and groups based on mobility status in a large population-based study of ageing in Ireland. A thresholding method, based on the individual trial number of mistakes, was employed to better visualise poorer SART performances, and was statistically validated with binary logistic regression models to predict mobility and cognitive decline after 4 years. Raw SART data were available for 4864 participants aged 50 years and over at baseline. The novel visualisation-derived feature bad performance, indicating the number of SART trials with at least 4 mistakes, was the most significant predictor of mobility decline expressed by the transition from Timed Up-and-Go (TUG) < 12 to TUG ≥ 12 s (OR = 1.29; 95% CI 1.14–1.46; p < 0.001), and the only significant predictor of new falls (OR = 1.11; 95% CI 1.03–1.21; p = 0.011), in models adjusted for multiple covariates. However, no SART-related variables resulted significant for the risk of cognitive decline, expressed by a decrease of ≥2 points in the Mini-Mental State Examination (MMSE) score. This novel multimodal visualisation could help clinicians easily develop clinical hypotheses. A threshold approach to the evaluation of SART performance in older adults may better identify subjects at higher risk of future mobility decline.
Highlights
Computer-based neurocognitive tests are commonly utilised in research [1], and increasingly, in clinical practice, for both the detection and rehabilitation of cognitive disorders in adults [2]
We considered the temporal evolution of mobility status, expressed by Timed Up-and-Go (TUG), usual gait speed (UGS), and history of falls at waves 1 and 3, and found that the distributions of TUG/UGS/falls at the two waves were statistically significantly different from each other, but significant differences were found for longitudinal TUG increment (TUG3 − TUG in wave 1 (TUG1)) and longitudinal UGS decrease (UGS1 − UGS3) between the subgroup of participants with only good Sustained Attention to Response Task (SART) performances at wave 1 and participants with at least one SART bad performance
We found that our new SART variable bad performances was a significant predictor of TUG decline in the employed binary logistic regression models, being associated with an increase per unit of around 30% in the odds of having TUG decline in the fully adjusted models
Summary
Computer-based neurocognitive tests are commonly utilised in research [1], and increasingly, in clinical practice, for both the detection and rehabilitation of cognitive disorders in adults [2]. The raw outputs from computer-based tests pose methodological and interpretation challenges, owing to the lack of optimal assays for the precise characterisation of latent neurocognitive processes, and shortcomings of many current methods to allow direct visualisation of multi-modal data that could help clinicians generate more meaningful hypotheses and predictions [3,4]. These challenges are only magnified in the case of computer-based repeated-measures neurocognitive data stemming from large-scale studies. Information averaging can result in loss of power and once a simplified predictor has been created and used in analyses, results should be interpreted considering the dimension of the derived variable, not at the level of the individual original variable [6]
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
