Combined Digital Clock Drawing Test and Delayed Word Recall and their relationship with PET Amyloid Biomarker Status in Cognitively Unimpaired Older Adults

Abstract

AbstractBackgroundDCTclock™ is an AI‐enabled digital cognitive assessment (DCA) that strongly discriminates between cognitively impaired and unimpaired individuals (receiver operating area under‐the‐curve AUC = 0.89). Among cognitively unimpaired individuals, DCTclock predicts greater PET Aβ burden and shows stronger discrimination (Cohen’s d = 0.76) between Aβ± individuals than the Preclinical Alzheimer’s Cognitive Composite (d = 0.30) (Rentz et al., 2021). The Digital Clock and Recall (DCR™) was recently created by adding 3‐word delayed recall to DCTclock to assess verbal memory impairment – an early hallmark of Alzheimer’s disease. We assessed whether recreating the DCR with the 3‐word delayed recall from the Mini‐Mental State Examination (MMSE) and DCTclock would improve classification of Aβ PET biomarker status among cognitively unimpaired older adults.MethodDCTclock, MMSE (including delayed recall), and Aβ PET imaging data were collected from 159 cognitively unimpaired older adults in the Harvard Aging Brain Study (age = 78.6±5.9; 98 females; MMSE = 28.9±1.4). Logistic regression classifiers were trained to classify Aβ biomarker status and were assessed with standard metrics and AUC as compared to age‐ and MMSE‐only models.ResultThe best‐performing model classified biomarker status among cognitively unimpaired individuals with an AUC of 0.76 (sensitivity = 0.65, specificity = 0.73, accuracy = 0.71)‐outperforming age‐only and total MMSE‐only models (AUCs = 0.61). Recursive feature elimination identified spatial reasoning, average speed, maximum speed on Copy Clock, oscillatory motion on Command Clock, and delayed recall score as key features. DCTclock or recreated DCR summary scores alone classified biomarker status with AUCs of 0.72 and 0.74, respectively.ConclusionThe DCR, which combines DCTclock with delayed verbal recall, may offer a means to classify Aβ biomarker status among cognitively unimpaired individuals in as little as 3 minutes. Classification models trained on a subset of DCR features performed better than all other models, including traditional neuropsychological testing. These results reinforce the potential for DCR solutions to streamline cognitive testing in the clinic and accelerate diagnosis for individuals with early‐stage Alzheimer’s and other dementias. Future work will examine the best possible combination of additional tests to further increase classification accuracy.