Abstract
BACKGROUND: Early identification of cognitive impairment and brain pathology associated with Alzheimer's disease (AD) is essential to maximize benefits from lifestyle interventions and emerging pharmacologic disease-modifying treatments (DMT). Digital cognitive assessments (DCAs) can quickly capture an array of metrics that can be used to train machine-learning (ML) models to concurrently evaluate different outcomes. DCAs have the potential to optimize clinical workflows and enable efficient assessment of cognitive function and the likelihood of a given underlying pathology.
METHODS: We assessed the ability of a multimodal ML-enabled DCA, the Digital Clock and Recall (DCR), to concurrently estimate brain amyloid-beta (Aβ) status and detect cognitive impairment, as compared with traditional cognitive assessments, including the MMSE, RAVLT, a DCA, Cognivue®, and blood-based biomarkers in 930 participants from the Bio-Hermes-001 clinical study.
RESULTS: Aβ42/40, p-tau181, APS, and p-tau217 poorly classified cognitive impairment (AUCs: 0.61; 0.63; 0.63; 0.70, respectively), but accurately classified Aβ status (AUCs: 0.81; 0.78; 0.85, 0.89, respectively). MMSE, RAVLT, and Cognivue poorly classified Aβ status (AUCs: 0.70, 0.73, 0.70, respectively). However, separate multimodal, DCR-based ML classification models, run in parallel, accurately classified both cognitive impairment (AUC = 0.83) and Aβ-PET status (AUC = 0.81).
CONCLUSIONS: DCAs that leverage digital technologies to generate advanced metrics, such as the DCR, enable accurate and efficient detection of cognitive impairment associated with AD pathology. They have the potential to empower health systems and primary care providers to help their patients make timely treatment decisions.