A Data-Driven Cognitive Composite Sensitive to Amyloid-β for Preclinical Alzheimer's Disease.

Background Integrating scores from multiple cognitive tests into a single cognitive composite has been shown to improve sensitivity to detect AD-related cognitive impairment. However, existing composites have little sensitivity to amyloid-β status (Aβ +/-) in preclinical AD. Objective Evaluate whether a data-driven approach for deriving cognitive composites can improve the sensitivity to detect Aβ status among cognitively unimpaired (CU) individuals compared to existing cognitive composites. Methods Based on the data from the Anti-Amyloid Treatment in the Asymptomatic Alzheimer's Disease (A4) study, a novel composite, the Data-driven Preclinical Alzheimer's Cognitive Composite (D-PACC), was developed based on test scores and response durations selected using a machine learning algorithm from the Cogstate Brief Battery (CBB). The D-PACC was then compared with conventional composites in the follow-up A4 visits and in individuals from the Alzheimer's Disease Neuroimaging Initiative (ADNI). Result The D-PACC showed a comparable or significantly higher ability to discriminate Aβ status [median Cohen's d = 0.172] than existing composites at the A4 baseline visit, with similar results at the second visit. The D-PACC demonstrated the most consistent sensitivity to Aβ status in both A4 and ADNI datasets. Conclusions The D-PACC showed similar or improved sensitivity when screening for Aβ+ in CU populations compared to existing composites but with higher consistency across studies.