Pick a PACC: Comparing domain- specific- and general- cognitive composites in Alzheimer disease research

Balancing the evolution of longitudinal cognitive assessment with practical considerations like maintaining legacy measures presents a vexing problem for many Alzheimer's Disease Research Centers (ADRCs) in the US. Changes to the Uniform Data Set (UDS) have left many ADRCs with heterogeneous cognitive data that can be challenging to harmonize. Using longitudinal data from the Knight ADRC, we used equipercentile equating to combine cognitive data collected across UDS versions 2 and 3. The combined data was then used to assess whether equated variables could create stable and informative domain-specific and general cognitive composite measures. Exploratory factor analysis (EFA) revealed four underlying clusters of tests that corresponded to episodic memory, semantic memory, working memory, and attention/processing speed. These factors appeared stable across test forms, with equivalent factors observed when EFA was conducted on UDS2 and UDS3 independently, and replicated when UDS2 and UDS3 were harmonized using equated variables. To understand the utility of these domain-specific factors, we also computed a commonly used preclinical Alzheimer cognitive composite (ADCS-PACC), a novel PACC (Knight-PACC), and a global composite consisting of 13 cognitive tasks. Prior work has proposed that these PACCs are sensitive to preclinical changes in AD, while the global composite would reflect general cognitive ability. We compared the relationships between each composite and neuroimaging metrics of amyloid, tau, brain structure, and clinical symptoms, cross-sectionally and longitudinally. We also simulated clinical trials using each composite and compared the sample size estimates needed to power a study expecting to observe a moderate effect size. Overall, global and PACC-based metrics of cognition slightly outperformed domain-specific composites, showing better sensitivity to AD biomarkers, although the working memory composite consistently performed poorly. We also found that data-driven, domain-specific composites can complement already-established general cognitive composites, with the exception of the present working memory composite. Power analyses revealed that the global, Knight-PACC, and attention and processing speed composites would require the smallest sample sizes to detect cognitive change in a clinical trial, while the ADCS-PACC required two to three times as many participants. Importantly, this work demonstrated the stability of cognitive domains across test versions. Finally, in line with prior work, we also highlight that attention and processing speed may be sensitive metrics of cognitive change in the earliest stages of AD.