Belinda Brown

Background
In cognitively unimpaired (CU) individuals, the PACC is widely used as a cognitive outcome measure and endpoint in observational studies and clinical trials. However, it has drawn criticism for being heavily weighted towards memory. Increasing evidence indicates a decline spanning multiple cognitive domains in CU individuals. Therefore, using principal component analysis (PCA), we derived data-driven domain-specific cognitive composites. And subsequently, compared them against their summed z-score counterparts in predicting progression to mild cognitive impairment (MCI).

Method
Baseline cognitive, demographic, and genotype data of 2,853 CU older adults (aged 41.6 to 98.3) was obtained from the Alzheimer’s Dementia Onset and Progression in International Cohorts (ADOPIC) Consortium. Using varimax-rotated PCA, tests significantly loading (≥ 0.5) onto each principal component were extracted to derive domain-specific cognitive composites. The resulting domain scores were normalised to a mean of 0 and SD of 1, with a higher score indicating better cognition. Cox regression was used to assess the association between progression to MCI and baseline demographics, cognition, and APOE ε4 carriage. Akaike information criteria (AIC) was used to compare the fitness of PCA-derived composites against the zPACC and z-score domain-specific composites.

Result
Baseline cohort characteristics are described in Table 1. PCA explained 68% of the variance and resulted in four independent cognitive composites (Figure 1): memory; executive function; attention and processing speed; and global cognition. At 15 years from baseline, 309 participants progressed to MCI, while 2,544 remained CU. Cox regression showed that the four cognitive composites, age and APOE genotype significantly predicted progression to MCI (Concordance = 77%, p < 0.001, AIC = 4105, Table 1). Additionally, the PCA-derived composites performed comparably, if not better than the summed z-score counterparts, the PACC (Concordance = 74%, p < 0.001, AIC = 4163) and domain-specific composites (Concordance = 75%, p < 0.001, AIC = 4142). Baseline older age, APOE ε4 carriage in a dose-dependent manner (Figure 2) and poorer cognition for each PCA-composite were independently associated with progression to MCI.

Conclusion
Together with APOE ε4 carriage, our PCA-derived domain-specific composites performed better than their summed z-score counterparts at predicting progression to MCI 15 years before symptom onset.