Brook Galna

Background
Posterior cortical atrophy (PCA) is a dementia subgroup commonly misdiagnosed due to unusual presentation and limited clinical awareness. Previously, the GaitDem study provided proof-of-concept for the use of accelerometery-based walking assessment in clinical and real-world settings in supporting differentiation Lewy body disease (LBD) and Alzheimer's disease (AD). Real-world walking assessment also provides insights into the impact of disease on everyday behaviours. Here, we aimed to assess the feasibility of accelerometery-based real-world walking assessment of PCA and describe differences between PCA and more prevalent neurodegenerative dementia syndromes, AD and LBD.

Methods
Fourteen participants with PCA (Age: 71 years (56-78); 57% female) wore an accelerometer (AX6, Axivity) affixed to their lower back for up to seven days. Using validated algorithms, real-world walking outcomes were derived including measures of walking quality (step velocity), volume (minutes spent walking, steps per day, bouts per day), pattern (mean bout duration) and variability (of bout durations). Data was compared to the GaitDem cohort, which included 36 people with AD (Age: 77 years (67-88); 58% female) and 46 with LBD (Age: 77 years (65-91); 17% female), following a similar protocol. Kruskal-Wallis Test assessed between-group differences with post-hoc Dunn tests. 26 controls (Age: 74(60-89), 58% female) were included for visual comparison (Figure 1).

Results
Twelve PCA participants completed seven days of real-world walking assessment; two completed 5-6 days. The PCA group walked faster, spent more minutes walking, took more steps and walking bouts per day (p <0.01 for all) than the LBD group; no significant differences were found for pattern and variability outcomes or between PCA and AD groups (p >0.05; Figure 1).

Conclusion
All PCA participants completed real-world walking assessment for the recommended period of >3 days, suggesting feasibility. Preliminary results suggest that the PCA group's real-world macro walking behaviours are more similar to AD than LBD. Despite a small sample, this novel data provides proof-of-concept. Only real-world macro walking outcomes are reported. Previously, clinic-based accelerometery outcomes relating to micro gait characteristics (e.g. spatiotemporal and signal-based features) were more sensitive to differences between LBD and AD; this will be further explored for PCA.

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.