Biomarkers

Authors

Julie K Wisch, Nicole S McKay, Matthew D Zammit, Bradley T Christian, Stephanie A Schultz, Peter R Millar, Nicolas R Barthélemy, Natalie S Ryan, Alan E Renton, Lisa Vermunt, Nelly Joseph-Mathurin, Zahra Shirzadi, Jeremy F Strain, Patricio Chrem, Alisha Daniels, Jasmeer P Chhatwal, Carlos Cruchaga, Laura Ibanez, Mathias Jucker, Gregory S Day, Jae-Hong Lee, Johannes Levin, Jorge J Llibre-Guerra, David Aguillon, Jee Hoon Roh, Charlene Supnet-Bell, Chengjie Xiong, Suzanne E Schindler, Guoqiao Wang, Yan Li, Robert Koeppe, Clifford R Jack, John C Morris, Eric McDade, Randall J Bateman, Tammie L S Benzinger, Beau Ances, Tobey J Betthauser, Brian A Gordon, Dominantly Inherited Alzheimer Network

Abstract

Alzheimers Dement. 2025 Dec;21 Suppl 2:e103008. doi: 10.1002/alz70856_103008.

ABSTRACT

BACKGROUND: Alzheimer Disease (AD) pathology evolves over decades, and understanding this progression is critical to the understanding of the disease and timing therapeutic interventions. Since individuals with Autosomal Dominant AD (ADAD) develop symptoms around the same age as their parent, it is possible to predict symptom onset and stage individuals by their estimated years to symptom onset (EYO). This approach does not generalize to other forms of AD, thus there is a pressing need for the timecourse of ADAD to be defined in broadly relevant terms. The objective of this project is to validate the Sampled Iterative Local Approximation (SILA) algorithm in a cohort with a known disease timecourse. SILA generates an estimate of time from amyloid positivity (A_time) based on longitudinal PET data.

METHOD: We evaluated A_time in a longitudinal ADAD sample (N = 316) with PET PiB data in three ways. First, we compared predicted age at amyloid positive (A+) to observed age at A+ for individuals who became A+ during enrollment. Next, using linear regression, we compared estimated age at A+ to estimated age at symptom onset (EYO=0). Finally, we used generalized additive models to compare the amount of variance in concurrent cognitive performance explained both A_time and EYO.

RESULT: We observed a mean average error of 1.15 years between actual age at A+ (N = 26) and the SILA-predicted A_time. Across all participants, SILA-estimated age at A+ explained 39% of the variance in estimated age at symptom onset (β = 0.918, p < 0.0001). Finally, we observed a nonlinear association between cognition and both A_time and EYO. A_time explained 19% of the variance in the general cognitive composite while EYO explained 43% of the variance.

CONCLUSION: SILA produces a valid estimate of time-from-amyloid positivity in ADAD. This work allows for disease stage in ADAD to be compared to staging for broad forms of AD, which was not previously possible using EYO. However, this work also illustrates that there is a high degree of heterogeneity in preclinical disease duration that is not explained by amyloid alone.

PMID:41451762 | DOI:10.1002/alz70856_103008