HAI Book 2025 - Flipbook - Page 681
Razlighi, Qolamreza
166
The Tau Progression Index (TPI): an individualized, clinically
applicable, multimodally-derived score to predict AD progression
Gloria Chiang1, Tracy Butler1, Bardiya Ghaderi Yazdi1, Sindy Ozoria1, Peter Chernek1, Jenseric
Calimag1, Sonja Blum1, Silky Pahlajani1, Nancy Foldi1, Mohammad Khalafi1, Devangere
Devanand2, Hengda He2, Hani Hojjati1, Saman Gholipour Picha1, Christian Habeck2, Yaakov
Stern2, Qolamreza Razlighi1
1
Weill Cornell Medicine, New York, NY, US
Columbia University Irving Medical Center, New York, NY, US
2
Background: Alzheimer9s Disease (AD) prognosis is extremely heterogeneous, even with a similar burden of global
tau and amyloid (A´) deposition in the brain, which makes it challenging to develop targeted therapeutic
interventions and to counsel families on disease prognosis.
Method: The TPI is based on 4 components: 1) Remote interaction (TPIri) between A´ and tau pathologies in
regions that are functionally and/or structurally connected, but spatially distinct; 2) Local interaction (TPIli)
between spatially co-localized A´ and tau pathology; and subject-specific 3) Functional connectivity (TPIfc) and 4)
Structural connectivity (TPIsc) between regions that shown to facilitate the spread of tau in the brain.
Results: From an ongoing study of 112 participants with early accumulation of Ab and Tau, longitudinal data were
available on 27 participants (2-3 years of follow-up). Using these data and all 4 TPI components as independent
variables, we built a LASSO model to predict future tau accumulation, controlling for covariates (age, gender,
APOE, infarcts, WMH, head-size). The obtained coefficients were used to compute our TPI, and its predictability
was assessed by its association with actual longitudinal tau accumulation (within-sample validation).
Furthermore, we compared our model, using subject-specific connectomes, with a conventional model using
group-averaged connectomes. As seen in Figure. 1, while both TPIs (obtained by subject-specific; r=0.8, p
