Data-Driven Approaches to Modeling Heterogeneity and Variability Across Asymptomatic Brain and Cognitive Aging, Mild Cognitive Impairment, and Alzheimer’s disease

• Author / Creator

  Drouin, Shannon

• Objective
  We apply data-driven approaches to identify predictors of heterogeneous trajectories across normal aging, Mild Cognitive Impairment (MCI), and Alzheimer’s disease (AD). In Study 1, we investigated predictors of left and right hippocampal (HC) volume trajectory classes. In Study 2, we identified the leading predictors of cognitive resilience, cognitive vulnerability, and brain/cognitive stability in varying contexts of morphometric brain changes. In Study 3, we examined the leading predictors of AD, MCI, and dementia from a set of risk factors/biomarkers including novel metabolomics markers.
  Methods
  Study 1 participants (n=351) were cognitively normal (CN) older adults from the Alzheimer’s Disease Neuroimaging Initiative (ADNI) with longitudinal imaging and baseline biomarker/risk factor data. We applied latent class growth analysis (LCGA) to identify separable HC trajectory classes. We then applied a machine learning (ML) algorithm to identify leading biomarker predictors (from 38) discriminating the highest and lowest trajectory classes.
  Study 2 participants (n=415) were CN older adults from ADNI with longitudinal imaging and baseline biomarker/risk factor data. We used LCGA for two foundational goals, identifying HC and cognitive trajectory classes. We applied ML algorithms to identify the leading predictors (from 42) of (a) cognitive resilience, (b) cognitive vulnerability, and (c) brain/cognitive stability.
  Study 3 participants included two samples of three age- and sex-matched cohorts from the Comprehensive Assessment of Neurodegeneration in Aging study: cognitively unimpaired (CU; n1=33; n2=32), MCI (n1=33; n2=33), and AD (n1=33; n2=21). For each sample, we used three ML algorithms and 111-112 risk factors/biomarkers, and metabolite predictors to identify the leading predictors of AD (CU-AD), MCI (CU-MCI), and dementia (MCI-AD).
  Results
  Study 1: We detected three trajectory classes for the left HC and three classes for the right HC (highest, middle, lowest). For the left HC, seven predictors from four modalities discriminated the lowest and highest HC trajectory classes: plasma Aβ1-40, plasma tau, plasma Aβ1-42, sex, education, depression, and body mass index. For the right HC, three predictors from two modalities discriminated the lowest and highest HC trajectory classes: sex, education, and plasma Aβ1-42.
  Study 2: We first detected two HC trajectory classes and two cognitive trajectory classes (low and high). Based on differential combined class membership, we identified three target subgroups: cognitively resilient (n=72), cognitively vulnerable (n=144), and brain and cognitively stable (n=92). Cognitive resilience was predicted by higher CSF Aβ1-42, higher education, lower plasma Aβ1-42, lower CSF p-tau, lower plasma Aβ1-40, and lower age. Cognitive vulnerability was predicted by lower education, higher plasma Aβ1-40, higher BMI, higher age, lower glucose, higher plasma Aβ1-42. Brain/cognitive stability was predicted by higher CSF Aβ1-42, lower polygenic risk score, female sex, higher plasma Aβ1-42, higher pulse pressure, and lower age.
  Study 3: We report leading predictors at a 40% model explanation criterion. AD was predicted by six biomarkers from three domains (sensory, imaging, and metabolomics) in Sample 1 and nine biomarkers from five domains (imaging, demographic, and clinical health, sensory, and metabolomics) in Sample 2. MCI was predicted by 12 biomarkers from three domains (metabolomics, clinical health, and imaging) in Sample 1 and 13 biomarkers from five domains (metabolomics, clinical health, vascular/metabolic, imaging, and demographic) in Sample 2. Dementia was predicted by nine predictors from four domains (sensory, imaging, metabolomics, and vascular/metabolic) in Sample 1 and nine biomarkers from seven domains (imaging, demographic, clinical health, metabolomics, gait/function, lifestyle, and vascular/metabolic) in Sample 2.
  Discussion
  The overall aim of this dissertation research was to apply data-driven approaches to the prediction of heterogeneous trajectories and outcomes in brain/cognitive aging and dementia. Study 1 demonstrated that HC trajectory classes represent secondary phenotypes of brain aging that are predicted by a wide range of AD-related risk factors/biomarkers. Study 2 demonstrated that (a) cognitive trajectories can supplement HC trajectories and represent alternative pathways of brain and cognitive aging and (b) these pathways can be predicted by a similar roster of AD-related risk factors/biomarkers. Study 3 identified the leading predictors originating from established risk domains as well as novel metabolomics predictors that are associated with MCI, AD, and/or dementia. Overall, the dissertation research highlights the relative importance of risk factors/biomarkers and candidate metabolites in the prediction of both desirable (high HC trajectory classes, resilience, stability) and undesirable (low HC trajectory classes, cognitive vulnerability, MCI, AD, dementia) aging trajectories and outcomes.

• Subjects / Keywords

  • Alzheimer's disease
  • multi-omics
  • biomarkers
  • risk factors
  • predictors
  • machine learning
  • data-driven

• Graduation date

  Spring 2024

• Type of Item

  Thesis

• Degree

  Doctor of Philosophy

• DOI

  https://doi.org/10.7939/r3-6eqc-tm19

• License

  This thesis is made available by the University of Alberta Libraries with permission of the copyright owner solely for non-commercial purposes. This thesis, or any portion thereof, may not otherwise be copied or reproduced without the written consent of the copyright owner, except to the extent permitted by Canadian copyright law.