Alzheimer's Disease Pathology Imputation and Risk Prediction Using Clinical Indices Público

Abstract

Alzheimer’s disease (AD) is a chronic progressive disorder that develops over years before manifesting impaired cognition, and early detection and intervention before the onset of noticeable AD symptoms might slow down the progression of cognitive decline. Since existing AD biomarkers are problematics and not widely available, this study aimed to develop models for imputing AD brain pathology and predict the risk of AD using common clinical indices. Data used in this study included clinical indices and postmortem pathology data contributed by 2000+ participants from two cohort studies who agreed on annual clinical visit and brain donation after death. In stage 1 of our study, we validated imputation models and chose the best-performing machine learning method: generalized linear regression model with elastic net regulation. In stage 2, we applied the imputation models to estimate baseline AD pathology using 57 clinical variables as predictors. In stage 3, we fitted Cox proportional hazard models and used the imputed pathology along with three demographic indices to predict the risk of cognitive impairment and AD dementia over years. Based on our data analysis results, imputed pathology was able to distinguish AD pathology-absent participants from AD pathology-present participants, and the clinical variables measured at baseline were effective predictors of baseline AD pathology. Moreover, imputed pathology along with three demographic indices were enough to make effective prediction on the risk of developing mild cognitive impairment or Alzheimer’s disease dementia. If the leveraged clinical indices—common, affordable, and convenient to be measured—can be used as new biomarkers that substitute the existing but problematic ones, many more elderly people would be able to benefit from early detection, intervention and prognosis of their potential risk of developing AD dementia or cognitive impairment.

Table of Contents

1. Introduction ................................................................................................................................................................................... 1

1.1. Alzheimer’s disease ...................................................................................................................................................................... 1

1.2. Motivations ................................................................................................................................................................................. 2

1.3. Study design ................................................................................................................................................................................ 4

1.4. Research goals and objectives........................................................................................................................................................ 5

2. Methods......................................................................................................................................................................................... 5

2.1. Clinical variables ......................................................................................................................................................................... 6

Table 2. Sample sizes per event type for the ROS/MAP cohorts used for fitting prediction models of the burden of AD pathology ............... 7

Figure 1. Correlation heatmap of clinical variables as predictors of the burden of AD Pathology .............................................................. 7

2.2. Assessment of AD brain pathology ............................................................................................................................................... 8

2.3. Data preparation ......................................................................................................................................................................... 9

2.4. Analytic approach ...................................................................................................................................................................... 10

2.4.1. Stage 1: Train and validate imputation models for AD pathology................................................................................................. 10

Figure 2. Stage 1 flowchart: model validation by training on near-death MAP cohort and testing on near-death ROS cohort .................... 10

2.4.2. Stage 2: Estimate AD pathology at study baseline ...................................................................................................................... 11

Figure 3. Stage 2 flowchart: AD pathology imputation by using GLM-EN models to train on near-death MAP & ROS cohort and to predict

pathology level in MAP & ROS at baseline. ......................................................................................................................................... 12

2.4.3. Stage 3: Predict AD risk by Cox proportional hazard model ..........................................................................................................12

Figure 4. Stage 3 flowchart: risk prediction models by training on baseline clinical variables and imputed pathology and predicting

occurrence of MCI and/or ADD events in year 3 and 5 .......................................................................................................................... 13

3. Results (include explanation of each figure/table) ............................................................................................................................ 14

3.1. Stage 1: Train and validate imputation models for AD pathology .................................................................................................... 14

Table 3. Comparison of machine learning method performance:

results of prediction R2 and AUC obtained from cross validation models in stage 1 ................................................................................ 15

Figure 5. Pathology estimation results in ROS testing samples at death and their discrimination with respect to profiled NIA-Reagan at death

for β-Amyloid, Tangles, Global AD pathology, and NIA-Reagan ............................................................................................................. 15

3.2. Stage 2: Estimate AD pathology at study baseline ........................................................................................................................... 16

Figure 6. Important covariates selected by GLM-EN in the prediction models for the burden of AD pathology ............................................17

Table 4. Training and testing results obtained from GLM-EN imputation models in stage 2 ......................................................................18

Figure 7. Imputation performance at baseline with respect to NIA-Reagan at death (stage 2).

Boxplots of predicted AD pathology at baseline with respect to profiled NIA-Reagan at death .................................................................. 19

3.3. Stage 3: Predict AD risk over years................................................................................... ...............................................................20

Figure 8. Risk prediction results of MCI and ADD events in year 3 and 5 by Cox proportional hazard regression models, using age, sex,

education and imputed AD pathology as predictors .............................................................................................................................. 21

4. Conclusions .................................................................................................................................................................................... 22

Appendix ............................................................................................................................................................................................24

Supplemental Table 1. Baseline characteristics of clinical variables by category .......................................................................................24

References .........................................................................................................................................................................................25 

About this Honors Thesis

Rights statement

• Permission granted by the author to include this thesis or dissertation in this repository. All rights reserved by the author. Please contact the author for information regarding the reproduction and use of this thesis or dissertation.

School	Emory College
Department	Quantitative Science
Degree	B.S.
Submission	Honors Thesis
Language	English
Research Field	Biology, Neuroscience Biology, Genetics Biology, Biostatistics
Palabra Clave	MCI (mild cognitive impairment) Cognitive aging Brain pathology Alzheimer's disease Cohort studies
Committee Chair / Thesis Advisor	Jingjing Yang, Emory University
Committee Members	David J. Cutler, Emory University Zhiyun Gong, Emory University

Última modificación

Primary PDF

Thumbnail	Title	Date Uploaded	Actions
	Alzheimer's Disease Pathology Imputation and Risk Prediction Using Clinical Indices ()	2022-04-12 23:35:44 -0400	Press to Select an action Download

Supplemental Files

Thumbnail	Title	Date Uploaded	Actions