Predicting sporadic Alzheimer's disease progression via inherited Alzheimer's disease-informed machine-learning

Abstract

Introduction: Developing cross-validated multi-biomarker models for the prediction of the rate of cognitive decline in Alzheimer's disease (AD) is a critical yet unmet clinical challenge.

Methods: We applied support vector regression to AD biomarkers derived from cerebrospinal fluid, structural magnetic resonance imaging (MRI), amyloid-PET and fluorodeoxyglucose positron-emission tomography (FDG-PET) to predict rates of cognitive decline. Prediction models were trained in autosomal-dominant Alzheimer's disease (ADAD, n = 121) and subsequently cross-validated in sporadic prodromal AD (n = 216). The sample size needed to detect treatment effects when using model-based risk enrichment was estimated.

Results: A model combining all biomarker modalities and established in ADAD predicted the 4-year rate of decline in global cognition (R² = 24%) and memory (R² = 25%) in sporadic AD. Model-based risk-enrichment reduced the sample size required for detecting simulated intervention effects by 50%-75%.

Discussion: Our independently validated machine-learning model predicted cognitive decline in sporadic prodromal AD and may substantially reduce sample size needed in clinical trials in AD.

Keywords: Alzheimer's disease; MRI; PET; autosomal-dominant Alzheimer's disease; biomarkers; machine learning; progression prediction; risk enrichment.

FIGURE 1

Flow-chart of the support vector regression (SVR) analysis pipeline. (A) Selected data from the DIAN-MC and ADNI-MCI sample are standardized and variance normalized to the respective healthy reference groups to ensure comparability of biomarker scaling across samples. (B) The SVR model is trained based on selected modalities in DIAN-MC in a nested cross-validation framework. (C) The trained SVR-models are blindly applied to the scaled ADNI-MCI biomarker data yielding a SVR score per subject. The SVR score is then evaluated as a predictor of baseline cognition and longitudinal cognitive decline in ADNI

FIGURE 2

SVR-based prediction of EYO in autosomal-dominant Alzheimer’s disease and feature selection probabilities. (A) Scatterplot showing the association between observed EYO scores and AFGC-predicted estimated years to symptom onset (EYO) scores in DIAN-MC. (B) Selection probabilities of the AFGC model indicate the percentage of final CV1 models (see step in Figure 1B) that included the respective region of interest/biomarker. Features were selected in the DIAN-MC cohort during each CV1 cycle based on the correlation with the outcome measure (ie, EYO)

FIGURE 3

SVR-based prediction of cognitive changes in ADNI MCI-Aβ+. Scatterplots showing the association between AFGC-derived SVR scores and longitudinal cognitive change in ADNI-MCI-Aβ+ for ADNI-MEM (A-D) and ADAS13 (E–H). Standardized β-values, partial R², and P-values are based on linear regression models adjusted for age, sex, and education