Simulating the outcome of amyloid treatments in Alzheimer’s disease from imaging and clinical data

Clément Abi Nader, Nicholas Ayache, Giovanni B. Frisoni, Philippe Robert, Marco Lorenzi, for the Alzheimer’s Disease Neuroimaging InitiativebioRxiv 2020 · DOI

TL;DR

This paper introduces a novel computational model that simulates the progression of Alzheimer's disease and the potential impact of amyloid-targeting treatments. By integrating imaging and clinical data, the model suggests that early intervention, even in the pre-symptomatic stage, is crucial for achieving statistically significant improvements in clinical outcomes.

What Problem Does This Paper Address?

Alzheimer's disease (AD) is a devastating neurodegenerative disorder affecting millions worldwide, with a significant and growing healthcare burden. Despite decades of research, effective treatments remain elusive. A major challenge is the disease's long, silent, pre-symptomatic phase, making it difficult to intervene effectively. While the "amyloid cascade hypothesis" suggests that amyloid-beta (Aβ) plaque deposition drives AD progression, clinical trials targeting Aβ have yielded mixed results. This lack of success is partly attributed to the timing of intervention and the need for better tools to predict treatment efficacy. This paper aims to address this by developing a quantitative framework to simulate intervention strategies and understand the critical role of intervention timing and drug dosage on disease progression.

Methods

The authors propose a novel computational model that leverages a Bayesian framework and stochastic variational inference to model the spatio-temporal dynamics of AD progression. The core of the method involves:

  • Latent Space Projection: High-dimensional multi-modal data (imaging and clinical scores) are projected into a lower-dimensional latent space, represented by four key z-scores: amyloid burden (z_amy), glucose metabolism (z_met), atrophy (z_atr), and clinical status (z_cli).
  • Latent Dynamical System: A system of Ordinary Differential Equations (ODEs) is used to model the relationships and transitions between these latent z-scores over time. This system captures the causal cascade of pathological changes.
  • Simulation of Interventions: The model allows for simulating hypothetical interventions by modulating the parameters of the ODEs. Specifically, they simulate amyloid-lowering interventions by adjusting the rate of change for the amyloid z-score.
  • Disease Severity Estimation: A time-shift parameter is used to quantify individual disease severity relative to the modeled progression trajectory.

This approach differs from previous Disease Progression Models (DPMs) by explicitly modeling causal relationships between biomarkers and enabling the simulation of intervention scenarios, rather than just describing progression.

Datasets

The study utilizes data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) database. This is a well-established, large-scale, longitudinal study that collects a wide range of data, including MRI, PET scans (for amyloid and glucose metabolism), and various neuro-psychological tests. The authors specifically used data from amyloid-positive individuals to ensure they were studying the "Alzheimer's disease continuum."

Key Insights and Ideas

  • Amyloid as a Primary Driver: The model confirms the amyloid cascade hypothesis, showing that increased amyloid burden (z_amy) is the initial trigger that leads to changes in glucose metabolism (z_met), brain atrophy (z_atr), and ultimately clinical decline (z_cli).
  • The Criticality of Intervention Timing: The most striking insight is the profound impact of when an intervention is administered. Earlier interventions, especially in the pre-symptomatic or very early symptomatic stages, yield significantly better outcomes than later interventions.
  • Pre-clinical Intervention Justification: The results provide a theoretical basis for testing amyloid-modifying drugs in individuals who are cognitively normal but have evidence of amyloid pathology.
  • Dose-Response Relationship: The model demonstrates that higher levels of amyloid lowering (e.g., 100% vs. 50%) lead to more substantial improvements in clinical outcomes, given the same intervention timing.

Key Results

  • Early Intervention is Key: The simulations suggest that to achieve statistically powered improvements (e.g., 80% power) in clinical endpoints, interventions need to be administered significantly earlier than currently practiced. For a sample size of 1000 individuals per arm, this could be up to 7 years before clinical diagnosis.
  • Impact of Sample Size and Intervention Timing: Figure 5 illustrates how sample size and intervention timing interact to determine statistical power. With smaller sample sizes (e.g., 100 subjects per arm), interventions need to occur even earlier (up to 11.5 years before conversion for ADAS11 with 100% amyloid lowering).
  • Model Compatibility: The model's simulated outcomes are consistent with findings from past clinical trials, lending credibility to its predictive capabilities. For instance, simulations for a 2-year intervention with 100% amyloid lowering align with the confidence intervals of MMSE improvements observed in the Verubecestat trial.
  • Clinical Endpoint Sensitivity: Different clinical scores show varying sensitivity to intervention. RAVLT forgetting, for example, appears to offer higher statistical power for detecting treatment effects compared to other scores.

Limitations and Future Work

  • Cohort Specificity: The study's reliance on amyloid-positive individuals from the ADNI cohort means the results may not be fully representative of the general Alzheimer's disease progression, particularly for amyloid-negative individuals.
  • Tau and Genetics: The model currently does not incorporate tau pathology or genetic factors like APOE4, which are known to influence AD progression. The authors suggest that future work should include these biomarkers to provide a more comprehensive understanding.
  • Other Aggravating Factors: The model could be extended to account for non-genetic risk factors (e.g., diabetes, obesity, smoking) to simulate personalized intervention strategies.
  • Clinical Trial Design Optimization: The authors propose that their model could be used to optimize clinical trial design by simulating scenarios based on different sample sizes, trial durations, and available funding, thereby estimating expected cognitive outcomes.
  • Data Modalities: While the model can handle missing data, it requires at least one visit with complete measures for initial estimation. Future work could explore more robust methods for handling missing data.

Summary updated

Abstract

In this study we investigate a novel quantitative instrument for the development of intervention strategies for disease modifying drugs in Alzheimer’s disease. Our framework is based on the modeling of the spatio-temporal dynamics governing the joint evolution of imaging and clinical biomarkers along the history of the disease, and allows the simulation of the effect of intervention time and drug dosage on the biomarkers’ progression. When applied to multi-modal imaging and clinical data from the Alzheimer’s Disease Neuroimaging Initiative our method enables to generate hypothetical scenarios of amyloid lowering interventions. The results quantify the crucial role of intervention time, and provide a theoretical justification for testing amyloid modifying drugs in the pre-clinical stage. Our experimental simulations are compatible with the outcomes observed in past clinical trials, and suggest that anti-amyloid treatments should be administered at least 7 years earlier than what is currently being done in order to obtain statistically powered improvement of clinical endpoints.
BibTeX
@misc{nader2020simulating,
  author = {Clément Abi Nader and Nicholas Ayache and Giovanni B. Frisoni and Philippe Robert and Marco Lorenzi and for the Alzheimer’s Disease Neuroimaging Initiative},
  title = {{Simulating the outcome of amyloid treatments in Alzheimer’s disease from imaging and clinical data}},
  year = {2020},
  doi = {10.1101/2020.09.02.279521},
  howpublished = {bioRxiv},
}
Figure