Dual Model Deep Learning for Alzheimer Prognostication

2025-12-22

AI Summary

Current Alzheimer’s disease prediction models rely on longitudinal data and lack uncertainty quantification, limiting their utility for initial clinical decision-making. To address this, we propose PROGRESS—a dual-model framework integrating a probabilistic trajectory network with a deep survival model that predicts disease progression trajectories and time-to-dementia conversion from mild cognitive impairment using only a single baseline cerebrospinal fluid (CSF) measurement, without requiring longitudinal follow-up or clinical history. We introduce the first calibration-aware uncertainty quantification mechanism, enabling robust performance across multi-center cohorts and overcoming heterogeneity arising from decades-spanning assay technologies. In leave-one-center-out validation across 3,000+ samples from 43 centers, PROGRESS achieves consistent C-indices >0.78—significantly outperforming Cox regression, random survival forests, and gradient-boosted survival models. Risk stratification reveals a sevenfold difference in conversion rates, demonstrating both high discriminative power and clinical interpretability.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Relational Probabilistic Models
• Cognitive Modeling & Cognitive Systems: Other Foundations of Cognitive Modeling & Systems

Application Category

• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Disease modifying therapies for Alzheimer's disease demand precise timing decisions, yet current predictive models require longitudinal observations and provide no uncertainty quantification, rendering them impractical at the critical first visit when treatment decisions must be made. We developed PROGRESS (PRognostic Generalization from REsting Static Signatures), a dual-model deep learning framework that transforms a single baseline cerebrospinal fluid biomarker assessment into actionable prognostic estimates without requiring prior clinical history. The framework addresses two complementary clinical questions: a probabilistic trajectory network predicts individualized cognitive decline with calibrated uncertainty bounds achieving near-nominal coverage, enabling honest prognostic communication; and a deep survival model estimates time to conversion from mild cognitive impairment to dementia. Using data from over 3,000 participants across 43 Alzheimer's Disease Research Centers in the National Alzheimer's Coordinating Center database, PROGRESS substantially outperforms Cox proportional hazards, Random Survival Forests, and gradient boosting methods for survival prediction. Risk stratification identifies patient groups with seven-fold differences in conversion rates, enabling clinically meaningful treatment prioritization. Leave-one-center-out validation demonstrates robust generalizability, with survival discrimination remaining strong across held-out sites despite heterogeneous measurement conditions spanning four decades of assay technologies. By combining superior survival prediction with trustworthy trajectory uncertainty quantification, PROGRESS bridges the gap between biomarker measurement and personalized clinical decision-making.

Problem

• Predicts Alzheimer's cognitive decline from single baseline biomarker
• Estimates time to MCI-to-dementia conversion with survival model
• Provides uncertainty quantification for personalized treatment decisions

Innovation

• Dual-model deep learning uses single baseline biomarker data.
• Probabilistic trajectory network predicts cognitive decline with uncertainty.
• Deep survival model estimates time to dementia conversion.