Forecasting Medium-Horizon Alzheimer's Disease Progression: Residual Gap-Aware Transformers for 24-Month CDR-SB Change from ADNI Clinical and Biomarker Histories

2026-05-19 · Source: stat.ML updates on arXiv.org · Field: Health & Wellbeing — Health & Medical Research, Medical Devices & Health Technology, Artificial Intelligence & Machine Learning · Depth: Expert, extended

Summary

A new model, the residual gap-aware transformer, has been developed to forecast medium-horizon Alzheimer's disease progression, specifically the 24-month Clinical Dementia Rating Sum of Boxes (CDR-SB) change. This model leverages harmonized longitudinal data from the Alzheimer's Disease Neuroimaging Initiative (ADNI), comprising 2,600 labeled anchors from 858 participants and 7,276 longitudinal rows. It combines a mixed-effects statistical reference with a transformer-based residual learning approach, incorporating participant-level random intercepts, observation-level triplet tokenization for irregular histories, and a learned nonnegative time-gap penalty in self-attention. The model was compared against a linear mixed-effects baseline, GRU-D, and STraTS across five participant-level random seeds. It achieved the best mean test performance across all reported metrics, reducing Mean Squared Error (MSE) by 13.1% and increasing prediction-observation correlation by 26.4% relative to the mixed-effects baseline, also outperforming GRU-D and STraTS.

Key takeaway

For AI Scientists and Machine Learning Engineers developing Alzheimer's disease progression models, consider adopting a hybrid statistical-neural approach. Your models should integrate a mixed-effects statistical reference to capture baseline disease burden and within-subject dependence, with a transformer-based residual learner that accounts for irregular longitudinal clinical and biomarker histories, including time-gap penalties in attention. This strategy can significantly improve prediction accuracy and correlation for medium-horizon outcomes like 24-month CDR-SB change, offering a more robust and clinically interpretable framework.

Key insights

Combining statistical anchoring with gap-aware residual learning improves medium-horizon Alzheimer's disease progression prediction.

Principles

• CDR-SB change is a meaningful medium-horizon progression target.
• Longitudinal biomarker history, even if irregular, is informative.
• Residual learning can enhance statistical reference models.

Method

The proposed method uses a mixed-effects model for baseline prediction, then a gap-aware transformer learns residuals from irregular longitudinal clinical and biomarker histories, employing observation-level triplet tokenization and a time-gap penalty in self-attention.

In practice

• Use 24-month CDR-SB change as a progression endpoint.
• Incorporate historical, irregular biomarker data.
• Employ participant-level train-test splits for robust evaluation.

Topics

• Alzheimer's Disease Progression
• Residual Gap-Aware Transformer
• CDR-SB Change Prediction
• ADNI Longitudinal Data
• Mixed-Effects Models

Best for: AI Scientist, Research Scientist, Machine Learning Engineer

Related on AIssential

• Transition-Based Digital Twin Modelling for Alzheimer's Disease under Sparse Longitudinal Data — Local transition modeling offers a data-efficient and robust strategy for personalized disease forecasting in neurodegenerative disorders.
• Treatment-Conditioned Diffusion for Forecasting Neurodegenerative Disease Progression — A treatment-conditioned diffusion framework accurately forecasts neurodegenerative disease progression by integrating imaging and pharmacological data, preserving anatomical detail.
• REVEAL: Multimodal Vision-Language Alignment of Retinal Morphometry and Clinical Risks for Incident AD and Dementia Prediction — Jointly modeling retinal images and clinical risk factors via VLM improves early AD and dementia prediction.
• Learning Directional Semantic Transitions for Longitudinal Chest X-ray Analysis — ProTrans models disease progression in CXRs as directional semantic transitions using vision-language pretraining and radiology reports.
• Scaling Recurrence-aware Foundation Models for Clinical Records via Next-Visit Prediction — RAVEN is a generative pretraining strategy for EHRs that predicts next-visit events while accounting for recurrence.
• AI-powered evaluation of dementia severity based on clinical data and visual scoring systems (MTA, ERICA, GCA) from MRI — AI-powered visual scoring and clinical data integration improves dementia severity evaluation, reducing expert dependency.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.