Original title: Multimodal risk trajectories reveal heterogeneous paths to dementia
Zhiqi Lee, Haowen Li, Tao Liu, Shiyuan Zhang +5 · August 2026
Dementia is not just one condition; it involves several different brain disorders. Predicting who will get which specific type before symptoms appear is very hard. To fix this, scientists built a tool called NetMoint. It works by combining blood proteins, structural brain scans, and measurements of blood flow in the head, even when some data is missing. The system tracks how risk changes across 1, 5, 10, and 20 years for three major types: Alzheimer's disease, vascular dementia, and frontotemporal dementia. It proved remarkably accurate across all three types. The tool showed that brain structure matters most for short-term risk, while blood proteins matter most for long-term risk decades away. It also discovered small, very high-risk groups with unique molecular patterns, such as lower levels of a protein named TGFB1 in Alzheimer's risk and higher levels of NDRG1 in frontotemporal dementia risk.
You are planning a hike and want to know if it will rain. Checking the sky now tells you about the next hour, but checking long-term forecasts tells you about next week. Here, the researchers combine a snapshot of the brain (like the sky) and a blood sample (like the forecast) to predict dementia risk over different time horizons. They found that the most important clue changes depending on how far ahead you look, and that a small group of people follow a distinct high-risk path.
Dementia is often diagnosed too late for effective treatment. This work shows that combining blood tests and brain scans can identify people on a high-risk path for specific dementia types up to 20 years before symptoms appear. That means clinicians could one day target prevention or monitoring to the right people at the right time, instead of waiting for irreversible damage.
The model was developed using UK Biobank data, which may not represent all populations. The independent validation used a smaller set of features and achieved lower accuracy (AUC 0.741), indicating challenges in generalizing across cohorts. The abstract does not report calibration or clinical utility measures, nor does it establish causality between the molecular signatures and dementia progression.
Ask anything about this paper. Answers come from its abstract, and say so when the abstract doesn't cover your question.
Multimodal risk trajectories reveal heterogeneous paths to dementia — Zhiqi Lee, Haowen Li, Tao Liu et al.
Abstract licence: CC0
Explanation generated by Synthesis from the abstract — it is not the original paper. This is a preprint, so it hasn't necessarily been peer reviewed.