# Dementia-R1 preprint reports prognosis gains over models up to 10x larger

_Published Monday, October 26, 2026 at 11:30 AM EDT · AI, Science · Latest · Tier 2 — Notable_

Researchers report that Dementia-R1, a framework for predicting dementia from clinical notes across patient visits, outperformed models up to 10x larger on the AMC cohort. The preprint reports an AUROC of 84.02%, a measure of how well predictions distinguish between outcomes.

The framework first trains a model to predict verifiable clinical indices extracted from patient histories, then uses reinforcement learning for prognosis. It also achieved an AUROC of 78.37% for Parkinson's disease dementia prediction in an independent hospital cohort. On the ADNI benchmark, its 7B model achieved an AUROC of 83.17%, the highest among the tested language-model baselines. These results concern prediction from clinical records rather than demonstrated patient outcomes.

## Sources

- [cs.CL updates on arXiv.org](https://arxiv.org/abs/2601.03018)

---
Canonical: https://techandbusiness.org/newswire/orYjRh3z-PejlY8m7sHf01
Published: 2026-10-26T15:30:38.294Z
Story chronology: 2026-10-26T04:00:00.000Z
Retrieved: 2026-10-26T18:03:36.864Z
Publisher: Tech & Business (techandbusiness.org)