Retinal deep learning biomarker estimates mortality risk and how it changes with age

The study developed RetiME, a deep learning biomarker built on foundation models that maps retinal photographs to mortality risk. It was trained on a European cohort and validated in three external cohorts in Southeast Asia, Eurasia and North America. RetiME was associated with all-cause and cause-specific mortality and predicted incidence of a range of systemic and ocular diseases. It also estimated each person's curve of mortality risk across future ages, which was associated with observed age at death. Clustering these curves gave cancer-predominant, cardiometabolic and neurodegenerative patterns. The model was trained on a single European cohort. The authors present it as only one component of a wider assessment of mortality risk.

Why it is interesting: A cheap, non-invasive retinal photograph is linked to both the level of mortality risk and how it changes with age. External validation covered Southeast Asia, Eurasia and North America, and genome-wide analyses identified loci associated with the score.

Source

Nature Biomedical Engineering, 2 November 2026

Paper

Re-timing mortality risk through individualized modelling of lifespan trajectories

Authors

Xu T, Peng Q, Li H, Wang M, Zhang J, Yu M et al.

DOI

10.1038/s41551-026-01822-4

Type

Peer-reviewed article

Design

Deep learning biomarker (RetiME) from retinal photographs, trained on a European cohort and externally validated in three independent cohorts (Southeast Asia, Eurasia, North America); n and follow-up years not stated in abstract; includes clustering, interpretability and GWAS analyses

Verdict

New finding

Driver

What linked data reveals

Driver

AI in medicine