The German National Cohort: Ophthalmic Assessment, Baseline Profile, and Potential for Artificial Intelligence-Based Eye Research

Objective: To describe the ophthalmic examination protocol within the German National Cohort (NAKO)/NAKO Gesundheitsstudie, to report the baseline profile of participants undergoing ophthalmic assessment, and to illustrate the potential as a population-based resource for artificial intelligence (AI) research in eye health. Design: Baseline analysis of ophthalmic data within the nationwide, population-based multicenter prospective NAKO study. Participants: Forty-eight thousand four hundred and sixty adults in the…

By the Truvace deskOctober 19, 20262 min read1 sourcePeer-reviewedTRV-2026-1889· cite & verify

"Retinal Epithelial Cells (RPE) Grown on Scaffold" by National Institutes of Health (NIH) is marked with Public Domain Mark 1.0. To view the terms, visit https://creativecommons.org/publicdomain/mark/1.0/.

In brief

Researchers described the eye examination protocol of the German National Cohort (NAKO) and profiled the 48,460 adults who completed the ophthalmic module. They used open-source deep learning models to grade fundus image quality and measure features such as cup-to-disc ratio. They also trained standard architectures to predict age, sex and blood pressure from the images.

The work matters because it positions a nationwide, population-based dataset as a base for developing and validating eye-health AI tools in the general German adult population. The paper presents a resource and benchmark-style results rather than evidence that any tool improves clinical care, so real-world benefit is still to be shown.

Main points

The ophthalmic module covered 48,460 adults out of 205,053 enrolled, aged 19 to 74, with a mean age of 48.9 years.
Self-reported ocular disease prevalence was 4.2% for cataract, 2.0% for glaucoma and 0.9% for macular degeneration.
68.2% of fundus images were rated gradable by four deep learning quality-grading models.
Age prediction reached a mean absolute error of 2.96, and sex prediction reached 0.84 accuracy.
Systolic and diastolic blood pressure predictions had mean absolute errors of 10.78 and 6.97.

The gain

Fundus images from about 50,000 adults in a nationwide cohort, paired with systemic health data, allowed standard deep learning models to predict age, sex and blood pressure, showing the dataset's value for building eye-health AI.

The rundown

All participants had standardized biomedical examinations and questionnaires, plus nondilated color fundus imaging, visual acuity testing and a brief ocular history. Mean logMAR visual acuity was 0.01 in the left eye and 0.03 in the right eye.

Open-source deep learning models assessed image quality and morphological features, which showed systematic differences across age groups. Standard deep learning architectures were then trained on the images to predict age, sex and blood pressure.

The authors conclude that the cohort supplies visual acuity measurements, systemic health indicators and color fundus images for about 50,000 participants. They say this lays groundwork for studying eye health in the general adult population in Germany.