Fundus-photo vascular features track glaucoma progression

Design

Retrospective cohort with deep-learning retinal vasculometry and mixed-effects modelling

Population

1,595 eyes (757 POAG, 838 glaucoma suspects), mean 14.5 visual fields over ~10 years

Primary outcome

Association of baseline vasculometry with visual-field mean-deviation slope

Effect

+0.079 dB/year per 1-SD venous fractal dimension (95% CI 0.034-0.123); similar for arterial metrics

Predicting which glaucoma patient will progress remains hard. This study tested whether vascular features measured automatically from a baseline fundus photograph relate to later visual-field decline.

Among 1,595 eyes (757 with primary open-angle glaucoma, 838 suspects) followed for about ten years with a mean of 14.5 visual fields, several deep-learning-derived vasculometry metrics were independently associated with the rate of visual-field mean-deviation change. For example, each standard-deviation increase in venous fractal dimension corresponded to a 0.079 dB/year difference in slope (95% CI 0.034 to 0.123), with similar associations for arterial fractal dimension and tortuosity (all P<0.001).

The appeal is that fundus photographs are cheap and widespread, so vascular biomarkers could one day refine who needs closer monitoring. The honest caveat is that these associations, though statistically robust, are small per patient and come from one retrospective cohort; they are a research signal, not yet a tool to change surveillance intervals.

• Retrospective study of 1,595 eyes with about ten years of visual-field follow-up.
• Deep-learning vascular features from baseline fundus photos were associated with visual-field slope.
• Each 1-SD higher venous fractal dimension corresponded to a 0.079 dB/year slope difference (P<0.001).
• Fundus photography is cheap and widespread, making such markers attractive if validated.
• Treat as a research signal, not yet a reason to change monitoring intervals.

Why it matters

It raises the prospect of using an ordinary fundus photograph to flag glaucoma patients likely to progress, where current prediction is weak.

Don't overread it

These are associations from a single retrospective cohort with small per-patient effects; the authors stress independent validation is needed before any predictive use.

The statistics, in plain English

The associations are statistically strong (P<0.001) but small in absolute terms, a fraction of a decibel per year per standard deviation, so at the individual level they shift probability modestly and need prospective validation before clinical use.

Read the rest in the app

You have read your two free briefings this month. The app carries all 27 specialties, every morning, free — and this finding is waiting in it.

Get Daily Doctor — free

Scan to keep reading on your phone. No account needed to start.