Forecasting Retinal Thinning and Visual Field Loss

Journal Highlights

Ophthalmology Glaucoma, January/February 2020

Download PDF

Progressive thinning of the circumpap­illary retinal nerve fiber layer (cpRNFL) thickness, as measured by optical coherence tomography (OCT), may indicate worsening optic nerve damage. Sedai et al. developed a multimodal model to forecast cpRNFL thickness at future visits.

For this observational study, the researchers enrolled 1,089 participants. Of these, 643 had glaucoma, 405 were glaucoma suspects, and 41 served as healthy controls. All underwent an ini­tial comprehensive ophthalmic exam-ination that included OCT scanning, and they were then monitored for 3.57 ± 1.69 years. The number of visits ranged from three to 30, and the mean interval between visits was 9.7 ± 9.0 months.

The researchers developed four forecasting models, based on the number of visits used (one to four); all four models used a combination of clin­ical, structural, and functional data, including deep learning–derived OCT features. The results were compared to a commonly adopted linear regression model, and the main outcome measure was the mean absolute difference and Pearson correlation coefficient between the true and forecasted values of the cpRNFL in the three cohorts.

Results showed that the most accu­rate forecasting model used three visits. The mean error was 1.10 ± 0.60 µm in healthy patients, 1.79 ± 1.73 µm in glaucoma suspects, and 1.87 ± 1.85 µm in patients with glaucoma. In contrast, the standard linear regression model showed a mean error of 1.55 ± 1.16 µm, 2.40 ± 2.67 µm, and 3.02 ± 3.06 µm, respectively, in the three groups. The Pearson correlation coefficient between the forecasted value and the measured thickness was p < 0.01 for all three groups.

In future work, the researchers plan to include visual functional parameters, which would provide a more complete outlook for individual patients.

The original article can be found here.