How Reliable and Reproducible Is Optical Coherence Tomography in Idiopathic Intracranial Hypertension? When Agreement Deteriorates in Advanced Papilledema.

BACKGROUND

Optical coherence tomography (OCT)-derived peripapillary retinal nerve fiber layer (pRNFL) thickness is central to monitoring papilledema in idiopathic intracranial hypertension (IIH). However, the structural distortion caused by optic disc edema impairs automated segmentation accuracy, particularly in advanced disease. The magnitude of discrepancy between automated and manually corrected pRNFL measurements across Frisén grades has not been fully characterized.

METHODS

In this retrospective observational study conducted at a tertiary neuro-ophthalmology clinic, 62 eyes were analyzed, including 32 eyes from 16 patients with IIH and 30 non-IIH control eyes from 15 control participants. Papilledema severity was graded by a masked neuro-ophthalmologist using the Frisén scale (Grades 1-5). pRNFL thickness was measured using the Heidelberg SPECTRALIS spectral domain OCT system. Automated segmentation outputs were independently reviewed and manually corrected by 2 masked graders: a clinical neuro-ophthalmology fellow and an ophthalmology resident, both with neuro-ophthalmology experience. Agreement between automated and manually corrected measurements was assessed using linear mixed-effects models. Intergrader reliability was evaluated with intraclass correlation coefficients (ICCs), stratified by the Frisén grade.

RESULTS

No statistically significant differences between automated and manually corrected pRNFL measurements were observed at Frisén Grade 1 (mean difference -1.1 µm, P > 0.05) or Grade 2 (-15.6 µm, P > 0.05). A trend toward significance emerged at Grade 3 (29.9 µm, P < 0.1), with large, statistically significant divergence at Grades 4 (182.4 µm, P < 0.001) and 5 (158.7 µm, P < 0.001). Sector-specific analysis revealed greater discrepancy in the temporal superior, temporal inferior, and nasal inferior sectors at Grade 3, with increasing nasal superior involvement at Grades 4-5. Intergrader ICC was good in control eyes (0.892) and lower-grade papilledema (Grades 1-2: ICC = 0.807) but decreased to moderate at higher grades (Grades 3-5: ICC = 0.449).

CONCLUSIONS

In this exploratory study, automated OCT segmentation demonstrated substantial divergence from manually corrected pRNFL measurements in higher-grade papilledema, with discrepancies exceeding 150 µm in advanced disease. The reliability of manual correction decreased to moderate levels with increasing Frisén grade. These findings underscore the need to interpret OCT-derived pRNFL measurements cautiously and in conjunction with the clinical examination in higher-grade papilledema.