Multimodal Characterization of Radiation-Induced Optic Neuropathy after Proton Radiotherapy

To evaluate the performance of multimodal ophthalmologic assessment compared with current visual acuity (VA)-based on Common Terminology Criteria for Adverse Events classification evaluation of radiation-induced optic neuropathy (RION) severity grade and to describe the relative prominence of functional, electrophysiological, and structural changes across postradiotherapy time windows.

Prospective cohort study.

Patients treated with proton radiotherapy for brain, skull base, or head-and-neck tumors between 2018 and 2023 who underwent standardized baseline and longitudinal ophthalmologic evaluations. Eyes with preexisting optic neuropathy were excluded. Of 238 patients, 89 met inclusion criteria (full assessment, no baseline deficit); 39 developed clinically diagnosed RION during follow-up.

Participants underwent repeated assessments including best-corrected VA (ETDRS), visual field (VF) perimetry (mean deviation [MD] and pattern standard deviation [PSD]), spectral-domain OCT retinal nerve fiber layer (RNFL) thickness, and pattern-reversal visual evoked potentials (VEPs) (P100 latency and amplitude). The median follow-up after proton radiotherapy was 3.83 (2.94-4.75) years. Single-variable and multivariable decision tree models were constructed to identify RION. Random forest models quantified feature importance and temporal dynamics across predefined postradiotherapy intervals (0-0.5, 0.5-1, 1-2, and 2-5 years).

Model performance (accuracy and F1 score) for identification of clinically diagnosed RION and relative importance of ophthalmologic parameters over time.

Of 39 patients with RION and 50 without, VF MD and PSD demonstrated the strongest individual performance for RION identification (accuracy 75%-79%; F1 score 0.78-0.81), outperforming VA, RNFL thickness, and VEP measures. The multimodal model combining all assessments achieved higher apparent-sample performance (accuracy 81.7% [95% confidence interval (CI), 73.9%-88.1%]; F1 score 0.83 [95% CI, 0.75-0.89]); cross-validated performance was more modest (mean area under the curve 0.64). Random forest analyses identified VF MD as the most influential parameter, followed by VEP measures and longitudinal RNFL change. Within the first year after radiotherapy, functional (VF) and electrophysiological (VEP) abnormalities were more prominent, whereas structural RNFL thinning became more evident between 1 and 2 years. Cross-sectional differences in VA were limited.

In patients with clinically diagnosed RION after proton radiotherapy, functional alterations predominated during the first year after treatment, whereas structural RNFL thinning became evident in the second year. Visual field-based assessment demonstrated greater sensitivity than VA-based grading alone; the cross-validated multimodal model achieved modest performance and did not clearly outperform VF testing alone. These findings should be considered hypothesis-generating. Prospective validation in larger cohorts is warranted.

The author has no/the authors have no proprietary or commercial interest in any materials discussed in this article.