Long-term prediction of radiation-induced optic neuropathy: A mixed-effects analysis of visual field kinetics following proton therapy.

Abstract

Radiation-induced optic neuropathy (RION) is a rare but serious complication of radiotherapy, leading to progressive vision loss. The temporal dynamics of RION are poorly understood, limiting effective monitoring and intervention. We developed a predictive mixed-effects model of visual field deterioration, a sensitive surrogate marker for clinically-reported RION, by integrating longitudinal clinical and dosimetric data, to anticipate long-term visual outcomes. Out of a prospective database of 238 patients, 179 eyes from 105 patients treated with pencil beam scanning proton therapy were included. All selected eyes had no significant visual field deficit at baseline, defined as a mean visual sensitivity loss better than -6 dB. Baseline clinical characteristics, detailed dosimetric data, and longitudinal visual field assessments were collected. Temporal changes in mean visual sensitivity were analyzed using feature selection through random forest models and linear regression. A nonlinear mixed-effects model was then developed to predict the trajectory of visual field deterioration over time. Visual field deterioration progressed significantly over time, with a quadratic model best capturing the kinetics. Mean sensitivity loss accelerated with increasing age and clinical target volume. Incorporating the full dose-volume histogram, the volume of the optic chiasma received at least 40 Gy (V₄₀/chiasma), improved model performance. Simulation based on this model showed that the probability of RION increased sharply over time: 4.6 % at 2 years, and 28.3 % at 5 years. This model confirms and expands upon prior work by showing that clinical factors can outweigh dosimetric ones in predicting RION progression. Our model was capable of predicting long-term visual outcomes even in patients with limited follow-up.