Multivariable Normal Tissue Complication Probability Prediction of Early Grade Radiation-Induced Optic Neuropathy Using Visual Field Deficit in a Prospective Pencil Bean Scanning Proton Therapy Cohort.

Abstract

Radiation-induced optic neuropathy (RION) occurs in about 4% of extraocular tumors. Current predictive models reliably identify grade-4 RION when blindness is irreversible. We developed a sensitive multivariable Normal Tissue Complication Probability (NTCP) model integrating quantitative visual outcomes and dosiomics to predict earlier grade 2+ RION in a pencil beam scanning proton therapy cohort. This prospective study included patients with paraoptic head and neck cancer and patients with central nervous system tumor who underwent standardized ophthalmologic assessments at baseline and semiannually. Dosimetry for the optic nerves, chiasm, and retinas was analyzed. RION was defined as a visual field deficit below -6 dB (grade 2+). Predictive models-including logistic regression and random forest-were developed with multicollinearity reduction and stepwise feature selection to identify the most relevant predictors of RION. Of 238 patients, 105 had no/minor visual deficit before irradiation (179 eyes, median age 58.1 years) and were analyzed; 42/105 (23.5%) eyes developed grade 2+ RION over a 3.6-year median follow-up, with onset at 12.1 months. Lyman-Kutcher-Burman models had poor predictive value (area under the curve = 0.5). RION correlated with clinical features, including age (R = 0.24, P < .001, weak), hypertension (R = 0.23, P = .002, weak), cholesterol levels (R = 0.24, P = .004, weak), and baseline visual field mean deficit (R = -0.39, P < .001, moderate). Random forest models incorporating dosiomics achieved an area under the curve of 0.8. Baseline deficit (possibly by tumor compression/surgery) and optic chiasm volume receiving over 50Gy (V₅₀ chiasma) were the strongest predictors. Patients with minor baseline deficits were 3.29-fold higher risk to RION (95% CI, 2.87-3.38) more likely to develop RION than those without deficits. A V₅₀ chiasma above 50% was associated with 1.59 times (95% CI, 1.54-1.61) increased risk of RION compared to V₅₀ chiasma < 50 Gy. Female patients had higher NTCP than males (29.8% vs 21.0%; 95% CI, 25.0%-34.6% and 17.1%-24.9%, respectively). NTCP was also increased in patients with hypercholesterolemia (36.7%, 95% CI, 34.2%-39.3% vs 24.7%, 95% CI, 22.1-27.3) and hypertension (36.3%, 95% CI, 33.9%-38.7% vs 24.2%, 95% CI, 21.7-26.7). Prospective and standardized visual assessment revealed a 23.5% risk of grade 2+ RION using visual field perimetry. Gender, vascular comorbidities, pre-existing damage from tumor compression or surgery to the optic pathways, and dose to the chiasma were risk factors for RION.