A Novel Predictive Model Utilizing Retinal Microstructural Features for Estimating Survival Outcome in Patients with Glioblastoma

Abstract

Purpose : Glioblastoma is a highly aggressive primary brain tumor with poor prognosis. Despite frequent vision concerns in patients, the associated visual sequelae remain inadequately characterized. This study aims to assess vision and its correlation with survival outcomes in glioblastoma patients.

Methods : We studied neuro-ophthalmic features in a cohort of glioblastoma patients, measuring the thickness of the retinal nerve fiber layer (RNFL) and ganglion cell complex (GCC), as well as the microvascular densities of the radial peripapillary capillaries (RPC), and the inner and outer retina. We analyzed visual outcome and developed a machine learning algorithm to predict survival outcome. Normal controls were recruited through convenient sampling.

Results : Nineteen eligible patients (9 male, 10 female) were studied with average age at diagnosis of 69 years (range: 47-76). Median time from tumor diagnosis and radiation to the 1st eye exam was 3.6 months(0-15) and 1.6 months(0-12), respectively. Average visual acuity by Snellen was 20/25-1 (20/20–20/50-2). Worse visual fields were observed in occipital tumors with average mean deviation -14.9, while frontal tumors had better fields with mean deviation -0.23. RNFL was normal in 3/3(100%) patients before, 5/5(100%) during, and 15/16 (93%) after radiation. Patients with overall survival (OS)<15 months demonstrated thinner RNFL and GCC compared to patients with OS≥15 months (p<0.0001). Also, the foveal avascular zone (FAZ) was larger in OS<15 months at ≥4 months after diagnosis (p=0.006). Average mean radiation dose for ipsilateral eyes was 1370cGy; contralateral eyes was 1180cGy (p=0.4166). Eyes ipsilateral to radiation had no statistical difference in visual acuity or retinal components compared to contralateral eyes. Machine learning models using partial lease squares discriminant analysis were trained to identified patients with long (≥15 months) event free and overall survival.

Conclusions : Glioblastoma patients experience varying degrees of visual field defects depending on tumor site. Retinal thinning and FAZ enlargement are evident in patients with OS<15 months. A machine learning algorithm, drawing from retinal microstructure and visual field measurements was established to predict event-free survival outcome with 80% accuracy. A larger study is needed to validate the model.

This abstract was presented at the 2024 ARVO Annual Meeting, held in Seattle, WA, May 5-9, 2024.