A novel predictive model utilizing retinal microstructural features for estimating survival outcome in patients with glioblastoma.

PURPOSE: Glioblastoma is a highly aggressive brain tumor with poor prognosis despite surgery and chemoradiation. The visual sequelae of glioblastoma have not been well characterized. This study assessed visual outcomes in glioblastoma patients through neuro-ophthalmic exams, imaging of the retinal microstructures/microvasculature, and perimetry.

METHODS

A total of 19 patients with glioblastoma (9 male, 10 female, average age at diagnosis 69 years) were enrolled. Tumor characteristic, neuro-ophthalmic exam data, Optical Coherence Tomography (OCT) and OCT-Angiography data of all patient eyes were analyzed using Microsoft Excel and a Machine Learning algorithm.

RESULTS

Best-corrected visual acuity ranged from 20/20 - 20/50. Occipital tumors showed worse visual fields than frontal tumors (mean deviation -14.9 and -0.23, respectively, p < 0.0001). Those with overall survival (OS)< 15 months demonstrated thinner retinal nerve fiber layer and ganglion cell complex (p < 0.0001) and enlarged foveal avascular zone starting from 4 months post-diagnosis (p = 0.006). There was no significant difference between eyes ipsilateral and contralateral to radiation fields (average doses were 1370 cGy and 1180 cGy, respectively, p = 0.42). A machine learning algorithm using retinal microstructure and visual fields predicted patients with long (≥15 months) progression-free and overall survival with 78 % accuracy.

CONCLUSION

Glioblastoma patients frequently present with visual field defects despite normal visual acuity. Patients with poor survival duration demonstrated significant retinal thinning and decreased microvascular density. A machine learning algorithm predicted survival though further validation is warranted.