Abstract
Purpose :
Ophthalmic trauma is a common condition that can lead to vision loss or blindness. Accurate prognostication is imperative for guiding treatment but can be difficult due to the variety and severity of injuries. In recent years, machine learning has emerged as a powerful tool for prognostication in various medical and ophthalmic conditions. However, its application in ophthalmic trauma has been limited. Our study aimed to develop and evaluate a machine-learning model for predicting visual outcomes in patients with ophthalmic trauma.
Methods :
Data was collected on the International Globe and Adnexal Trauma Epidemiology Study Registry from 25 centres in 11 countries, including India, Nepal, Indonesia, Guatemala, Iran, Columbia, the United States of America, Mexico, Singapore, Pakistan and Thailand. The prognostication model was developed using Generalized Linear Models via Lasso and Elastic-Net Regularization (glmnet). It predicted if a patient’s final best corrected visual acuity (BCVA) would be worse or better than 6/60. Final BCVA was taken as the BCVA recorded at the latest follow-up attended, in the absence of new injury or unrelated ophthalmic condition.
Results :
Data from a total of 4557 patients with ophthalmic trauma was used to develop the model. The male-to-female ratio was 3.59 and the average age was 32.73 years. The most important features for predicting a final BCVA worse than 6/60 were, in order of importance, open globe injury, surgical intervention being performed, presence of endophthalmitis, presence of lens injury, anterior chamber involvement, presence of scleral injury, zone 3 injury, presence of pre-existing past ophthalmic history and optic nerve involvement. Conversely, the most important features for predicting a final BCVA better than 6/60 were the unclassified setting of injury, presence of eye protection, zone 1 injury, and absence of globe injury. The area under the receiver operating characteristic curve (AUC) was 0.832, with 94.6% sensitivity and 59% specificity.
Conclusions :
Machine learning can be a useful tool for improving prognostication in ophthalmic trauma. With further refinement of the model and increasing study population size, improved accuracy of the model will be expected. Further research is needed to evaluate the applicability of the model in real-world settings and its utility in improving patient outcomes in ophthalmic trauma.
This abstract was presented at the 2023 ARVO Annual Meeting, held in New Orleans, LA, April 23-27, 2023.