Purchase this article with an account.
A. S. Shah, M. T. Andreoli, C. E. Kloek, R. Wee, C. M. Andreoli; Do Ocular-Trauma Models Accurately Predict Visual Outcome After Open-Globe Injuries?. Invest. Ophthalmol. Vis. Sci. 2008;49(13):620. doi: https://doi.org/.
Download citation file:
© ARVO (1962-2015); The Authors (2016-present)
To evaluate the accuracy of the Ocular Trauma Score (OTS) and the Classification and Regression Tree Analysis (CART) Model in predicting visual outcome after open-globe injury.
We reviewed 692 open-globe injuries presenting to the Massachusetts Eye and Ear Infirmary between 2000 and 2007. We eliminated cases where initial or final visual acuity could not be obtained and where follow-up was less than 28 days. We compared the OTS and the CART Model predictions with actual outcomes, and we calculated sensitivity, specificity, positive predictive value, and negative predictive value of each model.
Based on the inclusion criteria, we analyzed 519 cases of open-globe injuries. We found the demographics of this cohort to have a median age of 38.5 years (25th-75th quartile range was 24.8 to 52.0), 83.2% men, 357 rupture injuries, 162 laceration injuries, and an initial acuity of no light perception in 27 cases (5.2%). The predicted visual outcome based on OTS did not follow the expected distribution outlined in the original study; in most cases, OTS predicted far less visual potential. The CART Model predicted vision survival versus vision loss with 91.3% sensitivity but only 57.1% specificity. The modified CART Model, which predicts final visual acuity of 20/20 to 20/400 versus less than 20/400, had a sensitivity and specificity of the 86.1 and 14.3%, respectively. In line with these findings, the negative predictive value (i.e. the ability of the model to predict vision loss) was 15.7% in the CART model and 2.8% in the modified CART model. Despite these shortcomings, in our experience 50.8% of patients with an OTS 2 and 89.2% percent with an OTS 3 or 4 attained 20/40 or better vision. Additionally, both CART models showed greater than 97 percent accuracy in predicting 20/400 or better vision.
Our data suggest that neither the OTS nor the CART model predicts outcome in open-globe injuries as intended. It is difficult to determine where these models fail. The statistics and demographics of the OTS model are not published, and neither publication clearly outlines the initial management protocol for open-globe injuries. Thus, clinicians should use these models with caution in counseling and managing patients. We suggest that future models clearly indicate management guidelines to attain these outcomes, and we advocate using multi-center data to generate future models.
This PDF is available to Subscribers Only