Vision survival after open globe injury predicted by classification and regression tree analysis.

OBJECTIVE To assist ophthalmologists in treating ocular trauma patients, this study developed and validated a prognostic model to predict vision survival after open globe injury. DESIGN Retrospective cohort review. PARTICIPANTS Two hundred fourteen patients who sought treatment at the Wilmer Ophthalmological Institute with open globe injuries from January 1, 2001, through December 31, 2004, were part of the data set used to build the classification tree model. Then, to validate the classification tree, 51 patients were followed up with the goal to compare their actual visual outcome with the outcome predicted by the tree grown from the classification and regression tree analysis. METHODS Binary recursive partitioning was used to construct a classification tree to predict visual outcome after open globe injury. The retrospective cohort treated for open globe injury from January 1, 2001, through December 31, 2004, was used to develop the prognostic tree and constitutes the training sample. A second independent sample of patient eyes seen from January 1, 2005, through October 15, 2005, was used to validate the prognostic tree. MAIN OUTCOME MEASURES Two main visual outcomes were assessed: vision survival (range, 20/20-light perception) and no vision (included no light perception, enucleation, and evisceration outcomes). RESULTS A prognostic model for open globe injury outcome was constructed using 214 open globe injuries. Of 14 predictors determined to be associated with a no vision outcome in univariate analysis, presence of a relative afferent pupillary defect and poor initial visual acuity were the most predictive of complete loss of vision; presence of lid laceration and posterior wound location also predicted poor visual outcomes. In an independent cohort of 51 eyes, the prognostic model had 85.7% sensitivity to predict no vision correctly and 91.9% specificity to predict vision survival correctly. CONCLUSIONS The open globe injury prognostic model constructed in this study demonstrated excellent predictive accuracy and should be useful in counseling patients and making clinical decisions regarding open globe injury management.