Development of OphA-PRISM for Preoperative Prediction of Visual Outcomes and Operative Demands in Ophthalmic-Segment Aneurysm Microsurgery

Introduction: Surgical procedures for ophthalmic-segment aneurysms require multiple estimations of postoperative visual risk, successful treatment outcomes, and surgical demands. The authors created a model called OphA-PRISM to determine the risks associated with each clinical endpoint, as well as to determine whether there is additional predictive ability when using detailed anatomical modeling. Methods: The authors conducted a retrospective single-center study of 261 adult patients who were surgically treated from 1 January 1998, through 18 July 2025. Eligible patients were identified using a 408-variable registry containing 142 pre-incision candidate predictor variables. Repeated nested cross-validation evaluated the prediction of persistent visual morbidity (VISION), exposure burden (ACCESS), vascular control burden (CONTROL), reconstructive clipping requirements (RECON), and failure to achieve a composite favorable treatment result (CURE). Additional evaluation methods included calibration, probability error, anatomical comparator analyses, decision-curve analysis, and same-center temporal evaluation. Results: Twenty-three patients experienced persistent visual morbidity (VISION), while fifty patients failed to achieve a composite favorable treatment result (CURE). There were twenty-two instances where patients experienced visual morbidity despite having a patent ophthalmic artery. The out-of-fold area under the receiver operating characteristic curve (ROC-AUC) for predicting VISION was 0.804 (95% confidence interval [CI], 0.667–0.910), and the ROC-AUC for predicting CURE was 0.890 (95% CI, 0.821–0.950). In addition, the authors found that a diameter threshold greater than or equal to ten millimeters demonstrated discrimination similar to the CURE model (ROC-AUC, 0.883) and provided higher estimated net benefit than the model at review thresholds of 20–40%. However, the authors also reported that the model’s discriminatory ability for predicting ACCESS was lower (ROC-AUC, 0.608) and that it demonstrated poorer calibration, with a calibration slope of 0.42 and negative Brier skill (−0.022). Conclusions: The authors concluded that OphA-PRISM provides an endpoint-specific framework for developing anatomically based predictions prior to performing surgical procedures. Additionally, they suggested that predicting postoperative visual morbidity may require consideration of more than just whether the ophthalmic artery remains patent. Finally, the authors stated that complex models must demonstrate utility beyond routinely measured anatomical data and noted several limitations, including historical selection bias, an insufficient number of visual events, varying degrees of outcome ascertainment, and lack of independent external validation.