About: TrabeculetomyPred is a web-server to provide user-interface to machine learning (ML) models for predicting trabeculectomy outcomes at 5th year (post-surgery) in eyes of patients with juvenile onset open-angle glaucoma (JOAG) [Click to predict Trabeculectomy outcomes status].

We trained and developed ML models based on demographic, preoperative ocular and intraoperative surgical data of JOAG patients. The successful surgery was defined as postoperative intraocular pressure (IOP) ≤ 18mmHg and ≥ 50% reduction in IOP (from baseline) at five year of post-surgery follow-up. Feature selection techniques were used to select the most contributing features and ten-fold cross-validation was used to evaluate model performances. The ML models were evaluated, compared, and prioritized based on their accuracy, sensitivity, specificity, Mathew Correlation Coefficient (MCC) index, and mean area under the receiver operating characteristic curve (AUROC). The prioritized models were further optimized by tuning hyperparameters, and feature contributions were evaluated.

The age at diagnosis, preoperative baseline IOP, duration of preoperative medical treatment, tenon thickness, scleral fistulation technique, and administration of intraoperative mitomycin C (MMC) were identified as the main contributing parameters for efficient model training. The three models developed for a consensus-based outcome to predict the success of trabeculectomy, showed an accuracy > 86%, sensitivity >90%, and specificity >74%, using a ten-fold cross-validation.

The consensus-based ML models developed in the study have the potential in predicting the success of primary trabeculectomy that can assist in making a personalized decision for a proactive administration of surgery and thus patient counseling.