Identifying patients with poor visual outcomes after primary rhegmatogenous retinal detachment surgery using machine...

Area of Science:

• Ophthalmology
• Medical Informatics
• Machine Learning

Background:

• Primary rhegmatogenous retinal detachment (RRD) surgery aims to restore vision.
• Predicting poor visual outcomes is crucial for patient management and surgical planning.
• Existing scoring systems may have limitations in accuracy.

Purpose of the Study:

• To develop and evaluate machine learning (ML) models for predicting poor visual acuity after primary RRD surgery.
• To compare the performance of ML models against the established Primary Retinal detachment Outcomes (PRO) score.

Main Methods:

• Utilized data from the Japan-Retinal Detachment Registry, including 2658 patients with 6-month postoperative visual acuity.
• Developed prediction models using LightGBM, XGBoost, Random Forest, and logistic regression with 47 clinical features.
• Assessed model performance using the area under the receiver operating characteristic curve (AUROC).

Main Results:

• All developed ML models exhibited high discriminative accuracy (AUROC: 0.876-0.901).
• The Random Forest model achieved the highest accuracy (AUROC: 0.901), outperforming the PRO score (AUROC: 0.794).
• ML models effectively identified patients with poor visual prognosis (logMAR VA≥1.0 at 6 months).

Conclusions:

• Machine learning models offer a highly accurate method for predicting poor visual prognosis after primary RRD surgery.
• The Random Forest model shows promise as a superior tool for visual outcome prediction in RRD patients.
• These findings can aid in optimizing patient care and surgical strategies for RRD.