Prediction of vaults in eyes with vertical implantable collamer lens implantation.

Abstract

Purpose: To design formulas for predicting postoperative vaults in vertical implantable collamer lens (ICL) implantation and to achieve more precise predictions using machine learning models.

Design: Retrospective, observational study.

Setting: Eye Clinic Tokyo.

Methods: We retrospectively reviewed the medical records of 720 eyes in 408 patients who underwent vertical ICL implantation. The data included age, sex, refractions, anterior segment biometric data, and surgical records. We designed 3 formulas (named V1 to V3 formulas) using multiple linear regression analysis and tested 4 machine learning models.

Results: Predicted vaults by V1 to V3 formulas were 444.17 ± 93.83 μm, 444.08 ± 98.64 μm, and 444.27 ± 108.81 μm, with a mean absolute error of 127.97 ± 107.92 μm, 126.41 ± 105.86 μm, and 122.90 ± 103.00 μm, respectively. There were no significant differences in error among the V1 to V3 formulas, despite the fact that the V1 and V2 formulas referred to limited parameters (3 and 4, respectively) and the V3 formula referred to all 12 parameters. 2 of 4 machine learning models-Extreme Gradient Boosting and Random Forest Regressor-showed better performance in predicted vaults: 444.52 ± 120.51 μm and 446.00 ± 102.55 μm, and mean absolute error: 118.31 ± 100.55 μm and 118.63 ± 99.34 μm, respectively.

Conclusions: This is the first study to design V1 to V3 formulas for vertical ICL implantation. The V1 and V2 formulas exhibited good performance despite the limited parameters. In addition, 2 of the 4 machine learning models predicted more precise results.