Machine Learning Prediction of 1-Year Refractive Error After SMILE

Development and Internal Validation of a Multi-output Machine Learning Model for Predicting 1-Year Postoperative Refractive Prediction Error After Small Incision Lenticule Extraction and Comparative Virtual Planning Analysis Against ZEISS 4.0

This retrospective single-center observational study is designed to develop and internally validate a multi-output machine learning model for predicting 1-year postoperative refractive prediction error after small incision lenticule extraction (SMILE). The primary modeling target is 1-year spherical equivalent prediction error. Secondary targets include J0 prediction error, J45 prediction error, and postoperative uncorrected distance visual acuity in logarithm of the minimum angle of resolution. A secondary objective is to use the prediction framework to derive individualized nomogram recommendations and to compare these recommendations with ZEISS 4.0 planning in a virtual treatment-planning analysis.

Inclusion Criteria:

• Eyes that underwent SMILE for correction of myopia or myopic astigmatism
• Availability of preoperative examination data and surgical planning data
• Availability of postoperative follow-up data at approximately 1 year
• Age ≥ 18 years

Exclusion Criteria:

• Previous ocular surgery in the study eye
• Ocular comorbidity likely to affect refractive or visual outcome
• Intraoperative or postoperative complications that precluded standard outcome assessment
• Missing key variables required for analysis