Personalized prediction of post-SMILE refractive outcomes using a machine-learning nomogram

This study aimed to construct a personalized, machine learning-driven nomogram capable of predicting refractive outcomes following small incision lenticule extraction (SMILE). A total of 1253 eyes from 632 patients who underwent SMILE to correct myopia or myopic astigmatism were retrospectively reviewed. Comprehensive preoperative and postoperative parameters were collected. Feature selection was performed using both clinical expertise and statistical approaches. Four machine-learning models - linear regression, decision tree, random forest, and neural network - were trained to estimate postoperative manifest refraction spherical equivalent (MRSE). The nomogram was established based on the best-performing model and subsequently evaluated using an external independent cohort. Among the tested algorithms, the random forest model demonstrated the highest predictive performance, achieving a mean absolute error (MAE) of 0.18 D, root mean square error (RMSE) of 0.24 D, and an R2 of 0.92 in the validation dataset. The final individualized nomogram incorporated 11 key predictors and maintained strong predictive ability when applied to the external validation set. The overall postoperative MRSE averaged -0.08 ± 0.32 D, with 92.3% and 99.4% of treated eyes falling within ±0.50 D and ±1.00 D of intended correction, respectively. The machine learning-derived personalized nomogram provides a highly accurate and transferable tool for forecasting SMILE refractive outcomes. Its implementation may contribute to improved precision and predictability in clinical refractive surgery planning.