# Development of risk prediction model for small incision lenticule extraction

## TL;DR

This study developed a machine learning model to predict which myopia patients are suitable for SMILE eye surgery, using clinical data from over 2,600 patients.

## Contribution

A novel random forest model was developed and validated for predicting SMILE suitability with high accuracy.

## Key findings

- The random forest model achieved 96.0% accuracy in validation and 95.7% in testing.
- Key predictors included BAD-D, CBI, TBI, and spherical equivalent.
- The model showed excellent performance with an AUC of 0.976.

## Abstract

This study aimed to identify risk factors associated with small-incision lenticule extraction (SMILE) surgery and develop a risk prediction model to aid in determining patient suitability for SMILE.

This retrospective study included myopia patients from four medical centers in China, enrolled between January 2021 and December 2023. The data were randomly divided into training and test cohorts at an 8:2 ratio. A random forest (RF) model was developed and optimized using three-fold cross-validation, with feature importance assessed.

The study included a total of 2,667 patients, with 2,134 patients in the training cohort and 533 patients in the test cohort. Significant statistical differences were observed in the Belin/Ambrosio Enhanced Ectasia Display and the total deviation value (BAD-D), Corvis Biomechanical Index (CBI), Tomographic and Biomechanical Index (TBI), and spherical equivalent between patients suitable for SMILE and those not suitable, in both the training and test cohorts. The univariate analysis identified ten key features relevant to SMILE. The RF model developed from the training data demonstrated high performance, with an accuracy of 96.0% in the validation set and 95.7% in the test set, an F1 score of 0.967, and an area under the curve (AUC) of 0.976 (95% CI: 0.962–0.990).

SMILE is not appropriate for all patients with myopia. The RF model, based on clinical characteristics, showed excellent performance in predicting SMILE suitability and has potential as a valuable tool for clinical decision-making in the future.