Analyzing the effect of surgical and corneal parameters on the postoperative refractive outcomes of SMILE in myopic eyes based on machine learning

Am J Ophthalmol. 2024 Dec 27:S0002-9394(24)00583-X. doi: 10.1016/j.ajo.2024.12.017. Online ahead of print.

Abstract

Purpose: To analyze the effect of individual parameters on the postoperative refractive outcomes of small incision lenticule extraction in myopic eyes using machine learning methods.

Design: Retrospective Clinical Cohort Study METHODS: We included 477 patients (922 eyes) of small incision lenticule extraction at Tianjin Ophthalmology Hospital and divided the patients into two groups to analyzed the factors affecting postoperative refractive outcomes based on the label of postoperative spherical equivalent (SE) ≤ -0.50D. A total of 72 parameters, including 34 biomechanical parameters, 31 morphological parameters, 4 surgical related parameters, and 3 preoperative refractive parameters obtained clinically, were included as features in the XGBoost model. The XGBoost model randomly selected 42 patients (the number of patients in the group with poor refractive outcomes) from the group with good refractive outcomes and conducted 100 influential factor analysis. Normality tests were performed on the data using the Kolmogorov-Smirnov test, group differences were compared using independent sample t-test or Mann-Whitney U-test; the correlation between various factors and postoperative SE was analyzed using Spearman correlation analysis.

Results: The average preoperative SE and postoperative SE of 922 eyes were -5.01 ± 1.48 D and -0.06 ± 0.18 D. The 10 most important factors affecting postoperative refractive outcomes included corneal biomechanical parameters (highest concavity [HC] time, the time of maximum deflection amplitude [Deflection Amp Max (ms)], and stress-strain index [SSI]), surgery-related parameters (percentage thickness ablation [PTA], maximum lenticule thickness [LTmax], and residual stroma thickness[RST]), corneal morphological parameters (radius and steep radius of the anterior corneal surface), and preoperative refractive parameters (SE and spherical diopter [SD]). When PTA ≥ 25.09%, LTmax ≥ 139 μm, SE ≤ -7.00 D, or SD ≤ -6.75 D, the postoperative SE significantly increased (all P<0.05), with averages of -0.183 D, -0.171 D, -0.188 D, and -0.184 D. After controlling for age, intraocular pressure, and corneal thickness, the postoperative SE significantly increased when HC time was ≥ 17.422 and Deflection Amp Max (ms) ≥ 16.616, reaching -0.209 D and -0.202 D.

Conclusions: Excessive tissue cutting, greater HC time, Deflection Amp Max (ms), lower SSI, and high preoperative refractive power worsened postoperative refractive outcomes. Appropriate nomogram adjustments could improve results.

Keywords: corneal biomechanics; machine learning; nomogram; refractive outcomes; residual refractive error; small incision lenticule extraction.