Evaluating efficacy of 0.125% atropine using a myopia progression machine learning model.

Abstract

Purpose: To investigate the usefulness of a machine learning (ML) model that can predict the natural course of childhood myopia in evaluation of the inhibitory effects of 0.125% atropine on the progression of childhood myopia.

Study design: Retrospective, cohort study.

Methods: A retrospective analysis was conducted on children treated daily with 0.125% atropine eye drops. The children were classified into 6-, 12-, 18-, 24-, and 30-month group based on the treatment duration. Spherical equivalents (SE) at the last treatment time point were compared with the pretreatment and ML-predicted SE. The myopia suppression rate due to treatment was calculated based on the first- and ML-predicted SE.

Results: A total of 771 eyes (402 boys and 369 girls) from 397 children were included. The participants' mean age was 8.0 ± 1.5 years. The first SE of -2.87 ± 1.67 diopters (D), treatment led to a mean SE of -3.44 ± 1.90 D, showing a significant reduction in myopia progression compared to the ML model's prediction of -4.12 ± 1.75 D. Except for the 6-month group, the last SE was statistically significantly less myopic than the predicted SE, indicating that treatment suppressed progression compared to the natural course. The mean myopia suppression rate was 53.5%, with definite differences between the groups.

Conclusion: Treatment with 0.125% atropine may suppress myopia progression in children compared with the ML child myopia prediction model. The application of a machine learning model to predict myopia progression may assist in evaluating the efficacy of 0.125% atropine treatment.