Machine learning for predicting the treatment effect of orthokeratology in children

Abstract

PurposeMyopia treatment using orthokeratology (ortho-k) slows myopia progression. However, it is not equally effective in all patients. We aimed to predict the treatment effect of ortho-k using a machine-learning-assisted (ML) prediction model.MethodsOf the 119 patients who started ortho-k treatment between January 1, 2019, and January 1, 2022, 91 met the inclusion criteria and were included in the model. Ocular parameters and clinical characteristics were collected. A logistic regression model with least absolute shrinkage and selection operator regression was used to select factors associated with the treatment effect.ResultsAge, baseline axial length, pupil diameter, lens wearing time, time spent outdoors, time spent on near work, white-to-white distance, anterior corneal flat keratometry, and posterior corneal astigmatism were selected in the model (aera under curve: 0.949). The decision curve analysis showed beneficial effects. The C-statistic of the predictive model was 0.821 (95% CI: 0.815, 0.827).ConclusionOcular parameters and clinical characteristics were used to predict the treatment effect of ortho-k. This ML-assisted model may assist ophthalmologists in making clinical decisions for patients, improving myopia control, and predicting the clinical effect of ortho-k treatment via a retrospective non-intervention trial.