Exploratory multivariate adaptive regression splines modeling of myopic progression in children receiving atropine

BACKGROUND: The increasing global prevalence of pediatric myopia has led to the widespread use of atropine for myopia control. Despite its proven efficacy, variable treatment responses and complex interactions among refractive error, treatment duration, and dosage make individualized treatment challenging. Therefore, clinicians require decision support tools for personalized treatment management. METHOD: This retrospective study analyzed 1,545 pediatric myopic eyes treated with topical atropine at Shin Kong Wu Ho-Su Memorial Hospital (Taipei, Taiwan) between 2005 and 2008. The multivariate adaptive regression splines (MARS) approach was used to examine associations between baseline clinical characteristics, atropine exposure, changes in the spherical equivalent (SE), and myopic progression. Eight clinical predictors including age, sex, Baseline SE, intraocular pressure (IOP), IOP changes, treatment duration, and cumulative and average monthly atropine dosages were used. The model performance was evaluated via repeated cross-validation and benchmarked against standard linear regression. RESULTS: Baseline SE was the strongest predictor of myopic progression, followed by cumulative atropine exposure and average monthly dosage. The MARS model demonstrated robust predictive performance (root mean square error = 0.540) and identified the baseline SE, cumulative atropine exposure, and average monthly dosage as the principal predictors of myopic progression. The average monthly dosage showed a nonlinear V-shaped relationship with a transition point near 11.66 mg/month. CONCLUSION: The baseline refractive status and atropine exposure were the principal factors associated with treatment response and may support individualized myopia management. The observed V-shaped pattern may reflect the treatment transition characteristics of the historical high-concentration atropine era (0.1-1.0%), during which the atropine dosage was frequently adjusted according to the clinical response. This nonlinear pattern suggests that the treatment response varied across different levels of atropine exposure, warranting further clinical interpretation. Therefore, children receiving intermediate monthly dosages may represent a heterogeneous group undergoing physician-directed dose escalation, resulting in an apparent nonlinear relationship. Accordingly, this pattern most likely reflects real-world treatment strategies, rather than the intrinsic pharmacological effects of intermediate atropine dosages. These findings should be used for hypothesis generation and require prospective validation before influencing clinical dosing decisions.