Prevalence Patterns of Myopia Progression in Children and Teenagers in Southern China via Real-World Screening Data: Retrospective School-Based Study (Preprint)

Abstract

The prevalence of myopia has rapidly increased in recent decades, which has become a global public health problem.

We aim to explore the risk factors and established a machine learning model for the progression of myopia in children and teenagers through a true-world longitudinal data in southern China.

This retrospective study screened 16241 children and teenagers aged 6-18 years old at baseline annually over a period of 3 years. Three risk groups were defined by the continuity of myopia progression over two years. Risk factors of myopia progression were evaluated by logistic regression and prediction models were built by multi-layer perceptron. P value was two-tailed, where P < 0.05 was statistically significant, and 95% CIs were reported.

Age, school type, uncorrected distance visual acuity (UDVA), vision correction alteration (VCA), non-cycloplegic spherical equivalent (NSE) and vision correction type (VCT) were sorted by logistic regression analysis and importance of risk factors. The weights of importance in age, school type, UDVA, VCA, NSE, and VCT were 100%, 59.2%, 47.8%, 34.4%, 24.9% and 22.4%, respectively. Age of 6-18 years old, primary school students, UDVA (logmar) between (0.2, 0.6], without VCA and without vision correction caused 11.19-fold (95%CI 6.89-18.16), 2.46-fold (95%CI 1.64-3.69), 2.16-fold (95%CI 1.71-2.72), 1.73-fold (95%CI 1.51-1.99) and 1.52-fold risk (95%CI 1.27-1.81) to continuous myopia progression than 15-18 years old, high school students, UDVA (logmar) ≤ 0.20, with VCA,and with vision correction. Besides, every increase of 1 dioptre could result in a 1.07-fold risk of continuous myopia progression. The study established a multi-factor perceptron model to predict the progression of myopia, which could achieve an accuracy rate of 0.91(95%CI 0.90-0.93) and AUCs of 0.864-0.907.

This study established a machine learning predictive model for the progression of myopia based on large-scale screening data from the real world. We discovered that age, school type, UDVA, VCA, NSE and VCT were high-risk factors for myopia progression.