Study of myopia progression and risk factors in Hubei children aged 7-10 years using machine learning: a longitudinal cohort

Abstract

Background: To investigate the trend of refractive error among elementary school students in grades 1 to 3 in Hubei Province, analyze the relevant factors affecting myopia progression, and develop a model to predict myopia progression and the risk of developing high myopia in children.

Methods: Longitudinal study. Using a cluster-stratified sampling method, elementary school students in grades 1 to 3 (15,512 in total) from 17 cities in Hubei Province were included as study subjects. Visual acuity, cycloplegic autorefraction, and height and weight measurements were performed for three consecutive years from 2019 to 2021. Basic information about the students, parental myopia and education level, and the students' behavioral habits of using the eyes were collected through questionnaires.

Results: The baseline refractive errors of children in grades 1 ~ 3 in Hubei Province in 2019 were 0.20 (0.11, 0.27)D, -0.14 (-0.21, 0.06)D, and - 0.29 (-0.37, -0.22)D, respectively, and the annual myopia progression was - 0.65 (-0.74, -0.63)D, -0.61 (-0.73, -0.59)D and - 0.59 (-0.64, -0.51)D, with the prevalence of myopia increasing from 17.56%, 20.9%, and 34.08% in 2019 to 24.16%, 32.24%, and 40.37% in 2021 (Χ² = 63.29, P < 0.001). With growth, children's refractive error moved toward myopia, and the quantity of myopic progression gradually diminished. (F = 291.04, P = 0.027). The myopia progression in boys was less than that in girls in the same grade (P < 0.001). The change in spherical equivalent refraction in myopic children was smaller than that in hyperopic and emmetropic children (F = 59.28, P < 0.001), in which the refractive change in mild myopia, moderate myopia, and high myopia children gradually increased (F = 73.12, P < 0.001). Large baseline refractive error, large body mass index, and high frequency of eating sweets were risk factors for myopia progression, while parental intervention and strong eye-care awareness were protective factors for delaying myopia progression. The nomogram graph predicted the probability of developing high myopia in children and found that baseline refraction had the greatest predictive value.

Conclusion: Myopia progression varies by age, sex, and myopia severity. Baseline refraction is the most important factor in predicting high myopia in childhood. we should focus on children with large baseline refraction or young age of onset of myopia in clinical myopia prevention and control.

Keywords: Children; High myopia; Machine learning; Myopia progression; Refractive status; Risk factors.

Fig. 1

Prediction of the progression of myopia in children using the machine learning method. (a) ROC curves of refractive change predicted by 5 machine learning methods; (b) calibration curve of the XGBoost model; (c) decision curve of the XGBoost model

Fig. 2

Analysis of the importance of each feature using the XGBoost model

Fig. 3

Beeswarm plots of the XGBoost prediction model of myopia progression

Fig. 4

A nomogram to predict the probability of high myopia in grade 1 to 3 children in Hubei Province