A machine-learning approach to discerning prevalence and causes of myopia among elementary students in Hubei

OBJECTIVE: Our aim is to establish a machine-learning model that will enable us to investigate the key factors influencing the prevalence of myopia in students. METHODS: We performed a cross-sectional study that included 16,653 students from grades 1–3 across 17 cities in Hubei Province. We used questionnaires to discern levels of participation in potential factors contributing to the development of myopia. The relative importance of potential contributors was ranked using machine-learning methods. The students’ visual acuity (VA) was measured and those with logMAR VA of > 0.0 underwent a autorefraction test to determine students’ refraction status. RESULTS: The prevalence of myopia in grades 1, 2, and 3 was 14.70%, 20.54% and 28.93%, respectively. Myopia rates among primary school students in provincial capital city (32.35%) were higher than those in other urban (23.03%) and rural (14.82%) areas. Children with non-myopic parents, only one myopic parent, or both parents having myopia exhibited myopic rates of 16.36%, 25.18%, and 41.37%, respectively. Myopia prevalence was higher in the students who continued to use their eyes at close range for a long time and lower in those engaged longer in outdoor activities. The machine-learning model determined that the top three contributing factors were the students’ age (0.36), followed by place of residence (0.34), starting age of education (0.21). CONCLUSION: The overall prevalence of myopia was 21.52%. Children’s age and place of residence were the important influencing factors, but genetics and environmental were also played key roles in myopia development.