Statistical analysis

The data of the right eyes were used in the analysis (64% of the articles obtained data from one eye, and the right eye was one of the selection criteria used for data collection) (22). Chi-square tests for trend were used to assess the changes in the proportions of students with hyperopia and myopia at successive grade levels. The prevalence rates of hyperopia and myopia were compared between sexes.

In the cross-sectional analyses, we examined the relationship between academic performance and refractive statuses among students from Yangjiang in grade one to grade nine, which represents the grades of compulsory education in China, according to the following steps. First, multilinear regressions were performed in a cross-sectional study to calculate the relationships between academic performance and refractive errors (myopia or hyperopia). Second, subgroup analysis was also performed to determine the association between the refractive errors and scores for each subject. Third, multilinear regressions were used to analyze the relationship between academic performance and exact eye refraction in students with myopia and hyperopia.

The longitudinal analyses were an extension of the cross-sectional analyses. First, we performed the longitudinal analyses in Shenzhen, a higher-tier city in southern China. Second, we collected data using a cohort design to assess the longitudinal changes in the correlation over time. Third, we used a larger participant sample size from 114 schools. In longitudinal analyses, multilinear regressions were performed to calculate the relationships between academic performance and changes in eye refraction. Secondary longitudinal analyses were adjusted for additional covariates, including BMI, paternal myopia, maternal myopia, average screen time per day, average reading time per day, and average outdoor activity time per day.

In all multilinear regressions, academic performance was applied as the outcome variable, and gender, age, and BMI were included as covariates. All analyses were performed using SPSS (version 19.0, IBM Corp., New York, USA). The statistical tests were two-sided with a significance threshold of P<0.05.