Introduction
In recent years, the prevalence of myopia has been increasing worldwide and has gradually become a global public problem of great concern [
1]. There are significant ethnic and geographic differences in the distribution of myopia prevalence [
2]. East and Southeast Asians have the highest prevalence, such as China (37.7%) [
3], South Asians have a much lower rate, such as India (7.5%) [
4], black Africans have the lowest prevalence (4.7%) [
5], and white Europeans have intermediate prevalence, such as Norway (13.4%) [
6], Germany (11.4%) [
7], and Ireland (19.9%) [
8]. The prevalence of myopia in East Asians has increased by 23% over the past decade, with a slow increase in the prevalence of myopia in South Asians and minimal change in the prevalence of myopia in whites [
9]. It is expected that by 2050, the global myopic population will reach 4.76 billion, and the population with high myopia will reach 938 million, accounting for nearly 50% and 10% of the world’s population, respectively [
10].
Myopia is a multifactorial disease with a combination of genetic and environmental factors. Many studies have presented evidence regarding the risk factors for the onset and progression of myopia, such as near work, outdoor activities, excessive use of electronic devices, and parental myopia [
11]. However, among the above factors, only the causal relationship between education and outdoor activity time and the occurrence of myopia in school-aged children has been confirmed [
12], and other factors still need to be further validated by high-quality cohort studies and clinical randomized trials. Since there are also associations between the various factors affecting myopia, for example, an increase in time spent on electronic screens is often accompanied by an increase in near-work and a decrease in time spent outdoors, there are limitations when traditional statistical methods often fail to identify covariate covariance and possible confounders [
13]. Therefore, we need to explore new methods to reduce the influence of confounders, identify covariates, and measure the magnitude and importance of interactions between variables.
In this study, we conducted eye examinations and questionnaires on related factors for primary school students in grades 1–3 in Hubei Province, China, for 3 consecutive years (2019–2021), aiming to explore risk factors affecting myopia progression and to construct a personalized model to predict the probability of a child developing high myopia. This study will provide a reference for the development of myopia prevention and control strategies for adolescents.
Methods
Study population and sampling
Randomized stratified whole cluster sampling was used in this study. Hubei Province has 17 cities, including 12 prefecture-level cities, 3 directly administered cities, 1 autonomous prefecture, and 1 forested area. The whole group of grade 1–3 students from 2 elementary schools in each city was randomly included in the study. We came to schools for vision and refractive examinations every year, and they are followed closely for three years. Children with ocular diseases affecting vision, such as glaucoma, keratoconus, fundus lesions, strabismus, or a history of ocular surgery, as well as those who were using atropine eye drops or keratoplasty lenses, were excluded.
In the first year (2019), a total of 17,137 students in grades 1 to 3 in Hubei Province were sampled. A total of 15,512 students completed vision and refractive examinations in that year, and 14,213 (82.94%) valid questionnaires matching the students’ information were recovered. The number of students in the second and third years of follow-up was 13,568 and 12,766, respectively, and the response rates in 2019, 2020, and 2021 were 90.52% (15,512/17,137), 87.47% (13,568/15,512), and 94.09% (12,766/13,568), respectively. There was no significant difference in demographic characteristics between participants who completed the 3-year follow-up and those who were dislodged (P = 0.841).
Ethics, consent, and permissions
The study was approved and consented to by the Clinical Ethics Research Committee of Renmin Hospital of Wuhan University(WDRY2020-K211), following the tenets of the Declaration of Helsinki. The purpose of the study and the examination procedure was explained in detail to the students and their parents or legal guardians before the study began. A written agreement for informed consent was obtained from at least one parent, and verbal consent was obtained from each of the examined children at the time of the examination.
Ocular examination
Visual acuity(VA) was measured at a distance of 5 m using the Standard Logarithmic Visual Acuity Chart (National Standard of the People’s Republic of China, GB11533-2011). VA was converted to logMAR for analysis.
Cycloplegia was achieved by instilling at least five drops of 1% cyclopentolate in intervals of 5 min before obtaining autorefraction measurements (TOPCON RM-8800). During the refractometry, each eye was measured at least three times, and the mean was taken for statistical analysis. The spherical equivalent refraction (SE) was calculated as the spherical value of the refractive error plus half of the cylindrical value. Hyperopia was SE >−0.50 D and < + 0.50 D in both eyes; hyperopia was SE ≥ + 0.50 D in any eye; myopia was SE ≤−0.50 D in any eye; mild myopia was SE >−3.00 D and ≤−0.50 D in any eye; moderate myopia was SE >−6.00 D and ≤−3.00 D in any eye; and high myopia was SE ≤−6.00 D in any eye. The annual refractive change was defined as the total refractive change divided by the months of follow-up and multiplied by 12, with negative values indicating myopia progression. For example, if the total refractive change is −1.50 D over a 36-month follow-up period, then the annual refractive change is −0.50 D (−1.50 D/36 × 12).
Body index measurement
All subjects removed their shoes and hats when measuring height and weight. Height was recorded in centimeters (cm), and weight was recorded in kilograms (kg). Body mass index (BMI) was calculated as weight/height and recorded in kilograms per square meter (kg/m2). Based on the 2000 Centers for Disease Control and Prevention (CDC) growth icons of BMI percentile for each gender and age group [
14], students were categorized into four groups: slim (BMI < 5th percentile), normal weight (5th percentile ≤ BMI < 85th percentile), overweight (85th percentile ≤ BMI < 95th percentile), and obese (BMI ≥ 95th percentile).
All examinations were performed by trained ophthalmologists or optometrists following a standard study protocol. To ensure data quality, 5% of the students were randomly selected for repeat measurements of visual acuity, refraction, height, and weight. If the error between the two tests exceeded the permissible thresholds (visual acuity 0.1 log MAR, refraction 0.5 D, height 0.1 cm, weight 0.1 kg), corrective measures (e.g., additional training) were taken to improve the quality of the data.
Questionnaire survey
The questionnaires included general information about the students, which included age, gender, whether they were born prematurely, their eye habits, parents’ myopia condition, parents’ education level, daily outdoor time, daily near work time, time spent on electronic devices each day, frequency of eating sweets, and parents’ knowledge of myopia (The questionnaire is included as “Appendix 1”. ). The questionnaires were uniformly distributed to the students and their parents before the examination and were completed by students and parents.
Statistical analysis
The data were analyzed using the SE for the worst eye of each student. A database was created using Epi Data 3.1, and after data were entered in a double-blind manner and checked for errors, the data were analyzed using the Statistical Package for IBM Social Science Programs V25.0 (SPSS, Chicago, IL, USA) and R software (version 4.0.3,
https://www.r-project.org/). Descriptive statistics were expressed as the mean ± standard deviation (mean ± SD) for continuous variables and as the rate (%) for categorical variables. Multifactor linear regression was used to analyze the relationship between individual baseline characteristics and environmental factors (Table
1) and changes in students’ SE. Differences were considered significant at
P < 0.05 with a confidence interval (CI) of 95%.
Table 1
Variable assignments for environmental factors related to myopia
Duration of reading and writing each day | <1 h | 1 ~ 2 h | 2 ~ 3 h | ≥ 3 h |
Parents’ education level | Junior middle school equivalent or less | Senior middle school equivalent (Technical school) | Undergraduate degree (Junior college/college) | Postgraduate degree (Masters, PhD) |
Duration of electronic devices | <1 h | 1 ~ 2 h | 2 ~ 3 h | ≥ 3 h |
Duration of outdoor activity each day | <1 h | 1 ~ 2 h | 2 ~ 3 h | ≥ 3 h |
Most frequently used electrical device | television | computer | phone | tablet |
Whether to rest after a period of continuous reading | no | yes | | |
Whether to check vision every six months | no | yes | | |
Parents’ knowledge of vision care | Exactly not | A little | Some | Very well |
Supervision of children’s vision protection | never | sometimes | often | always |
Whether to wear glasses for children after myopia | no | yes | | |
Parents’ awareness of the hazards of myopia | light | moderate | heavy | severe |
Whether the child piddle | no | yes | | |
Frequency of eating sweets or carbonated beverages | never | once or twice a week | 3 ~ 5 times a week | Every day |
Multiple machine-learning approaches were employed in this work to estimate the annual myopia progression in children from year to year, such as the decision tree algorithm. 90% of the data, randomly selected, were defined as the internal validation group and the rest as the external validation group. In the internal validation group, fivefold cross-validation (80% for training and 20% for validation) was used to tune the parameters and construct an optimal machine-learning model. Then, the model was applied to the external validation group, the receiver operating characteristic (ROC) curves of several machine-learning approaches were calculated, and the model with the best prediction effect was selected to plot its calibration curve (CC) and decision curve analysis (DCA).
Discussion
Myopia progression is a process of an increase in spherical equivalent refraction. It is important to slow the progression of myopia in myopic children to avoid the development of high myopia and to minimize low vision and blindness due to complications of high myopia. As with the onset of myopia, myopia progression is a multifactorial process, with genetic and environmental factors playing independent roles and influencing each other. The amount of myopia progression varies by region, age, and sex. The mean age of the subjects in this study was 8.3 years, and the mean annual progression was − 0.68 D. It was similar to that in East Asian countries such as Hong Kong, China (−0.63 D) [
15] and Singapore (−0.47 D) [
16] and slightly higher than that in countries such as Australia (−0.31 D) [
17] and the United Kingdom (−0.41 D) [
18]. Cumulative progression at one, two, and three years of follow-up was higher in Asians than in Europeans − 0.31 D, −0.49 D, and − 0.58 D [
19]. The slopes of the progression curves at different ages of onset are essentially parallel at each age, with faster progression at younger ages. Across the age range, the greatest annual change in refractive error was seen in children aged 6–10 years, and the smallest change was seen in adults aged 26–30 years. Children under the age of 15 exhibited much faster myopia advancement than those beyond the age of 15 [
20]. If the age of onset of myopia is 7 years, the mean myopic progression in the following year is −0.58 D. For each additional year of age of onset, the annual progression of refractive error decreases by 0.07 D [
21]. We found that the rate of progression of refraction was relatively faster in females than in males (
P < 0.05). This may be due to the differences between male and female lifestyles [
22], where females usually spend more time near work, such as reading and doing homework, which is considered a risk factor for myopia, and boys spend more time on outdoor activities, which is considered a protective factor against myopia. In addition, differences in sex hormone levels between males and females and the rate of structural changes in the body may also account for gender differences in refractive changes in SE [
23].
We found a strong correlation between the rate of myopia progression and baseline refraction in 7- to 10-year-old students (
P < 0.001). In the XGBoost prediction model, baseline refraction was the second most important influencing factor. The mean age of the study population in this paper was 8.30 ± 0.94 years, and myopia occurring at this age reflects more of a genetic susceptibility to myopia [
24] and is relatively less influenced by environmental factors such as near work and outdoor activities. Lin et al. [
25] found that the bigger the SE determined by baseline optometry, the more noticeable the change of refraction, and the myopia progression rate rose by 0.135D per year for every 1D rise in baseline SE. Hu et al. [
26] found that preschool children with mild myopia at the initial visit exhibited a higher rate of myopia progression, which was the opposite of the pattern of change in school-age children, suggesting that factors associated with myopia progression in preschool children are different from those in school-age children. Verkicharla et al. [
20] found that children with severe myopia (≤−9.00D) had myopic progression fastest, followed by high (<−6.00D to −9.00D), moderate, and low myopia. This might be because myopia evolves to the point where the biomechanical qualities of the scleral extracellular matrix are altered, increasing the likelihood of refractive error [
27].
In this study, parental perception was investigated using a questionnaire, and the results showed that parental perception of myopia had a significant effect on the rate of myopia progression in their children (
P < 0.05). In the XGBoost model, parental perception was the third factor after BMI and baseline refraction. Despite the current increasing prevalence of myopia in school-aged children, parents still have misconceptions about the dangers and treatment of myopia. For example, some parents refuse to fit their already myopic children with eyeglasses, believing that myopia grows faster with eyeglasses [
28]. Choy et al. [
29] surveyed 1,396 children and their parents in Hong Kong and found that only 23.6% of already myopic students’ parents were aware that their children had refractive errors, and only 19.8% wore glasses. Among the myopic children who did not wear glasses for correction, only 50% of the children had a log MAR VA < 0.2, and the number of children whose refractive errors were corrected by appropriate prescription lenses for the above children with a log MAR VA < 0.2 reached more than 85%. Beneficial parental vision interventions, such as limiting children’s prolonged near-work time (≥ 180 min/d) and prolonged use of electronic devices (≥ 60 min/d), can reduce the incidence of myopia in children, especially at elementary school [
30]. Li et al. [
31] conducted a two-year randomized clinical trial in Guangzhou, China, and found that weekly parental education about eye care through social media could significantly prevent and control the incidence of myopia. Health education through social media significantly prevented and controlled the onset of myopia in children while delaying the average annual progression of myopia by 0.25 D. In the long run, the beneficial vision intervention behaviors of parents influenced children’s behaviors and reduced the risk of children developing into high myopia, which is consistent with the results of the present study.
The progression of myopia is a multifactorial process. Traditional methods can analyze a limited number of influencing factors and do so with low precision. Machine learning uses algorithms to find associations between data, which can provide faster output and avoid confounding influences between factors. In recent years, machine learning techniques have demonstrated unique advantages in the assisted diagnosis of diabetic retinopathy [
32], screening and diagnosis of glaucoma [
33], and diagnostic grading and treatment plan selection for age-related macular degeneration [
34]. Due to the stable development of various refractive elements in individuals at approximately 13 years of age, more accurate results can be obtained by applying their ocular parameters for prediction. In this study, we innovatively used various machine learning methods to construct a myopia progression prediction model and found that the most important influencing factors were BMI, baseline refractive error, parental eye care awareness, frequency of eating sweets, and age. Among them, large baseline refraction, large BMI, and high frequency of eating sweets were risk factors for myopia progression, while elder children and parental eye care awareness were protective factors for delaying myopia progression, effectively utilizing real-world data for the prevention and early diagnosis of high myopia. Higher BMI is significantly associated with high myopia but not with the prevalence of mild-to-moderate myopia in Korean children [
35]. Similarly, an Australian twin study found that female myopes were heavier than those without myopia in weight, but there was no significant difference in BMI between the two [
36]. Guan et al. [
37] found that age, uncorrected distance visual acuity, and SE were predictors of the occurrence of high myopia in school-age children. Among them, age is an important factor because it implies higher cumulative educational pressure, which is consistent with the results of our study. Children with rapid initial refractive error progression may require closer monitoring and follow-up, as well as early clinical therapeutic intervention, as they are more likely to develop high myopia during growth and development. Therefore, our findings can help in risk stratification and guide clinical decision-making in the management of myopia prevention and control in children. For children at higher risk, more aggressive treatments, including a higher frequency of clinical follow-up, use of low-concentration atropine eye drops, or low-concentration atropine in combination with keratoplasty lenses, are needed to control the rate of myopia progression and to avoid progression to high myopia.
In conclusion, we found that the rate of myopia progression varied according to age, sex, and myopia severity among children in grades 1–3 in Hubei Province and that we need to focus on younger children, girls, and those with high myopia in myopia prevention and control. Machine learning methods can be used to build a prediction model for high myopia using real-world data. Large baseline refractive error, large BMI, and high frequency of eating sweets are risk factors for myopia progression, while high parental awareness of eye care is a protective factor to delay myopia progression, which needs to be confirmed by observation of a longer cohort study in the future to guide myopia prevention and control in clinical practice.
Publisher’s Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.