A Serial Cross-Sectional Analysis of the Prevalence, Risk Factors and Geographic Variations of Reduced Visual Acuity in Primary and Secondary Students from 2000 to 2017 in Hong Kong

Background: We would like to investigate the prevalence trend, potential risk factors and geographic features of reduced visual acuity (VA) in primary and secondary schoolchildren of Hong Kong. Methods: This was a serial cross-sectional study using historical data of schoolchildren aged 6 to 15 years from the annual health checks conducted at Student Health Service Centers across Hong Kong, for the school years of 2000/2001 to 2016/2017. Results: The prevalence of reduced VA increased from 49.23% (95% CI, 48.99−49.47) in 2000/2001 to 54.34% (95% CI, 54.10−54.58) in 2011/2012 but decreased to 51.42% (95% CI, 51.17−51.66) in 2016/2017. Girls were less susceptible than boys at age 6−7 (and in grade primary 1−2), but more susceptible at older ages. The prevalence in junior grades increased while the risk effect of grade reduced over the past 17 years. Geographic variation on the risk for reduced VA existed and spatial autocorrelation was positive. The difference in prevalence of reduced VA between Hong Kong and mainland China has decreased in recent years. Cross-border students living in mainland China were associated with a lower risk for reduced VA. Conclusions: Further study was proposed to investigate the environmental association between students living in and outside Hong Kong with the prevalence of reduced VA. Multi-level research should also be conducted to investigate the influence of compositional and contextual factors on the prevalence of reduced VA.

To our knowledge, three surveys were done for estimating the prevalence of refractive errors or myopia among schoolchildren in Hong Kong (HK). A 2005−2010 survey by Lam et al showed that the prevalence of refractive errors was 34.2% and 66.6% at age 6 and 10, respectively [17]. Another survey found that the prevalence of myopia in local and international schoolchildren in 2001 were 85−88% and 60−66%, respectively, with the highest in the Chinese group (82.8%) and the lowest in the white group (40.5%) [18]. The third survey noted that the prevalence (± standard deviation, SD) of myopia at age 5−16 was 36.71% (±2.87%) from 1998 to 2000 [19]. However, they had three limitations.
First, their sample size was small because large-scale adoption of cycloplegic refraction was not feasible. In mainland China, two retrospective serial cross-sectional studies were conducted to estimate the prevalence of reduced VA in large populations, assuming that reduced VA would be a proxy measure for myopia [20,21]. Their rationale was based on the findings of three previous studies of testing the sensitivity and specificity of using unaided VA to screen for refractive errors among students in Singapore and Australia [22][23][24].
Second, they did not measure geographic variations on the risk for reduced VA and spatial autocorrelation in HK. Previous studies proved the associations of areas of residence (urban/rural region), housing types (flat room/building), living environment (frequency of seeing green) and economic status (socioeconomically advantaged/disadvantaged) with myopia, one major cause of refractive errors [25][26][27][28]. If these variables were similarly or dissimilarly clustered with the spatial location, there would be positive or negative spatial autocorrelation of reduced VA.
Third, they estimated the prevalence at a single time point but did not describe the dynamic changes over a series of time points, e.g., cross-boundary students with distinct sociodemographic features may have led to changes in prevalence over time. In mainland China, two retrospective serial cross-sectional studies were conducted to estimate the prevalence of reduced VA in Guangzhou, China from 1988 to 2007 and the whole of China from 1985 to 2010. In contest, no serial cross-sectional studies, whether prospective or retrospective, have been conducted in HK.
For understanding the epidemiologic trend of reduced VA, we would like to review the historical data of Health Programs at Student Health Service Centers (SHSCs) of the Department of Health, Hong Kong Special Administrative Region (DH, HKSAR) [29]. The objective of our study was to investigate the prevalence trend, potential risk factors and geographic features of reduced VA in primary and secondary students from the school year 2000/2001 to 2016/2017.

Study Area
The Student Health Service (SHS) of the DH, HKSAR has been offering annual health checks for primary and secondary students (including local and international schools) of HK since the 1995/1996 school year for primary students and the 1996/1997 school year for secondary students. This includes regular VA assessment, with the aim to detect any VA problems of students and whether these problems need referrals or have been appropriately corrected. Clinical data pertaining to visual assessment of these students were obtained from 12 SHSCs throughout HK.
The Hong Kong Special Administrative Region is divided into 18 administrative districts, which are: Central and Western, Wan Chai, Eastern, Southern, Yau Tsim Mong, Sham Shui Po, Kowloon City, Wong Tai Sin, Kwun Tong, Kwai Tsing, Tsuen Wan, Tuen Mun, Yuen Long, North, Tai Po, Sha Tin, Sai Kung and Islands [30].

Study Design
This is a retrospective, serial, cross-sectional study covering primary and secondary students aged 6 to 15 years attending the annual health check offered by the SHS from the school year of 2000/2001 to 2016/2017, with the exception of 2009/2010, since in that school year, SHS had to take part in the Human Swine Influenza Vaccination Program, and therefore, annual appointments were only provided to Primary 1 (P1) to Secondary 1 (S1) students.

Source of Data
Available records between 2000/2001 and 2016/2017 from the SHS database were extracted for relevant data; including age, sex, school type, grade of student, school district, location of SHSCs, type of housing, home district and VA test result. The relative age and sex composition of the study population was retrieved from the Student Enrollment Statistics of Education Bureau [31].

Definitions
VA testing was performed for distant vision using a Logarithm of Minimum Angle of Resolution (logMAR) VA display chart at 3 or 4 meters. Students with an unaided logMAR VA of 0.3 or better were recorded as having satisfactory vision. The test was done with visual aid if the students was wearing glass or contact lens. Reduced VA was defined as having an unaided logMAR VA of >0.3, or wearing glasses or contact lenses (excluding Ortho-K lenses).

Data Analysis
We used multiple imputation to create 20 data sets with imputed values for handling the missing data. The means and odds ratios (ORs) were calculated by using SAS PROC SURVEYMEANS and PROC SURVEYLOGISTIC, respectively. We combined the outcomes of 20 datasets and provided the final estimates with 95% confidence interval (CI) in SAS PROC MIANALYZE.
The prevalence of reduced VA was reported in crude rate, age-sex adjusted rate after adjusting the sampling weight of age and sex composition to poststratification weight for each school year, and standardized rate for selecting the study population for 2000/2001 as a standard population. We divided the sample data into 10 strata (5 age groups (6−7, 8−9, 10−11, 12−13 and 14-15) × 2 sex groups (boys and girls)), and combined them with the population data from Education Bureau to do poststratification analysis and direct standardization. Mann-Kendall trend test is a non-parametric test used to analyze the data for consistently increasing or decreasing trends in the series. We tested the null hypothesis that there was no monotonic trend for the prevalence of reduced VA over the study period. Subgroup analyses stratified by age (6−7, 8−9, 10−11, 12−13 and 14−15) and grade (P1-P2, P3-P4, P5-P6, S1-S2 and S3-S4) were also done to identify either consistency of or large differences in the prevalence and ORs of reduced VA among different categories of students.
We performed univariate logistic analysis with and without adjusting for age and sex and fitted all independent variables into multivariate logistic analysis to measure the strength of their association with reduced VA. The results were presented in four time intervals Chi-square independence test was used to test whether the presence of reduced VA and the 18 districts (home or school) were independent when Kruskal-Wallis test were used to test whether the prevalence of reduced VA were identical among the 18 districts (home or school). Mantel test was computed to test whether there was a relationship between the community distance from Central and Western District (where the city center is located) and the prevalence (or AORs) of reduced VA. Moran's Index was calculated to test whether there was a spatial clustering of the prevalence (or AORs) of reduced VA associated with the community distance from Central and Western District and also measure the overall spatial autocorrelation.
All data extraction, processing and analyses were performed using the R package version 3.5.1 (R Development Core Team, 2018) and SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). The results were reported in estimate with 95% CI and with p-value < 0.05 being considered statistically significant.

Ethics Approval
Ethics approval was granted by the Institutional Review Board (IRB) and Ethics Committees (EC) of the Hong Kong University/Hospital Authority Hong Kong West Cluster (UW 18-071) on 11 January, 2018. Availability of data and computing code: due to the ownership, the original dataset would not be available without authorization from the Department of Health in Hong Kong but the computing code are available from the corresponding author on reasonable request.

Demographics
The overall frequency of schoolchildren meeting the inclusion criteria were 6

Age and Sex
Overall, girls have higher risk for reduced VA relative to boys with stable OR of around 1.1 throughout the study period in univariate and multivariate logistic analysis.
If stratified by age, girls were less susceptible to reduced VA at 6−7 years but became more susceptible at 8 The result showed that an interaction effect between age (or grade) and sex was statistically significant. Sex would modify the effect of age (or grade) on reduced VA. Students with older age (or more senior grade) were associated with a higher risk of reduced VA, regardless of their sex. Being a girl meant a fixed reduction of risk regardless of age (or grade), but with an additional incremental risk according to the interaction term between age (or grade) and sex. The net effect of being a girl (compared to being a boy) was initially negative, but subsequently became positive, with regard to the risk of having reduced VA.

School District and Home District
Throughout the study period, chi-square independence test (p-value, <0.01) showed that the presence of reduced VA and the 18 districts (school or home) were not independent and had a significant relationship. Kruskal-Wallis test (p-value, <0.01) showed that the prevalence of reduced VA was nonidentical among the 18 districts (school or home), and at least one of them was differentiated from the others in relation to the prevalence of reduced VA.  Table 4. Relative change in odds ratio of school type (and grade) on reduced visual acuity before and after adjusting for age in logistic analysis.    We showed that the spatial autocorrelation was positive between the distance from home district to city center (at the Central and Western District) and the prevalence (or AORs) of reduced VA, that is, there would be clustering of home districts with similar estimates of prevalence (or AORs).

Prevalence
Our results showed that the prevalence of reduced VA in 2010/2011 were 29.97% for age 6−7 and 58.06% for age 10−11. Inferring from previous reports in mainland China of refractive errors accounting for 85−97% cases of reduced VA [4][5][6][7][8][9][10], our estimated prevalence of refractive errors in Hong Kong is lower than the earlier report by Lam et al. of 34.2% and 66.6% for the respective comparable age groups [17]. In contrast to that study, the ethnicity of our subjects was not limited to Chinese, nor were the recruitment sites limited to just six specific primary schools located in Hong Kong.
Prevalence of myopia from a previous survey in Taiwan was 21% and 61% at age 7 and 12, respectively [32]. In our study, the prevalence of reduced VA was 26.08% and 61.48%, at age 6−7 and 12−13, respectively, suggesting a lower prevalence of myopia compared to Taiwan, after making similar inference from the relationship between reduced VA and myopia as mentioned.
Data from mainland China showed that the prevalence of reduced VA amongst schoolchildren aged 7 to 18 years were 38.5%, 49.5% and 56.8% in 2000, 2005 and 2010, respectively, compared to our local estimates of 49.23%, 53.02% and 56.17% for the same corresponding period, showing decreasing differences over time between HK and mainland China schoolchildren from 2000 to 2010 [21].
Overall prevalence of reduced VA was negatively associated with the proportion of primary to secondary schoolchildren. Although subgroup prevalence of each age group remains the same, differences in annual birth rate can cause fluctuation of age composition of study population and influence the overall prevalence, so that a smaller proportion of younger or junior students would raise the overall prevalence. Thus, we used a direct standardization method to adjust the age and sex composition from sampling weight to standardized weight for reviewing the prevalence trend of reduced VA, which was not considered in the two previous studies from mainland China.

Age and Sex
Although univariate and multivariate logistic analysis showed that girls were more susceptible than boys to having reduced VA, our subgroup analysis actually showed that being a girl was less susceptible at age 6−7 or grade P1-P2, and only became more susceptible when older. Significant interaction term showed that there would be multiplicative effect of combining age (or grade) with sex to have a joint effect greater than the product of their individual effects. This finding correlates with the higher incidence of myopic progression among girls compared to boys noted in most studies [3,[33][34][35][36][37] and that this tends to occur as they reach adolescence [38]. Possible explanations included changes in lifestyle behavior influencing near vision usage, and the earlier onset of puberty among girls [39].

Student Health Service Centres
Since one SHSC served students coming from various school or home districts, we assumed that variations would exist among SHSCs. In the univariate analysis, SHSCs contributed to the variations on the prevalence of reduced VA. After adjusting for other variables, the risk effect of SHSCs reduced but still existed. We expected that the background of the students, for example, age, sex, home or school district, would partially but not fully explain the variations caused by SHSCs.

School Type and Grade of Students
The prevalence of reduced VA changed disproportionally among different age or grade subgroups. Throughout the study period, the magnitude of fluctuation of prevalence in younger age (6−9 years) or junior grade (P1-P4) was greater than that in older age (10-15 years)  showed that the extent of overestimating the association between grade (and school type) and reduced VA had enlarged. Since the confounding effect reflected the natural relationships between lifestyle, habits and other characteristics, we expect age might possibly be associated with other potential risk factors for reduced VA which are not covered in our study.

Living Environment and Social Classes
Living in villas or squatter was associated with a lower risk for reduced VA relative to public or private apartment blocks. A possible explanation may be that villas and squatter are usually located in less developed areas with a lower population and building density, while apartment blocks are more likely to be located in urbanized areas with a higher population and building density. This finding was consistent with a previous study showing that a greater number of floors/levels of housings or higher frequency of seeing green were related to increasing prevalence of myopia, one major cause of reduced VA [27,28].
Relative to subsidized HOS flats and public housing (or squatter), living in private housing (whether apartments or villas) was associated with a higher risk for reduced VA. This is similar to previous report of positive association between myopia and socioeconomic status, since only households with higher than average income can afford private housing in HK [26]. Interestingly, the ORs of private housing relative to subsidized HOS flats and public rental housing (or villas relative to squatter) decreased

Geographic Variation and Spatial Autocorrelation
Despite the fact that HK has a total area of 1108 km 2 only, the spatial location was not independent of the presence of reduced VA. There was geographic variation and positive spatial autocorrelation between the spatial location and presence of reduced VA on the map. In 2000/2001−2003/2004, school districts were not spatially clustered while home districts with the lowest AOR were located at the northern areas of HK (Tuen Mun, Yuen Long, North and Tai Po districts), where there is a lower population density and urbanization. During 2013/2014−2016/2017, school districts were also not spatially clustered, while home districts with the highest AOR were located at the southern areas of HK (Central and Western, Wan Chai, Eastern, Southern and Kowloon City districts), where there is a high population density and urbanization. Our findings are consistent with previous studies that showed living environment being a key factor for myopia prevalence, which is a major cause of reduced VA.
Compared to the student's home district, most school districts had lower AOR for reduced VA in . AORs tended to be close to 1 and their mean on average by school district, but far from 1 and their mean on average by home district. We expected that the effect of geographic variation on AORs tended to be mitigated by school district but increased by home district.
The result of measuring spatial autocorrelation differed between school and home districts. Mantel test and Moran's Index reported higher value of correlation coefficients for home district than school district. Home districts with similar AORs were likely to be clustered together, but this clustering did occur for school districts with similar AORs. Spatial location was significantly correlated to the presence of reduced VA by home district but not school district.
We deduced that the association of spatial location and attributes of the 18 districts contributed to the distinct results of AORs, geographic variation and spatial autocorrelation between school district and home district. The attributes of school districts (for example; culture, academic performance or teaching style) were less likely to be associated with the community distance from city center (located at the Central and Western District), in contrast to the attributes of home districts (for example; population density, urbanization and greenery ratio) which were more likely to be associated. Thus, home districts (but not school districts) with similar attributes were more likely to be clustered together in relation to the community distance from city center.

Cross-Boundary Students
Mainland China started to be an option of student's home district from 2013/2014 onwards. The AORs of schoolchildren living in mainland China (but commuting daily to HK to attend school) relative to the Central and Western District in HK were 0.499 and 0.477 in 2013/2014 and 2016/2017, respectively, so that cross-boundary students were less susceptible to reduced VA than local HK students (Supplementary Table S7). Moreover, the fertility trend in HK reported that the proportion of live births to non-HK citizen parents increased from 1.29% in 2001 to 29.18% in 2012, when HK became a popular destination for birth tourism for mainland Chinese (Supplementary Table S8). It is possible that the decreasing prevalence of reduced VA among the younger age or junior grade students starting from 2012/2013 may be related to the increased proportion of such students who often resides in less densely populated areas across the border from HK.
Since 2013, birth tourism has decreased greatly after new restrictions on cross-border entry were adopted by the HK Government. As the school entrance age is generally around 6, we postulate that the prevalence of reduced VA among the newly admitted students would increase again from 2019/2020 onwards, to reflect the previous local trend before the large influx of cross-border students. We recommended conducting further study to compare the environmental association on the prevalence of reduced VA between children residing locally in HK with those residing in mainland China (but attending school in HK).

Strengths and Limitations
The overall prevalence of reduced VA was negatively associated with the proportion of primary to secondary schoolchildren. Assuming that the prevalence of reduced VA in each age or grade subgroup remained the same, fluctuation in the composition of study population would change the overall prevalence, that is, a smaller proportion of younger or junior students would raise the overall prevalence. Thus, we used direct standardization method to adjust the age and sex composition from sampling weight to standardized weight for reviewing the trend of prevalence of reduced VA, which was not considered in the two previous studies from mainland China.
Our study considered interaction effect, confounding, geographic features and dynamic changes over time, which were not reported previously. Significant interaction effect identified that the risk associated with being a girl relative to a boy was increasingly shifted to a later stage of development. Age was the common cause to the grade of students and presence of reduced VA. Geographic variation on the risk for reduced VA existed and spatial autocorrelation was positive. Cross-border students had a lower prevalence of reduced VA than local students, and may have been related to the decreasing prevalence of reduced VA among younger students since 2012/2013.
The limitation was that we considered personal-level factors but did not cover the district-level factors and measure their risk effect in our analysis. We recommended conducting a multilevel research to examine how the compositional and contextual factors affect the prevalence of reduced VA by combining the district-level data from other official departments such as median monthly household income, household size, population density and greenery scale [40,41].

Conclusions
Since the prevalence of reduced VA was negatively associated with the proportion of primary to secondary schoolchildren, it is important to adjust the sampling weight of age and sex composition to standardized weight in a serial cross-sectional study. In subgroup analysis, the interaction effect between age (or grade) and sex was statistically significant. Girls were less susceptible than boys to reduced VA at age 6−7 or grade P1-P2 but became more susceptible at older ages. There was an increasing prevalence of reduced VA in junior grade so that the onset of reduced VA has been advanced earlier, thus lessening the risk effect of higher grades over the past 17 years. Schoolchildren in HK had lower prevalence of myopia than those in Taiwan, whereas the difference of prevalence of reduced VA between HK and mainland China has lessened with time. The association of housing type with reduced VA is likely affected by both living environment and social classes. Although the area of HK is small, there are geographic variation on the risk for reduced VA and positive spatial autocorrelation. Cross-border students living in mainland China was associated with a lower risk for reduced VA and further study should be considered for investigating the environmental association between students living in and outside HK with the prevalence of reduced VA. We did not include district-level factors in our analysis and suggested to perform a multilevel research to examine how the compositional and contextual factors affect the prevalence of reduced VA.
Supplementary Materials: The following are available online at http://www.mdpi.com/1660-4601/17/3/1023/s1, Table S1: Demographic information of schoolchildren included in the study, Table S2: Missing data pattern of variables of schoolchildren included in the study, Table S3: Subgroup analysis of the prevalence of reduced visual acuity, stratified by age and grade, Table S4: Subgroup analysis of the odds ratios (girls relative to boys) of reduced visual acuity, stratified by age and grade,