Neighbourhood Disadvantage and Vision Screening Failure Rates: Analysis of a School-Based Vision Program in Baltimore, Maryland

ABSTRACT Purpose To investigate the relationship between neighbourhood disadvantage and vision screening failure rates. Methods This analysis uses aggregate data from pre-kindergarten to eighth grade schools participating in a school-based vision programme in Baltimore, Maryland, from 2016 to 2019. Data on number of students screened and number of students who failed vision screening per grade level were recorded for each school. The Area Deprivation Index (ADI) was obtained for each school using the school’s ZIP+4 code. The association between vision screening failure rates by grade and school ADI was analysed using negative binomial regression models, adjusted for grade level and accounting for clustering by school. Results Nine hundred seventy-two grades across 117 schools were included in this analysis. Median national ADI percentile across the sample was 71 [interquartile range (IQR): 48–85] (100 = most deprived). The median grade-level screening failure rate across the entire sample was 33% [IQR: 26–41%]. School ADI was not associated with vision screening failure rate (incidence rate ratio (IRR) = 1.01 per 10 percentage point increase in ADI, 95% CI: 0.99, 1.03, p = 0.217). Conclusions In this study, there was no association between vision screening failure rates and school ADI. With one in three students failing screening in a high poverty public school district, these findings suggest a high need for vision services across schools in all neighbourhoods. Future work should investigate the impact of students’ home ADI and socioeconomic status on vision screening outcomes.


sIntroduction
Up to 28% of children in the United States have vision impairment, most commonly due to uncorrected refractive error. 1 Socioeconomic disparities impacting paediatric vision health have been well documented, with prior work showing that children of more disadvantaged backgrounds have less access to and lower utilization of eye care. They may also be underdiagnosed and undertreated for vision problems. [2][3][4][5][6] Beyond individual disadvantage, prior research has suggested that community-level disadvantage may also independently impact children's ability to access eye care and subsequent visual outcomes. [7][8][9] School-based vision programmes are one way to address barriers to eye care access by delivering vision screening, eye examinations, and glasses directly in schools, often in high-poverty communities. It is unknown whether a school's vision screening failure rates, which may reflect the student population's need for programme services, are associated with the disadvantage of the school's neighbourhood. This could have implications for public health planning and the prioritization of programme resources within a district containing schools located in communities of varying levels of poverty. The purpose of this study is to examine the relationship between neighbourhood disadvantage and school grade-level vision screening failure rates at schools participating in a school-based vision programme in Baltimore, Maryland.

Outcomes and variables
At each school, we recorded data on the total number of students screened and total number who did not pass the screening by grade level. To measure neighbourhood disadvantage for each school, we used the 2019 national Area Deprivation Index (ADI) percentile rankings based on school ZIP+4 codes ( Figure S1). 12,13 The ADI is a validated measure of area-based socioeconomic disadvantage at the census block/neighbourhood level used in health disparities research, accounting for domains such as income, education, housing quality, and employment. 14,15 In the national ADI percentile rankings, 100 represents the most disadvantaged neighbourhoods nationally. Schools were also grouped into ADI quintiles [first quintile (least disadvantaged): 1-46, second quintile: 47-65, third quintile: 66-76, fourth quintile: 77-89, fifth quintile: 90-100].

Statistical analysis
ADI rankings and number of grades across schools in the sample were descriptively reported. The number and percent of students screened and who failed screening by grade level and school ADI quintile were tabulated. Each unit of analysis is a grade at each school. We utilized population average marginal models to analyse the relationship between ADI (primary independent variable) and number of students who failed vision screening by grade level (dependent variable) using generalized estimating equations to account for school clustering effect and hierarchical data structure. 16 The models included negative binomial distribution, exchangeable working covariance structure, and robust variance. 17 Coefficients are expressed as rate ratios, with the number of students screened in that grade as the denominator. ADI was analysed as a continuous variable (Model 1) and using quintiles with the lowest quintile as the reference group (Model 2), separately. The models were adjusted for grade level. All analyses were conducted on Stata SE/ 15.1 (StataCorp, College Station, TX) with significance set at p < 0.05.

School characteristics
Out of 123 schools, 6 (5%) were excluded due to unavailable ADI data, leaving 972 grades across 117 schools in the analytic sample (Table 1). Median national ADI percentile across the schools was 71 [IQR: 48-85], although the total sample ranged from 5 to 98 ( Figure S2). The sample consisted of 98 (10%) pre-kindergarten, 110 (11%) kindergarten, 109 (11%) first, 110 (11%) second,  Table 2). In Model 2, using ADI quintiles and controlling for grade level, schools in more disadvantaged ADI quintiles had estimated 0-9% higher vision screening failure rates compared to schools in the least disadvantaged quintile, although these associations were not statistically significant.

Discussion
In this study, school grade-level vision screening failure rates did not vary by school geographic deprivation. One possible explanation for the lack of association between vision screening failure rate and ADI is that vision screening failure may not be as heavily impacted by access to eye care and social determinants of health, as compared to having a diagnosable eye condition. Because screening tests emphasize sensitivity, schools across the ADI spectrum may end up with similar general vision screening failure rates to ensure that any student who may need an eye examination is served. Additionally, students frequently need updated prescriptions, even if they have existing eyeglasses or access to eye care. 18 Thus, even students from high socioeconomic neighbourhoods may still fail vision screening. Nonetheless, with approximately one in three students failing screening in our sample, these findings suggest that there is a high need for vision services across schools in all neighbourhoods. There are several limitations. This study was based in one public school system, limiting generalizability. Further, our analysis was limited to one exposure variable and may not capture other factors pertinent to vision screening failure rates on a population-level. Future research may expand on this work by including a larger, nationwide sample of schools and additional important variables in their models. Finally, while this study utilized aggregated vision screening outcomes in a public school system with schools located in neighbourhoods that varied greatly in level of disadvantage, the results from this ecological study are not meant to be interpreted on the level of individual students. As our findings suggested a high need for vision services regardless of relative school socioeconomic status, future work is needed to determine another measure to aid in prioritizing schools for vision services.

Disclosure statement
A product used in the study described in this publication was manufactured by Warby Parker. Dr Collins was a paid consultant to Warby Parker at the time study was conducted. Dr Collins is also a Member of the Board of Directors at Warby Parker Impact Foundation. This arrangement has been reviewed and approved by the Johns Hopkins University in accordance with its conflict-of-interest policies. None of the following authors have any proprietary interests or conflicts of interest related to this submission: Hursuong Vongsachang, Xinxing Guo, David S. Friedman, and Gayane Yenokyan.

Funding
This work was supported by the National Center for Advancing Translational Sciences [TL1 TR003100].