Association Between Visual Impairment and Income Change: A Longitudinal Follow-up Study With a National Health Screening Cohort

We evaluated the inuence of visual impairment (VI) on income change using the longitudinal database of a Korean National Health Insurance Service cohort. A total of 5,292 participants ≥ 40 years old and registered as visually impaired persons were selected at a 1:4 ratio with 45,081 controls matched for age, sex, and income level. The income level of both the VI and control groups increased over time. In the VI group, the income levels 3, 4 and 5 years were higher than the initial value, while the income levels from 1 through 5 years were increased each year in the control group. The rate of change in income between time and VI were signicant. In the subgroup analysis considering age, sex, and severity of VI, the rate of change in income were signicant in < 65 years old subgroups. Regarding the severity of VI, a signicant interaction was found for the mild-to-moderate VI subgroup. Although both the VI and control groups showed increased income levels over 5 years, the degree of income increase in the VI group was relatively lower than that in the control group. This nding was prominent in the middle-age subgroup. These results strongly suggested that VI induced an income inequality.


Introduction
Visual impairment (VI), including blindness, is a serious disability that has a strong impact on quality of life. In particular, VI is well-known to be associated with lower socioeconomic status (SES) [1][2][3][4][5][6] . A relationship between VI and income level has been reported in diverse studies [2][3][4][5] ; the results, varying between non-signi cant and signi cant associations, implicate a complex link between the two factors.
It is frequently asserted that VI is both a consequence and a cause of low income 3,5−7 . Whereas some evidence suggests that low income is a major barrier to uptake of ophthalmologic care and leads to lower medical compliance 8-10 , other evidence indicates that VI reduces the earning potential of impaired patients and/or the household members who care for them 7,11 .
Due to the multifactorial nature of social-economic status, identi cation of factors affecting income level is non-trivial, and certainly, there are numerous confounding factors that need to be considered and controlled for 1,3,12 . In this respect, the majority of previous studies' cross-sectional design is a limiting factor, as it does not allow for analysis of any temporal relationship between visual impairment and income change 1,4,13−16 . In order to clarify the association between these two factors, a longitudinal study design with a large population-based database is required.
A comprehensive understanding of the link between VI and income can be utilized to make socioeconomic policy that effectively supports people with VI. Furthermore, it might help to break the social and personal links between those two factors. The current study's purpose was to evaluate VI's impact on longitudinal income change. To this end, changes in the income levels of VI and control groups in a representative sample cohort dataset of the Korean National Health Insurance Service (KNHIS) were followed annually for 5 years. The groups' initial income levels and demographic factors were matched, and additionally, subgroup analyses according to age, sex, and severity of VI were performed. The sociodemographic variables between the VI and matched control groups are summarized in Table 1. No signi cant inter-group differences were found in age, sex, and baseline income level, region of residence, obesity, smoking status, SBP or total cholesterol (all P > 0.05). However, the VI group was higher DBP (> 90mmHg), fasting blood glucose, CCI scores and less likely to have frequent alcohol consumption than the healthy controls (all P < 0.05).  Table 2 shows the changes in the mean values of income level, beginning with the initial level and proceeding over the 5-year course of the annual follow-ups. The income levels increased according to the duration of follow-up in both the VI and control groups. In the VI group, the income levels post 3-5 years were signi cantly increased relative to the initial value (P = 0.018, < 0.001 and < 0.001), while the income levels post 1 and 2 years were not (all P > 0.05). In the control group, the income levels were increased signi cantly each year from post 1 year through post 5 years (all Ps < 0.001).

Results
The rate of change in income between time and VI were signi cant (interaction effects, P = 0.003). The EV of income level for the VI group was estimated to be -0.014 (P = 0.771).
In our subgroup analysis for age and sex, time and VI's interaction effects impacting income level were signi cant for the < 60-years-old subgroups of both men and women (P < 0.001 and 0.035). The EV of income level for VI was estimated to be -0.019 and − 0.020, respectively, in these subgroups (P = 0.813 and 0.861). In the subgroup of men aged < 60 years old, the income level of the VI group did not differ from the initial level to post 1 year through post 5 years; by contrast, it signi cantly increased from the initial level to post 1-5 years (all Ps < 0.001) in the control group. Similarly, in the subgroup of women < 60 years old, the income level of the VI group decreased at post 1 year (P = 0.035) and did not differ thereafter, whereas the income level of the control group increased each year (post 1-5 year, each P < 0.001).
The ≥ 60-year-old men and ≥ 60-year-old women did not show any signi cant interaction effects between time and VI for income level. According to the severity of VI, the longitudinal changes of income level were analyzed as shown in Table 3. In the mildto-moderate VI group (n = 4,170), the income levels of post 3-5 years were higher than the initial value (P = 0.028, < 0.001 and < 0.001, respectively), while the control group showed increased income throughout the post 1-5 years (all Ps < 0.001). The rate of change in income between time and VI for income level was signi cant (P = 0.005), and EV of income level for VI was estimated to be -0.016 (P = 0.767) in this subgroup. Meanwhile, the severe VI group (n = 588) did not show any signi cant interaction between time and VI for income level (P = 0.229).

Discussion
In this study, in order to investigate the impact of VI on longitudinal income change, health insurance data was analyzed for a large national population cohort. We performed an in-depth analysis of the effects of VI on income changes in both VI and matched-control groups. The income level increased in both groups over the course of 5 years, but the increase was smaller in the VI group. Although we had matched the initial income level of the VI and control groups, the income gap between them widened over time. The interaction of time and VI for income level was signi cant in a linear mixed model. A similar nding was observed in a subgroup analysis of < 60-year-old participants.
Although there have been a few studies on VI and income level to date 6,13,15,16 , the economic consequences of VI have rarely been evaluated in representative samples of national populations. Also, the present study's longitudinal design contrasts with the majority of studies published thus far, which have been of cross-sectional design 3,4,6,14,17 . The results of cross-sectional studies typically have limited implications due to the lack of temporality of risk factor data; this means that the causal relationship between, for example, VI and income level, would have to be interpreted with caution. The present study, on the other hand, using the linear mixed model, was able to prove the interaction of time and VI for income level.
As is consistent with the fact that income and real-estate assets generally increase with age 18-20 , our results showed that income level increased in both the control and VI groups. The nding of income growth, in itself, in the VI group might be taken as an encouraging result. However, the rate of increase in income level was, characteristically, lower than that of the control group. Moreover, most of the VI group may well have had a lower baseline income compared with the control; indeed, several cross-sectional studies have noted low baseline income for VI relative to non-VI individuals 5,21,22 . And although we included initial-income-matched control participants, income growth was lower in the VI group.
This is indicative of a wider gap in total income/assets between the VI and control groups in the real world.
As could be expected, individuals with VI are known to have less economic capacity. Brezin et al. 13 found monthly household incomes to be lower for the low vision (€1255) and blindness (€1587) groups than for those having no visual problems (€1851). In Britain, the risk of poor vision has been associated with social class (i.e., unskilled manual workers) 23 . People suffering VI have been deemed to be at greater risk of unemployment, permanent disability, being a member of the working class, lacking skills-development opportunity, being less recognized for their work, and earning an inadequate income 24 .
Aging by itself is a source of disability and a universal risk factor associated with VI. Rates of VI and blindness have been documented to increase sharply with age, beginning at about 65 to 70 years 22,25,26 . In the present study, the effect of age on income change was adjusted for by subdividing and analyzing age based on the age of 60. Even after this adjustment, under the age of 60, both men and women showed a greater increase of income in the control group than in the VI group. Interestingly, for those over the age of 60, contrastingly, income change over time between the VI group and the control group was not signi cant. The younger age group, certainly, would be expected to be economically more active than the older age group. Therefore, for them, the impact of VI on employment status, working performance and income level could be especially strong. More research is needed to assess whether such income differences as shown in our under-60 group between VI sufferers and non-VI individuals can be explained by other socioeconomic differences.
In present study, income changes also were analyzed according to severity of VI. In the mild-to-moderate group (n = 4,170 for VI), most of the study subjects showed an interaction effect between time and VI for income level. However, the severe VI group (n = 588 for VI) showed no signi cant income-change differences over time. We considered that the relatively small size of the severe VI group was insu cient to secure the statistical power.
In addition to visual function, educational level and SES also interact with each other 2-7, 12 . VI, educational level, and SES act in similar though different ways to produce low income levels. That is, the effect of an individual's VI on his/her income change might not be direct only, but might also emerge from other, intermediary and perhaps complex determinants that remain, pending investigation.
There are some limitations to this study. First, limitations of available data precluded us from considering the leading causes of VI. Due to the study design's use of KNHIS data, there was no speci c data on VI causes. In epidemiological investigations, the major cause of VI has been mostly age-related macular degeneration in developed countries, and cataract in under-developed countries [27][28][29][30] . Analysis of the causes of VI could help to understand economic inequality caused by VI. Second, selection bias may have in uenced our results. The use of registers to estimate VI prevalence is in any case controversial, since a high proportion of subjects thus impaired do not register 31,32 . Moreover, in Korea, application for registration of disability is directly made by the individual him/herself; and there is a strong possibility that a high proportion of VI persons who do register are those who experience economic di culty. Finally, in this study, low vision and blindness were included together and analyzed as VI, but the economic disparity between subjects with "low vision" and "blindness" was far greater than what we expected.
Despite these limitations, this study reports, based on a longitudinal database, nationwide estimates of how VI affects income change according to subject age and severity of VI. We showed that in a large representative sample of Korea, the growth in the income level of the VI group was less than that of a control group matched for age, sex, region of residence and income level. Although additional research is needed to more thoroughly elucidate and target the drivers of disparity, our ndings identify areas requiring improvement for people with VI. Based on these data, a speci c nationwide database for the SES of VI could be compiled, and in turn, policies would be formulated to provide the most appropriate nancial and social assistance. impaired. With properly documented evidence of a VI, an assessment committee discusses the feasibility of the VI registration. In Korea, the degree of VI is typically divided into 6 grades according to the its severity; in the database, the data are then divided into two grades (mild-to-moderate VI group, grades III-VI; severe VI group, grades I-II).

De nition of Income
Income level was divided into deciles of population based on KNHIS annual premiums (Supplementary Table 1).
Medical Aid bene ciaries were added to the lowest income level 36 . Income change for participants was de ned as an income-level change between the income prior and closest to the day of VI registration and the income 5 years after that.

Statistical Analyses
The differences in the rates of the general characteristics were compared using the Chi-square test. The differences in the mean values of income between the initial index date and the 5-years-post-VI-registration date were compared using the paired t-test.
To estimate the interaction and estimated value (EV) of repeated measures data, a linear mixed model was used. Age, sex, region of residence, VI, and time of measurement were used as the independent variables and xed effects. BMI, SBP, DBP, fasting blood glucose, total cholesterol, smoking, alcohol consumption, and CCI scores were used as random effects. A rst-order autoregressive model was selected as the repeated covariance type, considering the correlation of each participant's iteration. The statistical analysis model of the linear mixed model is as follows: In the subgroup analyses, we subdivided the participants according to age and sex (< 60 years old and ≥ 60 years old; men and women). According to severity, VI was divided into mild-to-moderate VI and severe VI groups.
We performed two-sided analyses and determined statistical signi cance based on P values < 0.05. SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) was used in the analyses.