2.1. Study Participants
We used data from the National Health Insurance Service-National Health Screening Cohort (NHIS-HEALS) of the Republic of Korea, which has been constructed by the NHIS to enhance public research. The study protocol and detailed information of the NHIS-HEALS has been published elsewhere [17
]. In brief, a random sample of the participants in the National Health Screening Program, which is free and invites all residents aged ≥40 years in the Republic of Korea at least every 2 years, constitutes this cohort (n
= 514,866). Information on sociodemographic factors, lifestyle, past medical history, physical examinations, clinical laboratory test results, and mortality are included in the NHIS-HEALS.
In the present study, we used data of 362,285 individuals who participated in the National Health Screening Program in 2009 or 2010 because the health screening program was reorganized in 2009 and the number of survey items increased from 33 to 47 in that year. We further excluded individuals with physical (n = 687), auditory (n = 463), language (n = 209), intellectual (n = 117), and mental (n = 93) disabilities, brain lesions (n = 328), and other types of disability (including developmental, kidney, heart, and liver disabilities) (n = 404), leading to a final sample size of 359,984 individuals. We used information on disabilities from the National Handicapped Registry, which was provided to researchers by the NHIS-HEALS. The final study population comprised 359,523 individuals with no disabilities and 461 with visual impairment.
The Ethics Review Board of Seoul National University Hospital reviewed and approved the protocol of this study (E-1804-045-936). We conducted the present study using de-identified data (NHIS-HEALS) released to researchers for the purpose of public research.
2.5. Assumed Causal Pathway and Adjusted Covariates
We assumed a causal pathway for the association between visual impairment and mortality as depicted in Figure 1
, based on previous studies [5
], to assess the independent association by identifying potential confounders and mediators. In brief, we selected older age, systematic diseases and conditions (body mass index, waist circumference, systolic and diastolic blood pressure, serum levels of fasting glucose, creatinine, aspartate aminotransferase, alanine aminotransferase, and gamma glutamyltransferase, and history of stroke, heart disease, hypertension, type 2 diabetes, dyslipidemia, and other diseases including cancer), income, residing area, and lifestyle factors (smoking status and alcohol consumption) as potential confounders, which could affect both visual function and mortality and possibly bias the results. After identifying the potential confounders, we adjusted them in further analyses. We did not adjust for physical activity in the main models because physical activity may mediate the association between visual function and mortality partially (Figure 1
] and the association between visual impairment and mortality could be underestimated when adjusting for physical activity. Detailed information on selected covariates is presented in the Supplementary Material
2.6. Statistical Analysis
After visually checking that survival curves did not cross by visual function (no visual impairment, mild visual impairment, and severe visual impairment) using the Kaplan–Meier curves (Supplemental Material Figure S1
), we constructed Cox proportional hazard models adjusted for potential confounders to evaluate the association between visual impairment and all-cause mortality. We also constructed cause-specific proportional hazard models adjusted for the same covariates to assess the associations between visual impairment and cause-specific mortality (mortality due to cardiovascular disease, cancer, or other diseases). We estimated p
-values for trend by treating the variable for visual impairment as a continuous variable, not as an ordinal one. We also conducted the same analyses further adjusted for physical activity (none, 1–2, 3–4, or ≥5 times/week).
We then conducted stratified analysis using the same Cox models by the household income levels (0–3 deciles as lower income level, 4–7 as middle income level, and 8–10 as higher income level), sex, and body mass index (≥23 kg/m2 and <23 kg/m2) to identify vulnerable groups for decreased visual function.
We investigated whether physical activity would modify the association between visual impairment and mortality. It has been argued that interaction on an additive scale, rather than that on a multiplicative scale, is more suitable for evaluating biological interactions [23
]. Therefore, we estimated a measure for additive interaction (i.e., relative excess risk due to interaction (RERI)). If the risk of certain outcome for exposure X1
is A and the risk of outcome for exposure X2
is B, then the RERI is greater than zero in case the risk of outcome noted as C is greater than A + B when simultaneously exposed to X1
(RERI = C − (A + B)).
To conduct this analysis, we re-categorized physical activity as no or yes (≥once/week) and assigned yes (≥once/week) as a reference because it has been reported that preventive factors (e.g., physical activity) should be recoded as risk factors to adequately estimate measures for additive interaction [24
]. We also re-categorized visual impairment as no or yes (collapsed the categories for severe and mild visual impairment) and estimated the RERI between visual impairment (no or yes) and physical activity (no or yes) using logistic regression models adjusted for the same covariates. We assessed the null hypothesis that RERI = 0 using the Z-test and estimated the confidence interval (CI) and p
]. Because a RERI score >0 denotes greater risk than the sum of each main effect in the case when exposures occur simultaneously, in the present study, a RERI score >0 can be interpreted as higher mortality due to interaction compared to the additive effects of both visual impairment and no physical activity. We assessed the association between visual impairment and mortality in each stratum stratified by physical activity (none and ≥once/week).
In sensitivity analyses, we performed analyses after excluding individuals diagnosed with vision-threatening conditions such as detachment of retinal pigment epithelium (H33.0, H33.1, H33.2, H33.3, H33.4, and H33.5), central retinal artery occlusion (H34.1), degeneration of the macula and posterior pole (H35.3), optic neuropathy (H46), visual disturbances (H53.0, H53.2, H53.2, H53.3, H53.4, H53.5, H53.6, H53.8, and H53.9), and visual impairment (H54.0, H54.1, H54.2, H54.3, H54.4, H54.6, and H54.9) between 2002 and 2010 (n = 25,782) among those classified in the no visual impairment group. Finally, we conducted analyses further including individuals with physical, auditory, language, intellectual, and mental disabilities, brain lesions, and other types of disability classified in the no visual impairment group.
We performed the analyses using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA) and R version 3.5.2 (The Comprehensive R Archive Network, Vienna, Austria; http://cran.r-project.org