Do You See What Eye See? Measurement, Correlates, and Functional Associations of Objective and Self-Reported Vision Impairment in Aging South Africans

Abstract

Objectives:

Our study investigates measurement, correlates, and functional associations of vision impairment (VI) in an aging population in rural South Africa.

Methods:

1582 participants aged 40–69 reported on near (NVI) and distance vision impairment (DVI) and completed objective vision tests. Logistic and linear regression were used to evaluate sociodemographic, health, and psychosocial correlates of VI and assess relationships between VI and cognitive and physical function.

Results:

VI prevalence was considerably higher according to objective testing (56%) versus self-reports (18%). Older adults were especially likely to underreport impairment. Objective VI was associated with age, education, cardiometabolic disease, and female sex. Conversely, self-reported VI was associated with psychosocial factors. Objective NVI and both types of DVI were associated with worse visual cognition and slower gait speed, respectively.

Discussion:

Self-reported and objective VI measures should not be used interchangeably in this context. Our findings highlight extensive burden of untreated VI in this region.

Introduction

Vision impairment is a highly prevalent condition affecting at least 2.2 billion people worldwide (Fricke et al., 2018; World Health Organization, 2019). Largely a consequence of population aging, global prevalence of vision impairment (VI) is increasing even as preventative and therapeutic solutions have become more effective, affordable, and accessible (Bourne et al., 2017). Sub-Saharan Africa has the highest age-standardized prevalence of VI worldwide, suggesting that the region may experience the world’s greatest burden of VI as its population continues to age (Naidoo et al., 2020). Quality population-based data are urgently needed to assess prevalence and risk factors of VI in sub-Saharan Africa and prepare for immediate and downstream effects of VI in aging communities throughout the region. Our study examines self-reported and objective measures of near vision impairment (NVI) and distance vision impairment (DVI) in a population-based cohort in rural South Africa, administered as part of Health and Aging in Africa: A Longitudinal Study of an INDEPTH Community in South Africa (HAALSI).

Causes and Consequences of Vision Impairment in Adulthood

Vision impairment can drastically hinder autonomy and quality of life throughout the lifespan by limiting one’s ability to interact with and interpret the environment, communicate effectively, and perform essential activities of daily living, such as, reading, taking medications, cooking, driving, and dressing (National Academies of Sciences & Medicine, 2017). Unsurprisingly, VI has been linked to poor health outcomes in older age, including mortality (Bourne et al., 2017; Liljas et al., 2016; Varadaraj et al., 2019; Zhang et al., 2020). The association between VI and mortality might be mediated through effects of VI on several functional domains, including cognitive (Brenowitz et al., 2018; Maharani et al., 2018; Schubert et al., 2017), physical (Bouscaren et al., 2019; Klein et al., 2003), and psychosocial (Coyle et al., 2017; Liljas et al., 2016). VI has been shown to predict incident cognitive impairment and declines in physical function (Bouscaren et al., 2019; Laforge et al., 1992; Maharani et al., 2018, 2019), which in turn increase risks for falls, hospitalizations, and other accidents that contribute to mortality rates (Christ et al., 2014; Ehrlich, Hassan et al., 2019a; Lopez et al., 2011; Zhang et al., 2020). VI has also been associated with social isolation, depressive symptoms, and reduced well-being (Cosh et al., 2018; Coyle et al., 2017; Xiang et al., 2019), which are independently predictive of mortality (Schulz et al., 2000; Steptoe et al., 2013). Associations between VI and mortality may also be partially explained by shared underlying causes, such as cardiovascular disease and poor health behaviors (Lee et al., 2002).

NVI and DVI have distinct etiologies and may therefore have different sociodemographic and functional correlates. NVI is most commonly caused by presbyopia, a ubiquitous age-related decline in lens elasticity that makes it more difficult to focus on close objects (Fricke et al., 2018; Holden et al., 2008, 2015). In contrast, DVI is most often caused by refractive error, or myopia, which usually begins in childhood or adolescence. Onset of DVI in adulthood may indicate eye damage from injury or infection, or age-related eye disease, such as macular degeneration, diabetic retinopathy, cataracts, and glaucoma which are also associated with cardiometabolic disease (Ehrlich, Stagg et al., 2019; World Health Organization, 2019). Presbyopia and myopia are easily corrected with prescription lenses, and up to 90% of uncorrected VI is experienced by individuals in low and middle income countries (LMICs) (Ehrlich, Stagg et al., 2019).

Measurement of Vision Impairment in Population-Based Studies

There are considerable methodological differences in how VI is defined and measured across studies and populations (He et al., 2012). Many population-based studies have utilized self-reported vision difficulty as a proxy for VI to avoid the logistical challenges of administering visual acuity tests during field surveys (Zambelli-Weiner & Friedman, 2012). However, self-reported vision measures, like all measures of self-reported health, are limited by potential discrepancies between an individual’s perception regarding vision ability and his/her objective performance. Most studies on age-related vision loss have focused on a single type of vision impairment (NVI vs. DVI) and/or a single measurement method (self-report vs. objective) and have not attempted to compare prevalence and correlates of different measures of VI, especially outside high-income settings. It is important to understand the constellation of factors associated with different measures of VI to both optimize selection of screening tools for population-based studies and to tailor clinical interventions to address specific areas of need.

Although many researchers have used self-reported and objective VI measures interchangeably, there have been mixed results with regard to the level of concordance between measurement modalities (Coyle et al., 2017; El-Gasim et al., 2012; Pinto et al., 2014; Whillans & Nazroo, 2014). While some studies report acceptable levels of agreement between measurement types (Pinto et al., 2014; Whillans & Nazroo, 2014), others have reported socially-patterned measurement discordance (El-Gasim et al., 2012). In general, concordance between self-reported and objective VI has been lower in older, nonwhite, and lower-educated respondents (El-Gasim et al., 2012; Whillans & Nazroo, 2014). While older and black participants have mostly been shown to underestimate VI in self-reports, lower education has been associated with both patterns of measurement discordance. Varied levels of health self-awareness and health knowledge may contribute to education effects on self-report accuracy (Delpierre et al., 2009).

Measurement discordance may reflect several different factors, including unmeasured participant characteristics that influence self-reporting tendencies (Zebardast et al., 2017), as well as differences in the aspects of the visual experience assessed by self-reported versus objective vision measures (Whillans & Nazroo, 2014). Depressive symptoms, health pessimism, and low self-efficacy may affect the way different individuals experience and reflect on their own health or functional ability, even when they have objectively similar levels of health or functional performance (Cines et al., 2015; Cosentino et al., 2018; Coyle et al., 2017; El-Gasim et al., 2012; Kliegel & Zimprich, 2005). Objective and self-reported VI might also contribute to higher depressive symptoms and lower well-being (Cosh et al., 2018; Frank et al., 2019; Wettstein et al., 2021; Xiang et al., 2019), resulting in a cyclical relationship between psychosocial factors and self-rated VI. In sum, there are likely shared and unique factors associated with self-reported and objective VI that may be integral for how we interpret differences across measurement types.

Study Objectives and Hypotheses

The objectives of this study are threefold: (1) assess prevalence of objective DVI and NVI and their sociodemographic, health, and psychosocial correlates in an aging community in rural South Africa; (2) establish concordance between objective and self-reported VI measures and assess factors associated with self-report measures independent of objective VI; and (3) evaluate how different types of VI are associated with physical and cognitive function, controlling for covariates. Although our cross-sectional data do not allow us to formally test causal relationships, Figure 1 depicts a proposed conceptual model of the factors that may be associated with self-reported and objective VI. First, we hypothesize that objective vision acuity will be related to sociodemographic factors known to increase risk of VI, including older age, female sex, and lower education and wealth (National Academies of Sciences & Medicine, 2017). We further expect higher rates of objective vision acuity impairment to be associated with chronic health conditions, poor health behaviors, and rates of eye care treatment. However, an individual’s perception regarding their vision ability may reflect more complex dynamics. While self-reported vision will be modestly correlated with objective vision acuity, we also expect independent associations with psychosocial/self-report factors, reflecting bidirectional relationships between affective/well-being factors and an individual’s perceptions about global and domain-specific health. Chronic health conditions may be associated with higher rates of self-reported VI through their effects on objective vision health and global self-perceptions of health. Finally, we expect self-reported VI to be associated with demographic factors, independent of objective VI, due to differences in self-report tendencies as a function of these variables. We further hypothesize that both objective and self-reported VI will be independently associated with worse physical and cognitive function, even after controlling for shared associations with health, sociodemographic, and psychosocial mechanisms.

Figure 1.

Proposed conceptual model of the relationship between sociodemographic factors, psychosocial/self-report factors, and health factors with self-reported and objective vision. While objective vision acuity will be related to health and sociodemographic factors, self-reported vision will capture more complex dynamics, including bidirectional associations with psychosocial factors and other domains of self-rated health. Cognitive and physical function will be selectively associated with both types of vision measures.

Methods

Data and Participants

We utilize baseline data (2014–2015) from HAALSI, a population-based longitudinal study of social, biological, and economic determinants of health in rural South Africans. The HAALSI sample includes 5059 middle-aged and older adults living in the Agincourt subdistrict of Mpumalanga province, South Africa. The cohort is representative of individuals aged 40 and older and living within the study site for at least 12 months preceding the 2013 Agincourt Health and Socio-Demographic Surveillance Systems census, which covers a population of 116,000 people living in 31 villages. Details of the sampling and study procedures have previously been published (Gómez-Olivé et al., 2018); documentation and data are currently available for public download at https://haalsi.org/data. Participants provided informed consent, and study protocols were approved by institutional review boards at collaborating institutions. The current study draws on data collected during two separate visits associated with the baseline wave: (1) a home-based survey administered to the full sample, including health and demographic questionnaires, physical and cognitive function, anthropometrics, and biomarkers and (2) a laboratory evaluation administered to a subsample after the home-based survey, which involved more comprehensive testing including objective visual acuity. The laboratory sample was composed of a random sample of HAALSI cohort members <70 years of age. Response rate for the laboratory visit was 72% (2114/2948); reasons for nonparticipation included refusals (n = 485), missed appointments (n = 236), inability to participate (n = 81), death (n = 21), and out-migration (n = 11). Our sample is restricted to 1582 participants aged 40–69 who completed all self-reported and objective vision assessments. Relative to same-aged participants in the full HAALSI cohort, the analytic sample included a higher proportion of females but was qualitatively similar in terms of the distribution of other sociodemographic variables, self-reported VI and treatment, cognitive and physical function performance, and other covariates.

Measures

Vision Measures

Self-reported vision impairment.

NVI and DVI were assessed as part of the main home-based survey. Self-reported NVI was evaluated with the question “How good is your eyesight for seeing things up close, like reading ordinary newspaper print (with glasses or corrective lenses if you wear them)?” DVI was evaluated with the question “How good is your eyesight for seeing things at a distance, like recognizing a friend from across the street (with glasses or corrective lenses if you wear them)?” Each question had five response categories: excellent, very good, good, fair, and poor. For primary analyses, responses were recoded into a dichotomous variable grouping together fair and poor vision to indicate self-reported VI.

Objective vision impairment.

Adapted from the Study of Global Ageing and Adult Health (SAGE) (Kowal et al., 2012), vision acuity was tested using a computerized version of the “Tumbling E” logMAR chart (Taylor, 1978), which is appropriate for populations with lower levels of literacy (Bourne et al., 2003; Taylor, 1978). Participants were shown a series of capital letter “E”s rotated in different directions and were asked to indicate which direction the letters were facing. Rows of Es grew progressively smaller to assess increasing levels of visual acuity. Participants were asked to read the chart from 4 m away for distance vision and 40 cm away for near vision. Each eye was tested separately. If the participant missed more than 40% of Es in a trial, the test was discontinued, and the score for that eye was the smallest row that the participants read correctly. Participants were permitted to wear glasses or contact lenses if usually worn.

For each eye and distance level, results from the visual acuity test were categorized as blind, referring to visual acuity less than 3/60 (.05 decimal); low vision, referring to visual acuity between 3/60 and 6/18 (.05 and .33 decimal); and normal vision, referring to visual acuity at least 6/12 (.5 decimal). For the current analyses, we use the worst visual acuity level at each distance to indicate impairment, so if one eye had normal distance vision and the other had low distance vision, a person is coded as having low distance vision. Given low numbers of blind participants (Table 1), we collapsed low vision and blind into a single category to indicate VI.

Table 1.

Characteristics of the Sample.

Variables	n	%
Agegroup
40–49	468	29.6
50–59	660	41.7
60–69	454	28.7
Female	960	60.7
Near vision acuity test
Normal	805	50.9
Low	735	46.5
Blind	42	2.6
Self-reported near vision ability
Good to excellent	1360	86.0
Fair	192	12.1
Poor	30	1.9
Distance vision acuity test
Normal	1142	72.2
Low	407	25.7
Blind	33	2.1
Self-reported distance vision ability
Good to excellent	1428	90.3
Fair	150	9.5
Poor	4	.2
Wears glasses	140	8.8
Diagnosed with cataracts	157	9.9
Received cataract surgery	19	1.2
Education level
No formal education	530	33.6
Some primary (1–7)	638	40.4
Completed primary (8–11)	243	15.4
Completed secondary (12+)	167	10.6
Diabetes	144	9.5
Hypertension	842	53.7
Ever smoked	322	20.4
Self-rated health (good or excellent)	1155	73.0
Self-rated childhood health (good or excellent)	1395	88.2
	Mean	SD
Depressive symptoms [0–8]	1.31	1.57
Life satisfaction [0–10]	6.80	2.33
Social network size [0–7]	3.25	1.64

Correlates of Vision Impairment

Sociodemographic factors.

We examine age, sex, and education as possible correlates of self-reported and objective NVI. Education was recoded into a four-level categorical variable corresponding to no formal education, some primary (1–7 years), completed primary (8–11 years), and completed secondary (12+ years). We also examine socioeconomic status using the household wealth index divided into quintiles of wealth (Filmer & Pritchett, 2001; Riumallo-Herl et al., 2019). For analytic purposes, we define the bottom two wealth quintiles as low relative wealth.

Chronic health conditions.

We include health conditions that have been associated with age-related VI, including hypertension, diabetes, and history of smoking. All measures were collected during the main survey visit. Individuals were considered hypertensive if systolic blood pressure ≥140 and/ or diastolic blood pressure ≥90 or if the participant self-reported use of antihypertensive medications. Participants were considered diabetic if they self-reported diabetes treatment, or when blood samples were available, had blood glucose ≥7 mmol/l (126 mg/dL) if fasting, or glucose ≥11.1 mmol/l (200 mg/dL) if not fasting. Participants were coded as positive for smoking if they reported having ever smoked.

Eye conditions and vision correction.

Models control for self-reported diagnosis and treatment of cataracts (no cataracts, treated cataracts, and untreated cataracts), as well as self-reported use of glasses or corrected lenses (does not usually wear lenses and usually wears lenses), as these factors might influence the rate of endorsement of VI and performance on visual acuity tests. However, it is not known whether corrective lenses were prescribed for near or distance vision or whether lenses were worn during the test.

Psychosocial factors.

Depressive symptoms were assessed using an eight-item version of the Center for Epidemiologic Studies Depression Scale (CES-D (Eaton et al., 2004)), which was recently shown to have reasonable psychometric properties in the study population, Cronbach’s alpha = .66 and mean item-to-total correlation = .57 (Adams et al., 2020). Psychometric properties of the scale were identical in the current analytic sample (α = .66; item-to-total correlation = .57). Higher scores indicate higher endorsement of depressive symptoms (0–8). Social network size was evaluated by asking participants to name up to six adults they had been in contact with over the past six months, starting with the most important (Harling et al., 2020). If the participant was living with a spouse but did not name them during the interview, the spouse was added to the list for a total range of 0 to 7 contacts. To measure global life evaluation and self-appraisal (Kahneman & Deaton, 2010), we utilized the Cantril Self-Anchoring Ladder Scale (Cantril, 1965). Participants were asked to envision a ladder with steps numbered from 0 to 10, with 0 representing one’s worst possible life and 10 representing one’s best possible life. They were then asked to indicate where they stood at the current time, assuming that the higher the step the better they felt about their life. The 11-point self-anchoring scale has been widely used to assess global life satisfaction in a variety of populations and settings (Diener et al., 2010).

Subjective health.

Participants were asked to rate current health and health during childhood, using a five-level Likert scale from “Very Bad” to “Very Good”; responses were recoded into poor (very bad to moderate) versus good (good to very good) subjective health.

Functional Measures

Cognitive function.

We assess the impact of NVI and DVI on two measures of cognitive function. Global cognitive function (score range 0–24) was measured during the main household survey using 10-item immediate word recall, 1-minute delayed word recall, and 4-item orientation task requiring the participant to state the year, month, date, and president (Kobayashi et al., 2019). We evaluated visual cognitive function using the Oxford Cognitive Screen-Plus (OCS-Plus), a tablet-based cognitive battery with low literacy demands (Humphreys et al., 2017) administered during the laboratory assessment. Described in detail elsewhere (Farrell et al., 2020; Humphreys et al., 2017) and in the supplemental material, we created a visual cognition composite score using six subtests of the OCS-Plus with high visual acuity demands: picture naming, semantic identification, trail making, figure copy, and figure copy recall. Scores range from 0–83.

Physical function.

Gait speed was used as an indicator of physical function due to its established association with important health outcomes in older adulthood (Payne et al., 2017). Participants were asked to walk a 2.5-m course twice at their normal walking speed, and we use the average of these two walk times. Respondents who were unable to walk were not tested (n = 36). Walk times longer than 20 seconds were considered out of range and set to missing (n = 3).

Statistical Methods

We calculated Cohen’s kappa (κ) and percent agreement to assess concordance between self-reported and objective measures of VI. Objective and self-reported VI were assessed at separate study visits, and the timing interval varied across participants (mean interval = 5.6 months, SD = 2.8, range .1–13.4 months). To confirm that the time delay between visits did not affect measurement concordance, we examined whether time interval affected the strength of association between self-reported and objective VI measures and compared kappa statistics and percent agreement at increasing time intervals. We calculated basic descriptive statistics for all sociodemographic, health, and psychosocial covariates across vision groups and examined group differences using the chi-square test for categorical variables and t-tests for continuous measures. Pearson’s correlations were used to assess the strength of bivariate associations among covariates and vision measures, using continuous measures of age and years of education. Multiple logistic regression was used to identify unique predictors of objective VI and then identify which factors were associated with self-reported VI while controlling for objective VI. We report concordance (or C) statistics, equivalent to the area under the receiver operating characteristic (ROC) curve, as a measure of how well our models discriminated between those with and without VI (Steyerberg et al., 2010). By convention, C-statistics above .7 indicate reasonable/good discrimination and above .8 indicate strong/excellent discrimination (Hosmer et al., 2013). Finally, multiple linear regression was used to examine associations between VI measures and cognitive and physical function, controlling for covariates. Stata/MP 15.1 was used for statistical analyses, and significance was considered at p-values < .05.

Results

Sample Characteristics and Prevalence of Objective VI

Table 1 reports the characteristics of the sample. Overall, educational attainment was low, with almost 34% reporting no formal education. Over half of the sample was hypertensive and nearly 10% had diabetes. Most participants reported good current health and childhood health, and there was low endorsement of depressive symptoms. Wealth quintiles (derived from the parent HAALSI cohort) were equally represented, suggesting that this sample is socioeconomically representative of the larger cohort.

Figure 2 depicts the prevalence of each VI measure by decade of age, revealing markedly higher levels of VI when measured via objective testing compared to self-report. Prevalence of objective NVI increased steeply with age, with 25% aged 40–49, 54% aged 50–59, and 66% aged 60–69 testing in the impaired range on near vision tests. Objective DVI was less prevalent but still common, affecting 13% in the 40–49 group, 25% in the 50–59 group, and 47% in the 60–69 group. Corrective aids were relatively uncommon, with only 9% of the sample self-reporting corrective lens use. Diagnosis of cataracts was reported by 10% of the sample, although only 19 individuals (1%) reported having received cataract surgery.

Figure 2.

Prevalence of near vision impairment and distance vision impairment according to objective tests versus self-report.

Concordance between Self-Reported and Objective VI

We found modest correlations between self-reported and objective measures of NVI, r = .09, p < .0002, and DVI, r = .08, p < .0008. Cohen’s kappa confirmed low concordance between measurement types. For NVI, there was 53.9% agreement, κ = .07, p < .001, while for DVI, there was 70.1% agreement, κ = .07, p < .001. According to traditional benchmarks, κ ≤ .2 indicates zero to slight agreement between measures (McHugh, 2012). Agreement between self-reported and objective VI decreased with age. Percent agreement for NVI and DVI measures were 72% and 82% in the 40–49 age-group, 50% and 71% in the 50–59 age-group, and 41% and 56% in the 60–69 age-group (all κ < .1).

Low kappa statistics reflect the fact that only 13.6% of those with objective DVI, and 17.4% of those with objective NVI, acknowledged impairment in the corresponding self-report item on the questionnaire. For near vision, 88% of discordant responses were individuals reporting good near vision but testing in the impaired range on acuity tests. For distance vision, 80% of discordant responses were individuals reporting good distance vision but performing in the impaired range on acuity tests.

The strength of association between measurement types was virtually unchanged when we controlled for the time interval between study visits, r = .09, p < .001 for NVI and r = .08, p < .001 for DVI, suggesting that varying duration between assessments did not affect concordance between objective and self-reported measures. We also plotted κ-statistics and percent agreement between self-reported and objective VI for participants who were assessed within 3 months (15%), 3–6 months (46%), 6–9 months (25%), and ≥9 months (13%) and found no evidence for systematic decline in measurement concordance with increasing length of time between visits (see Supplemental Figure 1).

Correlates of Objective and Self-Reported VI

Summary statistics of sociodemographic, health, and psychosocial variables by the vision status group are described in supplemental materials and shown in Supplemental Tables 1 and 2. Correlations among all variables are shown in Supplemental Table 3. A few important patterns emerge from the bivariate analyses. Objective NVI was associated with older age, female sex, lower education, use of glasses, diagnosis of cataracts, hypertension, diabetes, higher depressive symptoms, poor self-rated health, smaller social networks, and lower life satisfaction. Objective DVI had similar correlates, except it was not related to use of glasses or life satisfaction and was significantly related to history of smoking. Relative to the objective VI measures, self-reported VI measures had similar bivariate correlates although there were differences in the magnitude of effects. Specifically, there were stronger correlations between age, education, and chronic health conditions with objective VI and stronger correlations between cataract diagnosis, depressive symptoms, self-rated health, and life satisfaction with subjective VI. It is also worth noting that use of glasses and diagnosis of cataracts were associated with socioeconomic factors, including education and wealth.

Results from multivariate logistic regression confirmed differences in predictors of self-reported versus objective NVI (Table 2). The strongest predictor of objective NVI was older age, with individuals aged 50–59 and 60–69 experiencing 2.9- and 4.8-times higher odds of NVI compared to the youngest age-group, respectively. Female sex, corrective lens use, and untreated cataracts were also associated with increased odds of NVI, while highest level of education and those with larger social networks were less likely to have objective NVI. Controlling for objective NVI, self-reported NVI was associated with female sex and diagnosis of cataracts, regardless of treatment. Higher depressive symptoms and larger social networks increased odds of self-reported NVI, while those with higher life satisfaction and good self-reported health were less likely to report NVI.

Table 2.

Logistic Regression Analyses Predicting Self-Reported and Objective Vision Impairment.

	Objective near vision impairment		Self-reported near vision impairment
Variables	Odds ratio	95% CI	Odds ratio	95% CI
Objective NVI			1.33	.95, 1.87
Age-group (ref: 40–49)
50–59	2.94**	2.21, 3.92	1.22	.79, 1.91
60–69	4.77**	3.41, 6.69	1.22	.73, 2.02
Female	1.57**	1.18, 2.10	1.84**	1.17, 2.89
Education level (ref: no formal)
Primary	1.04	.80, 1.34	1.21	.84, 1.74
Some secondary	.78	.54, 1.12	.85	.49, 1.48
Secondary +	.48**	.30, .77	.36*	.15, .86
Lower assets	1.10	.87, 1.40	.99	.70, 1.40
Wears glasses	1.92**	1.19, 3.08	1.63	.93, 2.84
Cataracts (ref: no cataracts)
Untreated cataracts	1.75*	1.10, 2.78	3.02**	1.83, 4.96
Treated cataracts	.62	.21,1.80	5.17**	1.63, 16.47
Hypertension	1.12	.89, 1.40	1.15	.82, 1.61
Diabetes	1.40	.95, 2.08	1.25	.76, 2.07
Ever smoked	1.17	.83, 1.65	1.19	.69, 2.05
Depressive symptoms	.99	.93, 1.07	1.12*	1.03, 1.23
Life satisfaction	.98	.93, 1.03	.88**	.82, .95
Social network size	.90**	.84, .96	1.16**	1.05, 1.28
Self-rated health	.90	.70, 1.17	.64*	.46, .90
Self-rated childhood health	.95	.67, 1.34	.39**	.26, .60
C-statistic	.71		.76
	Objective distance vision impairment		Self-reported distance vision impairment
	Odds ratio	95% CI	Odds ratio	95% CI
Objective DVI			1.22	.81, 1.84
Age-group (ref: 40–49)
50–59	1.81**	1.27, 2.57	1.03	.62, 1.73
60–69	4.03**	2.76, 5.89	1.06	.60, 1.88
Female	.76	.56, 1.04	1.53	.91, 2.59
Education level (ref: no formal)
Primary	.71*	.54, .94	1.24	.80, 1.91
Some secondary	.46**	.30, .71	.86	.44, 1.67
Secondary +	.52*	.30, .90	.50	.20, 1.24
Lower assets	1.12	.86, 1.46	1.05	.70, 1.59
Wears glasses	1.07	.66, 1.74	1.63	.89, 2.98
Cataracts (ref: no cataracts)
Untreated cataracts	1.36	.86, 2.16	4.12**	2.43, 7.00
Treated cataracts	.93	.31, 2.79	8.66**	2.72, 27.53
Hypertension	1.53**	1.18, 1.97	1.37	.91, 2.04
Diabetes	1.58*	1.07, 2.33	1.53	.88, 2.66
Ever smoked	1.18	.82, 1.69	1.58	.86, 2.88
Depressive symptoms	1.05	.97, 1.14	1.23**	1.11, 1.36
Life satisfaction	.99	.94, 1.05	1.02	.94, 1.11
Social network size	.94	.87, 1.02	1.08	.96, 1.21
Self-rated health	.89	.67, 1.18	.49**	.33, .73
Self-rated childhood health	.88	.61, 1.28	47**	.29, .76
C-statistic	.72		.78

Notes.

^*p < .05

^**p < .01.

Odds of objective DVI also increased with age, with adults in their 50s and 60s experiencing 1.8- and 4.0-times higher odds of DVI compared to adults in their 40s. Risk of DVI decreased with each additional level of education, and both diabetes and hypertension were associated with elevated odds of DVI. Conversely, self-reported DVI was associated with treated and untreated cataracts, higher depressive symptoms, and worse subjective health both now and during childhood. The full models of objective NVI and DVI performed well in terms of discrimination, with C-statistics of .71 and .72, respectively. C-statistics were higher for self-reported NVI (.76) and DVI (.78). This suggests that the selected covariates adequately identified those at high risk for both types of vision impairment but were better at discriminating high risk of self-reported VI.

Associations of VI with Cognitive and Physical Function

Supplemental Table 4 reports descriptive data for global cognitive function, visual cognitive function, and gait speed as a function of VI status. In bivariate comparisons, worse global cognitive function and slower walk speed were associated with each type of VI; worse visual cognition was associated with objective NVI, objective DVI, and self-reported NVI.

Coefficients for the effects of VI measures on functional outcomes are shown in Table 3. Fully adjusted models of cognitive function and walk speed are shown in Supplemental Table 5. When vision measures were included in the model without covariates, self-reported NVI, objective NVI, and objective DVI were associated with worse global cognitive function. However, only effects of self-reported NVI remained after adjustment for covariates, associated with a .66-point difference in global cognitive function. In models including only vision variables, visual cognitive function was associated with both objective vision measures. In the fully adjusted model, objective NVI was the only significant predictor of visual cognition, associated with a 3.5-point decrement performance. On the other hand, walk speed was associated with both DVI measures. Controlling for all covariates, objective and self-reported DVI were associated with .35 and .41 second slower walk times, respectively.

Table 3.

Multiple Linear Regression Analyses Predicting Cognitive and Physical Function.

	Global cognitive function	Visual cognitive function	Walk speed
Range	0–24	0–83	2.5–18
Mean (SD)	12.46 (4.02)	43.69(22.07)	6.15 (1.83)
Vision variables only	B (SE)	B(SE)	B(SE)
Objective near vision impairment	−.51(.21)*	−7.52(1.13)**	.10(.10)
Self-reported near vision impairment	−.82(.31)**	−.01(1.69)	.14(.14)
Objective distance vision impairment	−1.05(.23)**	−5.75(1.27)**	.48(.11)**
Self-reported distance vision impairment	−.57(.37)	−2.43(1.99)	.62(.17)**
N	1574	1579	1543
R 2	.03	.06	.03
Fully adjusted model	B (SE)	B(SE)	B(SE)
Objective near vision impairment	−.04(.21)	−3.5(1.03)**	−.05(.10)
Self-reported near vision impairment	−.66(30)	1.74(1.51)	.06(.15)
Objective distance vision impairment	−.29(.23)	−1.49(1.13)	.35(.11)**
Self-reported distance vision impairment	−.54(.35)	−2.91(1.76)	.41(.17)*
N	1488	1489	1471
R 2	.20	.34	.10

Notes. Fully adjusted models control for all demographic, health, and psychosocial variables.

^*p < .05

^**p < .01.

Discussion

Our study adds valuable population-based data on the burden of VI and its association with cognitive and physical function in a rural community in South Africa, which may be representative of other rural communities in the region. In our sample, 56% of adults aged 40–69 had some type of objective VI, including 49% with NVI, 28% with DVI, and 20% with impairment in both near and distance vision. These estimates are far higher than prevalence estimates from any high-income country reported in a recent meta-analysis (Bourne et al., 2017), which may be attributable to limitations in preventative and therapeutic treatments for vision loss. We also unveil several novel findings regarding measurement and correlates of age-related vision loss, which may have important implications for future studies of VI in LMICs. First, we reveal markedly low concordance between objective and self-reported VI measures. Older adults were especially likely to underestimate VI in the self-report measure, which may be due to intrinsic participant characteristics or differences in the constructs being assessed by these measures. In line with hypotheses, objective and self-reported VI measures have distinct sets of correlates, shedding light on the sociodemographic, health, and psychosocial factors associated with different types of VI. Finally, we find independent associations between specific vision measures and cognitive and physical function even after controlling for a variety of covariates, underscoring the important link between sensory and functional ability.

Prevalence and Correlates of Objective VI

In our sample, over 2% of participants were coded as blind and 26% were coded as impaired on distance vision tests. Although direct comparison remains difficult due to variability in measurement methods, our data align with prevalence estimates for sub-Saharan African adults aged 50 and older, where over 25% were shown to have some degree of DVI in 2015 (Naidoo et al., 2020). A somewhat surprising finding was the significant age-related increase in DVI, given that many previous studies have reported no increase or even decreasing levels of myopia throughout adulthood (Foster & Jiang, 2014). Although our data do not allow us to determine specific etiology of VI, the significant effects of age, hypertension, and diabetes on likelihood of DVI suggest that a considerable portion of DVI may be attributable to cardiovascular disease in this cohort.

NVI was even more prevalent in our cohort, providing further evidence for uncorrected presbyopia as the leading cause of VI worldwide (Holden et al., 2015). Levels of NVI ranged from 25% in the youngest age-group to 66% in the oldest age-group, estimates that are somewhat lower than reported rates in other parts of South Africa where over 80% of adults older than 35 had uncorrected presbyopia (He et al., 2012). Part of this difference may be due to our coding of individuals with 20/40 vision as “normal vision,” whereas other studies have classified 20/40 near vision as indicative of low vision. Hence, it is possible that some portion of individuals in our cohort with categorically normal vision were experiencing the first stages of declining near vision acuity.

Population-based studies have historically focused on distance vision due to assumptions that NVI is both easily correctable and less detrimental to function than DVI (Tahhan et al., n. d.). There may be further speculation that near vision function is less essential to well-being in LMICs, where literacy rates are low and occupational demands are less tied to reading and computer work. To the contrary, our data show that most adults in our cohort with need for near vision correction did not own corrective lenses and that both subjective and objective NVI were strongly linked to well-being and health, including measures of social engagement, cognitive function, depressive symptoms, self-appraisal, and perceived health. These findings affirm previous studies documenting the effect of NVI on quality of life and productivity in other LMICs, including rural Africa (Frick et al., 2015; Patel et al., 2006).

A rather striking finding from our study was low level of vision correction, with less than 10% reporting regular use of glasses or corrective lenses, and only 10% of participants with diagnosed cataracts reporting surgical correction. Surprisingly, glasses wearers in our sample performed worse on NV acuity and no better on DV acuity, suggesting inadequate vision correction even among those who have lenses. Unmet need for vision care in rural communities, including those in South Africa, was previously highlighted in a multinational study where less than 10% of adults with need for refractive correction had necessary spectacles (He et al., 2012). Globally, most causes of VI could be prevented or corrected with simple ophthalmologic treatments (Bourne et al., 2013). Our data emphasize the importance of implementing large-scale vision correction programs to improve well-being in aging communities in South Africa and other LMICs (He et al., 2012).

Concordance between Objective and Self-reported VI

Another reason for insufficient vision correction may be low awareness of VI according to self-report. Prevalence of self-reported NVI (14%) and DVI (9.7%) was much lower than estimates generated by the corresponding objective vision tests. This was also reflected in the pattern of discordant responses, where most cases of disagreement between self-reported and objective VI measures were individuals reporting good vision but testing in the impaired range. While some studies have also found underestimation of VI when utilizing self-report versus objective vision measures (Varadaraj et al., 2019), others have reported the opposite pattern (Coyle et al., 2017; Whillans & Nazroo, 2014). It is important to note that prior studies reporting higher concordance and/or overestimation of VI in the self-report were conducted in higher-income settings where rates of objective VI were much lower, utilization of eye care was much higher, and participants had higher levels of literacy and education. These factors may make it easier to detect and accurately report VI. In our study, the tendency to deny or underestimate VI was greater for NVI vs. DVI and increased with age. Other studies have shown that older adults tend to underreport health problems, including vision, hearing, and cognitive difficulties (Ikeda et al., 2009; Whillans & Nazroo, 2014). There are several possible interpretations of this pattern. One explanation is that older adults lack awareness of problems due to reduced cognitive capacity and/or less demanding daily activities which mask underlying problems. Alternatively, older people may normalize VI because of inherent expectations that vision loss is a typical part of aging. Social desirability may encourage older adults to deny impairment to be perceived as healthier or younger by the interviewer. Finally, older adults may use downward social comparison when reporting on their vision ability, selectively comparing themselves to same-aged peers who are worse off as a means of maintaining a positive outlook on their own health (Robinson-Whelen & Kiecolt-Glaser, 1997; Whillans & Nazroo, 2014). Future research should continue to examine participant and methodological factors that influence accuracy and interpretation of self-reported health variables deployed in population-based surveys.

Correlates of Self-reported VI

Clear differences in correlates of objective versus self-reported VI suggest that these measures capture different aspects of vision impairment. In agreement with our conceptual model, self-reported vision was strongly associated with psychosocial factors and other self-report domains, including depressive symptoms, social engagement, self-rated health, and global life evaluation. Our cross-sectional data do not allow us to disentangle temporal order, and indeed, many of these associations could be bidirectional. Age-related vision impairment and associated loss of function could lead to higher rates of depressive symptoms and reduced well-being (Cosh et al., 2018; Frank et al., 2019); conversely, depressive symptoms may also lead to higher rates of self-reported VI (Frank et al., 2019) by increasing risk factors for VI (e.g., health-seeking behaviors and health conditions), and/or increasing negative self-reflection leading to higher rates of symptom endorsement.

Neither measure of objective VI was a significant predictor of its corresponding self-report measure in multivariate models. This pattern has been shown in other health domains, such as cognitive function, wherein self-report measures are more strongly associated with affective factors than they are with objective measures of the same construct (Cosentino et al., 2018). Others have shown clustering of self-reported health measures across domains, suggesting that there may be a common thread underlying self-report tendencies that is independent of participants’ actual physical health, and may be partially explained by affective factors and personal belief systems (Cosentino et al., 2018; Luszcz et al., 2015).

An unexpected finding was the positive association between social network size and self-reported NVI. Prior research suggests that vision loss may result in gradual disengagement from social life in older age (Alma et al., 2011; Coyle et al., 2017; Ehrlich, Stagg et al., 2019; Liljas et al., 2016), likely due to a combination of reduced mobility, communication disruptions, and fear of navigating outside the home. One possibility for our contradictory findings is that those with greater perceived need for care might accumulate a larger network of social contacts whose primary function is to provide physical or functional support. A final factor shown to influence rates of self-reported DVI and NVI was history of vision care, as indicated by diagnosis of cataracts (regardless of surgical correction) and/or use of glasses. This may reflect more proactive health-seeking behaviors in individuals with greater awareness of vision difficulties or the fact that VI is more likely to be endorsed when an individual has received an official diagnosis from a medical professional. Overall, our data indicate that self-reported vision measures capture more complex dynamics than objective vision acuity tests, including global perceptions about health, affective factors, and receipt of eye care.

Associations of Vision Impairment with Cognitive and Physical Function

DVI and NVI were differentially associated with important aspects of cognitive and physical function. Strong cross-sectional associations between VI and cognitive performance are well-documented (Chen et al., 2017). Surprisingly, effects of most VI measures on global cognition were fully attenuated after adjusting for demographic, health, and psychosocial factors; only self-reported NVI was associated with slightly worse cognitive performance. It is important to note that our global cognition measure was composed entirely of auditory-based tasks, so VI should not directly impact one’s ability to perform the tasks. In contrast, the visual-based cognitive function composite, which included tasks relying heavily on visual perception and psychomotor control, was strongly associated with objective NVI even after controlling for covariates. In fact, objective NVI was associated with a drop in visual cognitive performance similar in magnitude to the effect of a decade of age. Given that objective NVI was selectively associated with cognitive abilities requiring visual perception, but not with verbal or auditory tasks, our data suggest that observed associations between VI and cognitive function may be partially attributed to degraded quality of sensory input during cognitive task performance (Glass, 2007). Additional waves of data are needed to determine whether high rates of VI, and prolonged reductions in sensory input, may contribute to more global cognitive impairments and the pace of age-related cognitive decline in this cohort.

By contrast, our measure of physical function, gait speed, was associated with both objective and self-reported DVI. Effects of DVI measures on walk speed were comparable in magnitude to the difference in walk speed for participants in their 60s relative to participants in their 40s (see Supplemental Table 5). It is not surprising that impaired vision ability may hinder an individual’s walk speed (Helbostad et al., 2009). However, associations between self-reported DVI and gait speed suggest that one’s self-perception about visual ability may also be related to physical function in older age. Individuals with low visual function, and more importantly, individuals with poor perceived visual function, might have low self-efficacy when attempting physical tasks such as a timed walk and therefore perform these tasks with reduced confidence. Alternatively, associations between DVI and walk speed could be explained by shared underlying causes, such as chronic health conditions, aging, and poor diet (Lee et al., 2002). However, our models of physical function controlled for effects of age and common cardiometabolic diseases, suggesting that exposure to these conditions cannot fully explain associations between DVI and gait speed observed here. Overall, our data highlight important associations between VI and cognitive and physical performance in middle and older age and demonstrate how the magnitude of effects are determined by the specific demands of the task.

Limitations and Strengths

Our study, which utilizes data from a unique cohort in rural South Africa, has several strengths as well as some limitations. Strengths of this study include the use of a well-defined cohort in an understudied region, as well as its use of multiple measures of VI and functional outcomes to better understand the impact of VI on the community. The primary weakness lies in the cross-sectional nature of the data, which precludes formal investigation of the causal pathways underlying associations between health and psychosocial factors, objective and self-reported VI, and functional outcomes. However, we have provided a conceptual model that describes the most plausible connections between these constructs which can be empirically tested with future longitudinal waves. Broad measures of visual acuity make it difficult to determine the etiology of vision loss and may not capture all aspects of vision decline that are assessed in self-report. We have limited information on eye care utilization, efficacy of corrective treatments, and reasons for undertreatment (e.g., lack of awareness of deficit, limited eye care providers, and knowledge about available services). Future research should include detailed ophthalmological examinations to determine the major causes of VI in this setting, which would facilitate understanding of timing of onset as well as the proportion of VI that may be correctable with straightforward interventions.

Conclusions

Near and distance vision ability were associated with health and well-being in our rural South African cohort. An important conclusion from our study is that self-reported VI is not a reliable indicator of objective VI and should not be used interchangeably in this population. Studies utilizing self-report may underestimate objective vision acuity impairments but may capture other unmeasured factors and/or different aspects of the vision loss experience. Given the high prevalence and low correction of VI observed here, more needs to be done to mitigate increasing burden of a consequential, yet highly treatable, health condition affecting aging populations worldwide.