Do eye diseases increase the risk of arthritis in the elderly population?

Participants

Data was obtained by The Irish Longitudinal Study on Ageing (TILDA), a longitudinal study with two waves of population-representative data from Irish participants aged 50 and over who were sampled using a geographic cluster-based RANSAM sampling system [14, 15]. The study targeted individuals aged 50 and over who lived in the Republic of Ireland in residential accommodation (excluding nursing homes and other similar institutions). Full informed consent was provided by a total of 8,504 participants, and Trinity College Dublin Research Ethics Committee granted ethical approval of this.

Between October 2009 and February 2011, participants were contacted for a Wave 1 interview [16], and Wave 2 interviews were conducted between April 2012 and January 2013 [17]. The response rate adjusted for eligibility was 62.0%. Participants were excluded for incomplete covariate data, which resulted in a sample size of 8,423 for cross-sectional analyses. In the data pre-processing of baseline, participants (n = 2589) with missing data for covariates of CHOL, LDL, HDL, and TRIG, and participants (n = 2337) were missing BMI covariate data. Considering that the elimination of large censored data of covariates can create a bias and reduce the sample size, k-nearest neighbor imputation was used to deal with the incomplete covariates [18]. Participants who reported arthritis at baseline (n = 1927) or lost to follow-up at Wave 2 (n = 1261) were excluded from longitudinal analyses, which resulted in a sample size of 5,235.

Measures

Demographic data for gender, age, smoking status, and marital status were collected in Wave 1 interviews.

Eye diseases and arthritis were investigated during Waves 1 and 2. The senior ophthalmologists and orthopedic specialists diagnosed and collected clinical parameters, including cataracts, glaucoma, other eye diseases, arthritis and its subtypes of osteoarthritis, rheumatoid arthritis, and other types of arthritis in Wave 1, and reaffirmed whether participants had arthritis in Wave 2. Other eye diseases refers to eye diseases other than cataracts and glaucoma.

Mental health and wellbeing were evaluated using scales and questions in Wave 1. The Center for Epidemiologic Studies Depression Scale [19], Hospital Anxiety and Depression Scale [20], CASP-19 Scale [21], Aging Perceptions Questionnaire [22], Penn State Worry Questionnaire [23], UCLA Loneliness Scale [24], and Computer-aided personal interview questionnaire were utilized and 109 variables were selected to characterize mental health and wellbeing.

Physical health, disability, functional impairment and helpers were assessed using a series of tests and questionnaires in Wave 1. Activities of Daily Living, Instrumental Activities of Daily Living, diseases, disability etc., were characterized by 168 variables (125 variables from the physical health aspect and 33 from the disability, functional impairment and helpers aspect).

Health measures were conducted at fixed health assessment centers in Dublin or Cork for the collection of objective measures of physical health. Respondents who were unable or unwilling to attend a health assessment center were given the option of participating in a shorter, home-based assessment. Some measurements, including visual acuity, could only be measured at the health centers. A total of 96 variables were included in the baseline, including height, weight, body mass index (BMI), waist and hip circumference, grip strength, visual acuity, blood pressure, blood lipids, and other indicators, Mini-mental state examination (MMSE), picture memory test, and visual reasoning were employed for this purpose.

Health behaviors were characterized by the International Physical Activity Questionnaire in Wave 1, which included a total of 16 derived variables.

Covariates were chosen based on theoretical, practical, and previous empirical evidence associated with arthritis [25–27]. Age, gender, smoking status, marital status, osteoporosis, depression, time and degree of physical activity, BMI, blood pressure, TRIG, LDL, HDL, and CHOL were all included.

All reported variables in these five aspects can be viewed at tilda.tcd.ie.

Statistics

Binomial logistic regression quantified the associations (i.e. odds ratios) between eye diseases (cataracts, glaucoma, and other eye diseases) and arthritis (all arthritis, osteoarthritis, rheumatoid arthritis, other types of arthritis) at the baseline and in Wave 2. A Hosmer-Lemeshow test was subsequently conducted and Nagelkerke R² was calculated in order to evaluate the models’ goodness of fit [28]. Likelihood ratio tests assessed the significance of the covariates [29].

Spearman correlation analyses were performed to gain further evidence of the relationship between eye diseases and arthritis. The contribution and importance of each subtype of eye diseases in relation to arthritis were quantified using mean Gini-index which was derived from a random forest algorithm (1,000 decision trees), which is a reliable machine-learning method [30].

A series of mediation models were used to reveal the potential mechanisms which mediated the effects on the outcomes [31]. As previously mentioned, there were five aspects of potential mechanisms which focused on: mental health and wellbeing; physical health; disability, functional impairment and helpers; indicators from health measures; and health behaviors. 389 derived variables were included as candidate mediators for mediation analyses where the potential mechanisms of the remaining direct effect of eye diseases and the indirect/mediated effects of potential mediators could be detected to the greatest extent. Censored data for each candidate mediator in a specific mediation model was eliminated independently in order to circumvent any potential bias and reduction to the sample size. Further details relating to measures and analyses are available upon request.

To perform stratified analyses and gender-related grouped analyses, Kruskal–Wallis one-way ANOVA test (H test, a pairwise comparison using Dwass-Steel-Critchlow-Fligner test, p-value adjustment using the Benjamini and Hochberg method), and binomial logistic regression were used for association detection and quantification. It is notable that the effect size was calculated based on different analyses to supplement the adjusted p-value for the quantification of significant difference.

Details of the software and packages that were used to conduct statistical analyses: R (version 3.6.3) was used for statistical analyses. Packages used in the study: Hosmer-Lemeshow test (ResourceSelection package); likelihood ratio test (lmtest package); Wilcoxon test, Kruskal–Wallis one-way ANOVA test, Dwass-Steel-Critchlow-Fligner test, Benjamini and Hochberg method was used for p-value adjustment, and corresponding effect sizes calculation (ggstatsplot package); Random forest algorithm (randomForest package); mediation analyses (mediation package).

Demographics of participants on the baseline

The baseline characteristics of the Irish participants aged 50 and over are shown in Table 1. All eye diseases demonstrated significant difference between the no arthritis and arthritis groups: cataracts (p<0.001, V=0.113, 95% CI: 0.092 to 0.133, small effect size); glaucoma (p=0.003, V=0.031, 95% CI: 0.006 to 0.056, small effect size); and other eye diseases (p<0.001, V= 0.144, 95% CI: 0.121 to 0.167, small effect size).

Despite the fact that all eye diseases showed statistical significance, the effect sizes were dissimilar, which indicates that cataracts were the most important, followed by other eye diseases and glaucoma, in terms of their contribution to the outcomes.

Cross-sectional and longitudinal results

Binomial logistic regression allowing association quantification (i.e. odds ratios) was performed to evaluate the effects of eye diseases on the development of arthritis.

For the cross-sectional studies, the prevalence of arthritis was 26.97% (n=2272). The crude models attested that cataracts, glaucoma, and other eye diseases had a significant association with 153.2% (OR= 2.532, 95%CI: 2.196 to 2.918, p<0.001), 56.8% (OR=1.568, 95%CI: 1.162 to 2.101, p<0.01), and 123.5% (OR=2.235, 95%CI: 1.979 to 2.522, p<0.001) higher odds of arthritis, respectively. Following adjustments made for a series of covariables, ORs as the quantitative indicators of associations between cataracts, glaucoma, other eye diseases and arthritis, displayed fluctuations but generally remained stable (Table 2), thereby confirming the robustness of the aforementioned discovered associations. With the benefit of the refined questionnaires, association assessments were conducted between eye diseases and arthritis subtypes. As the resultant data from crude models shows in Table 2, cataracts increased the incidence of osteoarthritis, rheumatoid arthritis, and other types of arthritis by 117.0% (OR=2.170, 95%CI: 1.818 to 2.579, p<0.001), 71.2% (OR=1.712, 95%CI: 1.374 to 2.115, p<0.001), and 80.0% (OR=1.800, 95%CI: 1.818 to 2.579, p<0.001), respectively. At the same time, the developmental processes of osteoarthritis and rheumatoid arthritis were unaffected by glaucoma, which significantly augmented the risk of other types of arthritis by 128.9% (OR=2.289, 95%CI: 1.225 to 3.918, p<0.01). Other eye diseases also showed a significant association with 82.5% (OR=1.825, 95%CI: 1.560-2.129, p<0.001), 74.3% (OR=1.743, 95%CI: 1.445-2.091, p<0.001), and 88.4% (OR=1.884, 95%CI: 1.416-2.486, p<0.001) higher odds of osteoarthritis, rheumatoid arthritis, and other types of arthritis, respectively. The robustness of these associations was corroborated by adjusting a series of covariables, and the associations of reporting significant differences remained of statistical significance (Table 2).

For the longitudinal studies, the prevalence of probable arthritis was 9.59% (n=502). The current crude models suggested that cataracts, glaucoma, and other eye diseases increased the risk of arthritis after two years by 131.8% (OR=2.318, 95%CI: 1.748 to 3.038, p<0.001), 123.1% (OR=2.231, 95%CI: 1.306 to 3.626, p<0.001), and 91.1% (OR=1.911, 95%CI: 1.501 to 2.415, p<0.001), respectively (Table 2). As anticipated, these associations remained significant statistically when the crude models were adjusted for covariables.

In addition, following prudent consideration of the results of the goodness-of-fit test and the likelihood ratio test (Supplementary Table 1), model 1, model 2, and model 3 were deemed to be different during sensitivity analyses, which thereby validated the robustness of the detected associations.

Contributions of eye diseases to arthritis

Correlations between each eye disease and arthritis, including all subtypes, were reaffirmed in Waves 1 and 2. The resultant data from the correlation analyses matched the derived results from the aforementioned binomial logistic regression (Figure 1A). Baseline analyses conducted using Spearman coefficient confirmed that all eye disease types had a significant correlation with arthritis (all p<0.01). For analyses of arthritis subtypes, cataracts and other eye diseases showed a significant association with all arthritis subtypes (all p<0.01), and glaucoma only affected other types of arthritis (p<0.01). Wave 2 analyses for correlations further corroborated that all eye diseases had a significant association to a higher probability of arthritis.

In order to quantify each eye disease’s contribution to arthritis incidence, mean Gini-index derived from random forest algorithm was used. As the resultant data in Figure 1B shows, cataracts were the most valuable impactor on the developmental process of arthritis in Waves 1 and 2, followed by other eye diseases and glaucoma (mean Gini-index of cataracts, glaucoma, and other eye diseases: 39.29, 1.79, 36.74 in Wave 1; 5.20, 2.11, 1.31 in Wave 2). For baseline analyses aimed at arthritis subtypes, cataracts were discovered to be the most significant contributor to osteoarthritis development (mean Gini-index = 12.58), while other eye diseases made a much higher contribution to rheumatoid arthritis (mean Gini-index = 4.22) and other types of arthritis (mean Gini-index = 0.89) than cataracts and glaucoma.

Underlying mechanisms mediating the increased risk of arthritis in Wave 2

Among all the eye diseases, the study has proven that cataracts have the most significant impact on increased arthritis risk. Therefore, further investigation of the underlying mechanisms between cataracts and arthritis was conducted. As previously mentioned, in order to extensively reveal these underlying mechanisms, 389 variables from five aspects characterizing elderly adults’ health were investigated using mediation analyses for the detection of potential mediators. This method was based on the previous study [32], and the five aspects of the study were: mental health and wellbeing; physical health; disability, functional impairment and helpers; indicators from health measures; and health behaviors.

In the mental health and wellbeing aspect, 29 variables significantly mediated associations between cataracts and arthritis from 109 derived variables, and the top 10 mediators ranked by proportion of mediation are shown in Figure 2A. Most significant variables were derived from the CASP-19 scale and Aging Perceptions questionnaire, showing that worse control domain of life quality, consequences, and control negative sub-dimensions of aging perceptions were the remarkable mediators for the relationship between cataracts and arthritis.

With mediation analyses for the physical health aspect, 33 variables were shown to be significant mediators from 125 variables, and the top 10 mediators are displayed in Figure 2B. Steadiness-related variables impacted the aforementioned relationships most significantly (proportion of mediation: 0.222, 95% CI: 0.139 to 0.329, of steadiness of walking), followed by memory issues, teeth missing, and long-term health problems, illness, disability or infirmity. It is notable that cataract surgery’s potential effect on arthritis risk was also tested and the resultant data suggested that there are no mediation effects (p = 0.630). Further analyses aimed at the detection of potential differences of arthritis prevalence in cataract participants also failed to achieve statistical significance (with arthritis vs. without arthritis: 19% vs. 17%, p = 0.623).

In the aspect of disability, functional impairment and helpers, 22 variables demonstrated significant effects on the outcomes from the 33 candidate mediators, and the top 10 mediators are shown in Figure 2C. Greater physical limitations was the main mediator of associations between cataracts and arthritis (proportion of mediation: 0.281, 95% CI: 0.194 to 0.395), followed by long-term illness (proportion of mediation: 0.149, 95% CI: 0.086 to 0.222).

For the health measures aspect, 31 indicators corroborated the significant mediators affecting outcomes from 96 variables, and Figure 2D displays the top 10 mediators. Functional mobility impairment was shown to have the most significant impact on outcomes (proportion of mediation: 0.214, 95% CI: 0.125 to 0.398, of speed of timed up and go test), followed by indicators pertaining to memory and muscle strength (grip strength of dominant and non-dominant hand), and glycosylated hemoglobin (HbA1c), CHOL.

In aspect of the health behaviors, only two mediators were statistically significant, from the 16 included candidate mediators (Figure 2E). The resultant data showed that sleep quality, including insomnia and poor spirit during the daytime, mediated the associations between cataracts and arthritis.

Gender-related differences

The present study further investigated whether gender difference had an effect on the risk of arthritis in Wave 2. When only subjects suffering from cataracts, glaucoma, or other eye diseases were included in the analyses, none of these diseases were reported to be of any significant statistical difference between males and females in terms of arthritis prevalence in Wave 2 (Supplementary Figure 1).

Role of the funder/sponsor

The funding sources played no role in the designing or conducting of this study.

Availability of data and materials

All data which was generated or analyzed during this study is included in this published article.