Abstract
Purpose.:
To determine if the odds of mobility disability increases at a different rate among visually impaired (VI) as compared with nonvisually impaired (NVI) over an 8-year period.
Methods.:
A total of 2520 Salisbury Eye Evaluation Study participants were followed 2, 6, and 8 years after baseline. VI was defined as best-corrected visual acuity worse than 20/40, or visual field of approximately less than 20°. Self-reported difficulty with three tasks was assessed at each visit: walking up 10 steps, walking down 10 steps, and walking 150 feet. Generalized estimating equation models included a 6-year spline, and explored differences in mobility difficulty trajectories by including an interaction between VI status and the spline terms. Odds ratios (OR) and 95% confidence intervals (CI) compared mobility difficulty for each task by VI status.
Results.:
At baseline, the VI were significantly more likely to report difficulty mobility tasks than the NVI (ORdifficultywalkingup10steps = 1.37, CI: 1.02–1.80; ORdifficultywalkingdown10steps = 1.55, CI: 1.16–2.08; ORdifficultywalking150feet = 1.50, CI: 1.10–2.04). The trajectory of mobility disability did not differ by VI status from baseline to the 6-year visit. However, the difference between the VI and NVI declined at the 8-year visit, which may be due to loss of VI participants at risk of developing mobility difficulty.
Conclusions.:
The VI were more likely to report mobility disability than the NVI, but the trajectory of mobility disability was not steeper among the VI as compared to the NVI over the study period.
In addition to the variables described above, the SEE study included age, sex, and self-designated race (white or black). The baseline values of these covariates were used in the analysis. Previous research has indicated that the risk of both visual impairment and disability increases nonlinearly with age.
8 To capture this nonlinear association, age at baseline was categorized as: 65 to 69, 70 to 74, 75 to 79, and ≥80 years.
We also examined the following covariates as time-varying: body mass index (BMI), smoking status, number of comorbid condition, presence of diabetes, the presence of depressive symptoms, and Mini-Mental State Examination (MMSE) score. The values of these covariates for an individual were allowed to change at each study visit. BMI was calculated as weight (kg) divided by height (m) squared measured at each visit, and was categorized into three groups: underweight (<18.5), normal weight (18.5 to <25), and overweight/obese (≥25). Smoking status was assessed via self-report and categorized as never smoker and current/former smoker.
Comorbid conditions are known to negatively impact mobility.
18,19 Therefore, participants were asked questions about their comorbidities using the lead in “has a doctor ever told you that you have…” The conditions asked about include: arthritis, hip fracture, back problem, heart attack or myocardial infarction, angina or chest pain, congestive heart failure, intermittent claudication pain in the legs, high blood pressure, emphysema, asthma after age 50, stroke, Parkinson's disease, cancer or malignancy, and vertigo or Meniere's disease. For these analyses, we categorized the number of comorbid conditions as: 0, 1, 2, or 3+ comorbid conditions.
As diabetes can lead to visual impairment as well as mobility disability, this comorbid condition was examined separately from the comorbidity covariates described above. The presence of diabetes was recorded if hemoglobin A1C values were above 7% or if a doctor had ever told the participant that they had diabetes. The presence of depressive symptoms was assessed using the seven-item depressive symptom subscale of the General Health Questionnaire.
20,21 This scale was designed to identify severe depressive symptoms in the general public. An individual is categorized as having depressive symptoms if they respond “yes” to one or more of the seven questions about worthlessness, suicidal thoughts, and hopelessness. Cognitive status was determined using the MMSE.
22 Scores on this test can range from 0 to 30, and cognitive impairment is suggested by scores less than or equal to 23.
Univariate analyses were used to compare the distribution of potential confounders by visual impairment status at baseline. We obtained P values from χ2 for categorical variables or t-tests for continuous variables comparing the VI with the NVI. Since visual impairment status was used as a time-varying covariate, we examined how the number of participants with visual impairment and nonvisual impairment changed by study visit, including the percentage of incident VI. We also determined the mean visual acuity (logMAR) and mean number of central visual field points missed by visually impairment status and study visit.
Generalized estimating equation (GEE) models were used to account for the correlation between the repeated measures.
23 Three separate regression models were run, one for each of the mobility outcomes: difficulty walking up 10 steps, difficulty walking down 10 steps, and difficulty walking 150 feet. From these models, odds ratios (OR) and 95% confidence intervals (CI) were estimated comparing the odds of reporting difficulty with each of these tasks among the VI to the odds among the NVI using an exchangeable working correlation structure. Our initial models included covariates that were significantly associated with visual impairment status from our univariate analyses. The most parsimonious statistical model was determined based on comparison of model fit using the quasi-likelihood under the independence model criteria.
24 In addition to visual impairment status and time since baseline (in years), our final models included baseline age categories, sex, and race, as well as time-varying values for smoking status, BMI categories, number of comorbid conditions categories, the presence of depressive symptoms, and the presence of diabetes. We modeled years since baseline using spline terms with a knot at the 6-year visit.
For each outcome, we included interaction terms between visual impairment status and the spline time covariates, which would allow for the slope to differ between the VI and NVI over each section of the spline. This allowed us to test our primary hypothesis that the change in the odds of reporting mobility disability over the study period differed between the VI and the NVI.
To check for emmigrative selection bias, we modeled the odds of being lost to follow-up compared with the odds of not being lost to follow-up at each study visit after baseline. These models included covariates for visual impairment status as well as the other covariates in our primary analyses from the visit prior to dropout. We ran three sets of models, one set for each of the mobility disability outcomes and included these variables as predictors of being lost to follow-up (i.e., three sets of models, with each set including one type mobility disability as a predictor of being lost to follow-up), and an interaction term between mobility disability and visual impairment status. This interaction term was included to determine if there was differential loss to follow-up of VI and disabled participants after adjusting for our other observed covariates.
We also ran sensitivity analyses where the visual acuity criteria used to define visual impairment was shifted from best-corrected distance visual acuity worse than 20/40 to acuity worse than 20/60, as this alternate cut point is the visual acuity criteria used for the ICD-10.
15 However, our results were similar our primary analyses and the inferences were unchanged, and only results from the primary analyses are presented.
All data were analyzed using statistical software (STATA Statistical Software: Release 12.1; STATA Corp., College Station, TX; and SAS; SAS, Inc., Cary, NC).
The results from this study indicated that from baseline to the 6-year visit, the VI were more likely to have mobility difficulty than the NVI. But, there was no evidence that the trajectory was different among the VI, compared with the NVI. This was contrary to our a priori hypothesis that the trajectory of mobility disability would be steeper among the VI as compared to the NVI over the 8-year study period.
Our results also indicated that for this study population, the odds of reporting of mobility disability over time were not linear. In fact, we observed that the percentage of individuals (both VI and NVI) reporting mobility disability did not continue to increase after the 6-year visit. Additionally, the difference in the odds of reporting mobility disability between the VI and NVI was no longer significant by the 8-year study visit. These findings were unexpected.
Therefore, we examined the possibility that we were observing a “healthy survivor” bias, meaning participants who were disabled and VI were preferentially lost to follow-up. We found no evidence of differential loss of the VI from baseline to the 6-year visit, suggesting the absence of a difference in the trajectory was not due to emmigrative selection bias. However, we observed that from the 6- to the 8-year visits the VI were more likely to be lost to follow-up than the NVI, and the disabled were more likely to be lost than the nondisabled. The interaction between visual impairment and disability was not significant, indicating that individuals who were VI and disabled were not more likely to be lost and suggesting the absence of differential loss to follow-up of these individuals. While it is possible that these cross-sectional models were not powered to detect this interaction at the last study visit, unadjusted analyses did not indicate a differential loss of VI and disabled participants (data not shown). Despite this result, we cannot rule out the possibility that the decline in mobility disability from the 6- to the 8-year visits was a result of loss to follow-up of VI participants who were at greatest risk of developing mobility disability. Overall, we found no indication that similarity of the mobility disability trajectory between the VI and NVI was due to emmigrative selection bias.
Our observation of no difference in mobility disability trajectory by visual impairment status was surprising. Previous cross-sectional studies reported that the VI are more likely to have mobility difficulty than the NVI,
6–11 and as a result we expected this difference would increase over time. But our observed difference in the odds of reporting mobility disability by visual impairment status at baseline may suggest that this difference occurs early in the process of visual loss, and possibly before the accumulation of other comorbid conditions. However, in the SEE study comorbidities were common. At baseline, over 30% of the NVI and more than 40% of the VI had three or more comorbid conditions. Further research including younger populations with few comorbidities would be needed to test this hypothesis. It is also possible that over time, the VI develop compensatory strategies to reduce their mobility difficulty, but we could not examine that hypothesis. At the time of the study, Salisbury, Maryland, did not have a low vision rehabilitation program.
In addition to visual impairment, other factors in our regression model were associated with the report of mobility disability. We found that individuals reporting mobility disability were more likely to be of older age, female sex, black race, past or current smokers, under- or overweight (based on BMI), have one or more comorbid conditions, report depressive symptoms, and have diabetes. These results were expected, as the covariates examined in this study were chosen based on our review of relevant literature that identified these factors as potential confounders of the association between visual impairment status and mobility disability.
2,18,25–34
Our results may have implications for low vision rehabilitation efforts. These efforts should include improvement of mobility, such as with the use of assistive devices and mobility and orientation training. However, the inclusion of mobility training in low vision rehabilitation settings is infrequent. In a survey of over 1000 US low vision rehabilitation clinics, only 28.7% of the polled programs offered mobility and orientation training.
35 We found that the negative impact of visual impairment on mobility disability largely remains as people age, and believe this result emphasizes the importance of including mobility training as a part of low vision rehabilitation efforts.
We thank Alison Abraham, Karen Bandeen-Roche, and Pradeep Ramulu for providing technical advice and expertise on the analyses used in this manuscript.
Supported by grants from the National Institute on Aging (National Institutes of Health): AG10184 and T32AG000247. The authors alone are responsible for the content and writing of the paper.
Disclosure: B.K. Swenor, None; B. Muñoz, None; S.K. West, None