Fourteen (larger gray circles) of the 510 item measures had significant differential item functioning. Items below the identity line were easier in the LVROS sample compared with the legacy sample.
eAppendix. Low Vision Research Network Study Group Members
eFigure 1. Standard Error Estimated From Item Responses in LVROS Sample and Combined Data Set (LVROS and Legacy)
eFigure 2. Association of Visual Ability as Measured by Patient Difficulty Ratings of Goals With logMAR Visual Acuity
eFigure 3. Association of Patient Reading Ability With logMAR Visual Acuity
Goldstein JE, Chun MW, Fletcher DC, Deremeik JT, Massof RW, for the Low Vision Research Network Study Group. Visual Ability of Patients Seeking Outpatient Low Vision Services in the United States. JAMA Ophthalmol. 2014;132(10):1169-1177. doi:10.1001/jamaophthalmol.2014.1747
Most patients with low vision are elderly and have functional limitations from other health problems that could add to the functional limitations caused by their visual impairments.
To identify factors that contribute to visual ability measures in patients who present for outpatient low vision rehabilitation (LVR) services.
Design, Setting, and Participants
As part of a prospective, observational study of new patients seeking outpatient LVR, 779 patients from 28 clinical centers in the United States were enrolled in the Low Vision Rehabilitation Outcomes Study (LVROS) from April 25, 2008, through May 2, 2011. The Activity Inventory (AI), an adaptive visual function questionnaire, was administered to measure overall visual ability and visual ability in 4 functional domains (reading, mobility, visual motor function, and visual information processing) at baseline before LVR. The Geriatric Depression Scale, Telephone Interview for Cognitive Status, and Medical Outcomes Study 36-Item Short-Form Health Survey physical functioning questionnaires were also administered to measure patients’ psychological, cognitive, and physical health states, respectively.
Main Outcomes and Measures
Predictors of visual ability and functional domains as measured by the AI.
Among the 779 patients in the LVROS sample, the mean age was 76.4 years, 33% were male, and the median logMAR visual acuity score was 0.60 (0.40-0.90 interquartile range). Correlations were observed between logMAR visual acuity and baseline visual ability overall (r = −0.42) and for all functional domains. Visual acuity was the strongest predictor of visual ability (P < .001) and reading ability (P < .001) and had a significant independent effect on the other functional domains. Physical ability was independently associated with (P < .001) overall visual ability as well as mobility and visual motor function. Depression had a consistent independent effect (P < .001) on overall visual ability and on all functional domains, whereas cognition had an effect on only reading and mobility (P < .001).
Conclusions and Relevance
Visual ability is a multidimensional construct, with visual acuity, depression, physical ability, and cognition explaining more than one-third of the variance in visual ability as measured by the AI. The significant contributions of the nonvisual factors to visual ability measures and the rehabilitation potential (ie, ceiling) effects they may impose on LVR are important considerations when measuring baseline visual ability and ultimately LVR outcomes in ongoing clinical research.
Low vision rehabilitation (LVR) improves patients’ ability to function with chronic disabling vision impairment. Low vision rehabilitation includes ascertaining the patient’s functional goals, implementing a goal-oriented treatment program to enhance the patient’s vision, teaching the patient to use impaired vision more effectively, and teaching nonvisual strategies that improve overall functional ability despite limited vision.1- 4
Measuring changes in overall functional ability as a result of LVR requires agreement on the definition of the variable(s). Functional ability refers to a person’s overall ability to perform activities with a criterion level of ease. When applying this definition for rehabilitation, the chosen activities should be those that are important for the person to achieve and maintain a desired life state. Functional ability is a multidimensional construct that includes a person’s physical ability, cognitive ability, sensory ability, and various subdivisions of each of these general domains.5 In the case of LVR, we are interested in a person’s overall ability to perform activities that depend on vision, a variable we call visual ability. However, few activities depend only on vision; therefore, reports of fatigue, chronic pain, and depression, in addition to visual limitations, are essential parts of the LVR evaluation.2,6- 8
The multifaceted nature of visual ability and the personalized nature of the rehabilitation plan challenge the development and validation of clinically meaningful and responsive LVR outcome measures, which are needed to conduct clinical research and are requisites to evidence-based LVR practices. Traditional strategies for measuring LVR outcomes include (1) performance measures,9- 12 (2) physician or therapist ratings of the abilities of patients,13,14 and (3) patient self-reports of ability.15- 17 Because the treatment objective of LVR is to improve visual ability within the boundaries of individual patient preferences, health states, and environmental constraints, most LVR outcome studies18,19 have used patient-reported outcome instruments to measure treatment effectiveness.
With the advent of modern psychometric methods (eg, Rasch analysis), it is no longer necessary to use patient-reported outcome instruments with a fixed set of items. The current trend is to develop a large calibrated item bank for the population of interest and, with computer assistance, present a subset of items that are selected in an adaptive fashion.20- 22 The Activity Inventory (AI) is an adaptive patient-reported outcome instrument designed to measure LVR outcomes.23- 25 The items describe specific vision-dependent cognitive and motor activities (tasks) that are nested under more general descriptions of the activities’ goals (eg, reading bills, writing checks, and using a computer, all serving the goal of managing finances). Patients’ ratings of the importance and difficulty of goals determine which tasks will be rated by the patient. One advantage of the AI is that only items important to the patient and difficult to perform are targeted in the rehabilitation plan and therefore contribute to the outcome measure. Another advantage of the AI is that the item bank need not be static; items can be created and calibrated for different cultures or changes in society and technology.16
The AI is well validated as a psychometric measure of visual ability in patients with low vision.6,23- 26 Different summary measures estimated from AI responses reflect the confounding properties of the multiple dimensions of visual ability.19 The aims of this study are to examine the patient variables that contribute to visual ability and functional domain measures and determine how nonvisual traits of patients modify these measures.
The Low Vision Rehabilitation Outcomes Study (LVROS) is a collaborative, prospective, observational study designed to measure the effectiveness of LVR provided in 28 outpatient centers throughout the United States. Recruitment, participation, and data collection methods have been described elsewhere.27 The protocol and informed oral consent procedure were approved by the institutional review board of The Johns Hopkins University and adhered to the tenets of the Declaration of Helsinki. In addition, study sites complied with the requirements of the Health Insurance Portability and Accountability Act and, when required, obtained separate institutional review board approval. All patients gave informed oral consent for study participation.
Data from 2 patient samples were used for the initial analysis. In both samples, eligible patients were 18 years or older and new patients to the physician.27 Patients who were unable to communicate with the research assistant interviewers by telephone were excluded. There were no visual impairment or other eligibility requirements. The LVROS sample has 779 patients recruited from 28 centers in the United States and enrolled from April 25, 2008, through May 2, 2011. The legacy sample has 2398 patients recruited from the Wilmer Eye Institute LVR Service from April 4, 2001, through September 14, 2007.
Each center identified new patients scheduled for appointments who, after written authorization, were contacted by the coordinating center at The Johns Hopkins University. Oral consent was then obtained by telephone, and patients were administered several questionnaires over the telephone by research assistants from The Johns Hopkins University, including the AI to measure overall visual ability and its different functional domains,2,23- 25 the Telephone Interview for Cognitive Status (TICS),28 the Geriatric Depression Scale (GDS),29 the Medical Outcomes Study 36-Item Short-Form Health Survey (SF-36) physical functioning component to measure physical limitations,30 and a standardized check-box intake survey to acquire patient trait information.27
The 510-item AI is administered adaptively by computer-assisted interview. The hierarchical structure of the questionnaire organizes the functional interview into goals and tasks surrounding daily living, social, and recreation-related activities. The patient is asked to rate the importance of each goal (not important, slightly important, moderately important, or very important). If the goal is not important, the interviewer moves onto the next goal; otherwise, the patient is asked to rate the difficulty of the goal (not difficult, slightly difficult, moderately difficult, very difficult, or impossible). If a goal is at least slightly important and at least slightly difficult, the patient is asked to rate the difficulty of subsidiary tasks (using the same difficulty ratings) or identify the task as not applicable. The AI consists of 50 goals and 460 tasks nested under the goals, which provides a tailored visual function measure for each patient.
Usual LVR was provided by the physicians and the rehabilitation team where applicable after the baseline interview. Clinical findings from examinations and disorder diagnosis were recorded and entered online into a secure web server. Centers agreed in advance on the parameters for testing.27 Data from the legacy sample were obtained similarly. Telephone interviews included the AI and the intake survey. Clinical findings were collected via medical record review and entered in the study database.23- 25
Trait distributions for the LVROS and legacy samples were summarized to determine whether the 2 data sets could be combined for the purpose of increasing the precision of the item measure estimates in the calibrated item bank. Rasch analysis, using the Andrich Rating Scale Model (Winsteps statistical software, version 3.65), was performed on the combined data sets of baseline AI responses to estimate item measures and response category thresholds, as well as estimating a set of person measures for each patient. Differential item functioning (DIF) analysis was performed to evaluate disagreement in AI item calibrations between the data sets. Overall visual ability was estimated for each patient from difficulty ratings of AI goals and measures of functional domains of visual ability (reading, mobility, visual information processing, and visual motor function) from difficulty ratings of different subsets of tasks. The item measures and response category thresholds were anchored to values estimated from the analysis of patient difficulty ratings across all AI goals and tasks.
Rasch analysis also was performed on the responses to the GDS and SF-36 to estimate continuous interval–scaled measures for depression and physical functioning, respectively. Rasch analysis was not performed on the TICS responses because the raw scores were heavily skewed toward the ceiling; therefore, raw scores were used as the continuous variable for cognitive ability, which is expected to be a monotonic function of an interval measure that would be estimated from Rasch analysis.31
Pearson correlations were used to identify potential covariates and determine the associations with visual ability measures and other continuous trait variables. To confirm previously observed associations and evaluate the effects of other categorical variables (eg, sex, living situation, and driving status), 1-way analysis of variance (ANOVA) was performed with overall visual ability as the dependent variable. Exploratory analyses of covariance (ANCOVAs) using fixed models were performed. The covariates chosen were continuous variables that exhibited higher Pearson correlations with the dependent variable and significant effects in the ANOVA. Bivariate and stepwise multivariable regression models for visual ability and functional domain measures were developed incorporating independent variables that were significant predictors of the dependent variable. Bonferroni correction was used to determine significance criteria for the 4 continuous independent variables.
Patient characteristics, visual impairment measures, and baseline functional ability were obtained from 779 LVROS patients. Because of the adaptive nature of the AI, the median number of respondents to each item was 149 (interquartile range, 58-266). To expand the database, a legacy data set of 2398 patients was combined with the LVROS data set to produce a sample of 3177 patients with low vision for calibrating AI item measures and response category thresholds. For the legacy data, the median number of patient responses to each item was 388 (interquartile range, 155-718).
Table 1 compares baseline characteristics of the LVROS and legacy patient samples. Mean logMAR visual acuities (VAs) were nearly identical for the legacy and LVROS samples (0.55 vs 0.60, respectively). Mean visual abilities were not significantly different for reading (P = .87) or visual motor function (P = .10). The mean measures for goals (P = .03), mobility (P < .001), and visual information processing (P < .001) were significantly different; however, all differences were less than the mean SEs of the visual ability estimates (Table 1).
A comparison of estimates of AI item measures between the LVROS and legacy patients is shown in the Figure. The DIF analysis using the Mantel-Haenszel method identified 14 of 510 items with significant DIF. Most of these 14 items were related to reading mail, financial management, and computer use and were easier in the LVROS sample.
By combining the databases, the median number of patients responding to an AI item increases to 548 (interquartile range, 213-976). Item measure precision is summarized with the SE. The SE for a given item is inversely proportional to the square root of the number of persons who respond to that item and represents within-person variance. The mean SEs for item measures estimated from the combined data sets are approximately half the value of the SEs for item measures estimated from LVROS responses alone (eFigure 1 in the Supplement). Given the overall lack of DIF and agreement in the item measures between samples (Figure), the AI data sets were combined to generate more precise calibrations of item measures and response category thresholds. Visual ability and functional domain measures for LVROS patients were estimated with item measures and response category thresholds anchored to values estimated from the combined data sets.
Table 2 provides the Pearson correlations of the continuous variables. No significant correlations were found between measures of VA and depression, VA and physical functioning, contrast sensitivity (CS) and depression, and CS and physical functioning. No significant correlations were found between age and measures of reading, mobility, visual motor function, VA, and CS. The strongest correlations were found among the different AI measures, the AI measures (except mobility) and VA, the AI measures (except mobility) and CS, and VA and CS.
Results of ANOVA and ANCOVA of patient visual ability (dependent variable) and patient characteristics (independent variables) are listed in Table 3 and Table 4. Because of ceiling effects, many of the trait distributions were not normal; therefore, continuous independent variables were categorized into quartiles for this analysis. Significant main effects were seen with all independent variables, but after correction for multiple comparisons, the effects of age, living situation, and self-reported fluctuations of vision did not achieve the α criterion of .05. Post hoc multiple comparisons using the Bonferroni method reveal that traits assessed with categorized continuous variables have significant ordered effects (not threshold effects) on visual ability. The ANCOVA results, with the covariates listed in Table 4, reveal that depression, physical functioning, and VA have significant independent effects on ability to achieve AI goals. No interactive effects were observed between depression and physical functioning (P = .71 for VA and P = .92 for TICS) and cognition and depression (P = .63). Correcting for multiple comparisons, cognitive function did not have a significant independent effect on visual ability measures (P = .02). Requiring support services from family or friends has a significant effect on visual ability (P < .001); however, living alone does not (P = .39).
Table 5 provides the regression analysis coefficients for continuous independent variables identified as having independent effects on visual ability. Regression model coefficients also are displayed for the same continuous independent variables when functional domain measures are dependent variables. Cognition was included as an independent variable in the regression analysis because of suggestions from previous work.27,32,33 Confirming the ANCOVA results, SF-36 physical functioning (P < .001), GDS (P < .001), and logMAR (P < .001) measures are significant independent predictors of visual ability. The TICS scores do not have a significant effect on overall visual ability (P = .80), although they have a significant independent effect on reading. Reading is most strongly predicted by logMAR (P < .001), GDS (P < .001), and TICS (P < .001), whereas SF-36 physical functioning (P = .12) does not have a significant effect. The SF-36 physical functioning (P < .002), GDS (P < .001), and logMAR (P < .001) measures are predictive of mobility, visual information processing, and visual motor function, whereas the TICS scores predict mobility (P = .007). On the basis of the magnitude of β, logMAR VA measures explain the most variance in the dependent variables, with the exception of mobility, for which SF-36 physical functioning measures explain the most variance. Linear combinations of these 4 health state variables (VA, physical functioning, depression, and cognition) explain the greatest variance in visual ability at the goal level (r2 = 0.367), followed by reading (r2 = 0.338), visual motor function (r2 = 0.309), mobility (r2 = 0.251), and visual information processing (r2 = 0.240). Bivariate analyses illustrate the linear trend between visual ability and logMAR VA (eFigure 2 in the Supplement) and reading and logMAR VA (eFigure 3 in the Supplement).
Our findings agree with prior research reporting that VA, physical functioning, and depression are independent contributors to measures of visual ability in patients.8 Cognition also contributes to measures of visual ability but mostly in the area of reading and to a lesser extent mobility. The need for support services is an independent predictor of visual ability, but age, sex, living situation, and self-reported fluctuations in vision are not. Overall, we observed that VA in patients is the strongest independent predictor of visual ability measures at the goal and domain levels. However, the regression analyses constructed from VA, physical ability, depression, and cognitive function measures account for 24% to 37% of the variance in visual ability. The large unexplained variance indicates that we still need to explore other visual impairment domains, other patient traits, and a variety of environmental factors.
The nonvisual traits that modify visual ability emphasize the importance of performing a health systems review to determine the presence and magnitude of the characteristics that may affect LVR.7,8 Patients seeking LVR are older and present with chronic health concerns that commonly are progressive and often not modifiable. Our findings indicate that the number of comorbidities is not a good predictor of visual ability. Rather, measures of the patients’ ability to perform activities, such as walking and climbing steps, is a significant predictor of visual ability and should be used rather than the number of comorbidities when assessing the effects of physical functioning.34 Mobility and visual motor function domains are significantly affected by physical functioning; however, VA and depression also have independent effects. This finding is consistent with the strong evidence that individuals with even mild VA loss have a greater risk of falls.35- 37 Consistent with our findings, prior research supports independent effects with similar magnitudes of physical functioning, VA, and depression on household management and mobility concerns in populations with low vision.8 When physical functioning components are the primary limiting factor in improving visual ability, ceiling effects on LVR effectiveness should be expected.
Several studies38- 40 have reported increased rates of depression in populations with low vision. In this study, the GDS measures reveal that 23.0% of patients seeking outpatient LVR services have at least mild depression and an additional 1.6% have severe depression. Our findings indicate that depression has an independent effect on visual ability and all functional domains, a result nearly identical to that seen in 2 previous studies7,8 that examined the effects of nonvisual factors. Note that our results are cross-sectional and that the relationship between vision loss and depression has not been proven to be causal. If loss of visual ability is the sole cause of depression and LVR is effective, the emotional state has the potential to improve markedly. If, however, depression is related to other health problems, a ceiling may be imposed on LVR outcomes.
Analysis of the TICS measures revealed that 10% of patients present with at least mild cognitive impairment. Although cognitive loss was not a significant predictor of overall visual ability, it had an independent effect on mobility and reading function. Because reading is the primary concern of most patients seeking LVR services, cognitive functioning should receive attention during the clinical evaluation because it can limit the patient’s rehabilitation potential and outcome.41,42
Patients seeking outpatient LVR services in the United States have moderate impairment, with median VA Snellen values ranging from 20/71 to 20/80 (legacy and LVROS groups, respectively). Study results indicate that VA is the strongest predictor of visual ability and all functional domains. Although VA is a strong predictor, it accounts for only 15.4% of the variability in visual ability. Other vision variables, such as scotomas and visual field loss, may contribute to the unexplained variability, especially for mobility.43 Although CS is not an independent linear predictor of visual ability, we are not concluding that CS is unimportant to the patient’s ability to function.
Depression, VA, physical functioning, and, less so, cognition are the strongest independent predictors of patients’ visual ability measured with the AI at baseline. For visual ability to be a clinically meaningful outcome measure, it requires an understanding of the measure and effect modifiers to proceed with clinical trials outcomes research in LVR.
Group Information: The Low Vision Research Network Study Group members are listed in the eAppendix in the Supplement.
Submitted for Publication: December 13, 2013; final revision received February 11, 2014; accepted February 14, 2014.
Corresponding Author: Judith E. Goldstein, OD, Lions Vision Research and Rehabilitation Center, Wilmer Eye Institute, the Johns Hopkins University School of Medicine, 600 N Wolfe St, Baltimore, MD 21287 (email@example.com).
Published Online: June 19, 2014. doi:10.1001/jamaophthalmol.2014.1747.
Author Contributions: Drs Goldstein and Massof had full access to all the data in the study and take responsibility for the integrity of the data and the accuracy of the data analysis.
Study concept and design: Goldstein, Fletcher, Massof.
Acquisition, analysis, or interpretation of data: All authors.
Drafting of the manuscript: Goldstein, Chun.
Critical revision of the manuscript for important intellectual content: All authors.
Statistical analysis: Goldstein, Massof.
Obtained funding: Deremeik, Massof.
Administrative, technical, or material support: Goldstein, Chun, Fletcher.
Study supervision: Goldstein, Massof.
Conflict of Interest Disclosures: None reported.
Funding/Support: This study was supported by grants EY012045 and EY018696 from the National Eye Institute, National Institutes of Health, and a grant from Reader’s Digest Partners for Sight Foundation (Dr Massof).
Role of the Sponsor: The funding sources had no role in the design and conduct of the study; collection, management, analysis, and interpretation of the data; preparation, review, or approval of the manuscript; and decision to submit the manuscript for publication.