| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
|
|
|
|||||||
|
|||||||||
1 From the Dana Center for Preventive Ophthalmology, Johns Hopkins School of Medicine, Baltimore, Maryland; 2 Department of Ophthalmology, University of Arizona, Tucson; and 3 Department of Ophthalmology, University of Wisconsin, Madison.
 |
Abstract |
|---|
 Top Abstract IntroductionMethodsResultsDiscussionAppendix 1References |
|---|
METHODS. A sample of randomly selected block groups in Tucson and Nogales,
Arizona, were selected for study. Participants were interviewed at
home; a majority of the interviews were conducted in Spanish. The home
interview included questions from the NEI-VFQ-25 and HHANES. Presenting
acuity was done using ETDRS methodology, followed by a standardized eye
examination by an ophthalmologist. The authors analyzed the internal
consistency of the NEI-VFQ-25 responses using Cronbach’s 
coefficient and the construct validity by assessing the relationship
between presenting acuity and scale scores, adjusting for age and
gender. A second model was also explored to determine whether other
demographic variables affected scale scores; differences in reporting
between the Spanish and English versions was observed in this model,
used in a subset of the population that minimized interviewer effect.
RESULTS. Of the 4774 participants in the study, 99.7% had completed
questionnaires, not completed by proxy. The highest nonresponse rate
occurred in the Driving scale, with 25% of participants not driving
for reasons other than problems with vision. Internal consistency was
high, with Cronbach ranging between 0.65 and 0.86 for scales with
multiple items. Adjusting for age and gender, those with presenting
acuity worse than 20/40 scored significantly lower than those with
presenting acuity 20/40 or better, for all scales. The demographic
variables with the most consistent association across the NEI-VFQ-25
scales were presenting acuity, income, and gender. No significant
differences in reporting were found between the Spanish and English
versions of the questionnaire in the subset of the population.
CONCLUSIONS. In this study of Hispanic people age 40 years or older, the NEI-VFQ-25 was sensitive to presenting acuity and other demographic variables, such as age, gender, and income. The findings from this psychometric analysis provide evidence of the reliability and validity of some of the scales in the 25-item NEI-VFQ when used among people with a range of visual acuity level, providing other explanatory variables are also considered.
|
Introduction |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
The Activities of Daily Vision Scale (ADVS) and VF-14 were created to capture the decrements in physical function specifically associated with visual impairment. Recovery of visual function after cataract surgery was associated with improvements in the ADVS,3 whereas the VF-14 was associated strongly with functional impairment caused by cataract.4 In previous work we have shown that the ADVS is sensitive not only to vision loss, but also age, gender, and racial variation.5 As research interest in this area has grown, several "quality of life" instruments have been developed to capture decrements not just in the domain of physical function, but in psychological function, social function, and other domains as well. In this spirit, the National Eye Institute has sponsored research into the development of a vision specific quality of life instrument, the NEI-Visual Function Questionnaire (NEI-VFQ).6
In contrast to the ADV and VF-14 instruments, which were developed for use in cataract patients, the NEI-VFQ was developed for use among patients with a range of visual disabilities and impairments, with the goal of creating a comprehensive assessment of quality of life associated with visual function. The 51-item NEI-VFQ has been validated for a group of persons with chronic eye disease.7 A shortened version of the NEI-VFQ is now in use, which consists of a subset of 25 items from the original 51-item scale. The properties of this shortened version have not been described, but it has been used frequently, in part because of the shortened time for administration.8 9 10
The purpose of this study is to assess the psychometric behavior of the NEI-VFQ-25 in a population of older Hispanic people living in the United States, to explore other demographic factors that affect participant response, and to observe the comparability of the Spanish and English versions of the instrument. Our population differs from the population in which the instrument was developed in that our population is relatively healthy and has a common ethnic and cultural background that reflects their Spanish heritage. We present data that examine the reliability and validity of the 25-item NEI-VFQ used among this population.
|
Methods |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
Home Interview and Clinical Measures
The majority of home interviews (80%) were conducted in
Spanish. The questionnaire consisted of specific questions on
education, socioeconomic and health status, health care utilization,
history and duration of diabetes, history of vision problems, visit to
an eye care professional, Native-American ancestry, the 25-item
NEI-VFQ, and a series of questions from the Hispanic Health and
Nutrition Examination Survey (HHANES).12
The questionnaire
was translated in-house from English to Spanish and then was
back-translated to verify accuracy.
Presenting and best corrected visual acuity was measured at the clinic site. The following methods were used for assessing visual acuity for each eye: an autorefractor (Humphrey Auto-Refractor; Humphrey Instruments, San Leandro, CA) was used as a starting point for a full subjective refraction; distance acuity was tested with the Early Treatment Diabetic Retinopathy Study (ETDRS) chart13 at 3 m, illuminated at 130 cd/m2. Participants who failed to read the largest letters at 3 m were retested at 1.5 m and then at 1 m. Visual acuity was scored as the total number of letters read correctly, transformed to LogMAR units. Failure to read any letters was assigned an acuity of 1.7 LogMAR units, which is equivalent to an acuity of 20/1000. Presenting acuity was measured with the participant’s habitual distance correction. An E chart13 was used for participants who were illiterate. Data collection started in April of 1997 and ended in September of 1999.
25-Item NEI-VFQ
The NEI-VFQ-25 was created from a subset of the original 51-item
NEI-VFQ. The shortened questionnaire consisted of items from the
following scales: General Health, 1 item; General Vision, 1 item;
Ocular Pain, 2 items; Near Vision, 3 items; Distance Vision, 3 items;
vision-specific Social Functioning, 2 items; vision-specific Mental
Health, 4 items; vision-specific Role Functioning, 2 items; Dependency
due to Vision, 3 items; Driving, 2 items; Peripheral Vision, 1 item;
and Color Vision, 1 item. Ratings for each question consisted of the
following types: "amount of difficulty," "amount of time,"
"amount true," and "amplitude". Ocular pain consisted of 1
"amount of time" item and 1 "amplitude of pain" item; mental
health consisted of 1 "amount of time" item and 3 "amount true"
items. All other scales consisted of items of the same type.
Item responses were adjusted for directionality (high values reflect participants with good vision or health) and were transformed to a scale of 0 to 100. Adjusted items belonging to a scale were averaged together to create a single scale score. Participants were excluded from the analysis if they had stopped doing the activity for reasons other than poor eyesight or if someone other than the participant had answered their questionnaire. The number of participants responding to items and the distribution of item responses were examined.
Statistical Analysis
To estimate the internal consistency of the NEI-VFQ-25 for our
population, we calculated the Cronbach for multi-item scales and
average inter-item correlation. Cronbach’s coefficient is the
proportion of a scale’s total variance that is attributable to a
latent variable underlying the items (e.g., near vision, distance
vision). Acceptable coefficient is between 0.7 and 0.9: an that
is too low signifies low homogeneity among items, whereas a very high
coefficient reflects redundancy.14
Coefficient
increases not only as the strength of the correlation between items
increases, but also as the number of items in the scale increase.
Average inter-item correlation is the average of a scale’s
item-to-item correlation coefficients. The inter-item correlation will
be less than or equal to and is not influenced by the number of
items in a scale.15
If a scale has high and low
inter-item correlation, it is most likely because of a greater number
of items in that scale rather than a high correlation between items.
Item–scale correlation coefficients were observed for additional information on scale homogeneity. If an item was not part of a subgroup (e.g., reading newsprint and General Health), the correlation was directly calculated; if an item was part of a subgroup (e.g., reading newsprint and Near Vision), the item was correlated with the sum of the other items in that subgroup. In general, we expect that items belonging to scales will show higher correlation with that scale than items not belonging to the scale. Scales with only two items will reflect the degree to which the items correlate with each other; thus, we do not necessarily expect the correlation coefficients to be high for two-item scales. Streiner suggests that an adequate item–scale correlation be above 0.20.
Statistically, an item correlated with another single item will have lower correlation than an item correlated with the sum of two or more items with similar variance and direction. For more equivalent comparisons between items in and out of the scale, we calculate an inflation factor for the within-scale item correlations. The derivation of this inflation factor is shown in the Appendix.
Construct validity is concerned with the theoretical relationship
between variables. A valid construct will behave in an expected manner,
with regard to another variable.16
17
We assess validity
by estimating the statistical significance of presenting acuity in the
better eye (hereafter referred to as "presenting acuity") and the
NEI-VFQ-25 scales. In our model, presenting acuity was represented by a
binary variable: acuity of 20/40 or better and acuity worse than 20/40.
We also account for age and gender. The linear model is as follows:
 | (1) |
 |
 |
 |
 |
 |
 |
We expect participants with worse than 20/40 presenting acuity to have significantly lower scores in NEI-VFQ-25 scales than those with better acuity. Scales that describe activities that require central visual acuity, such as Near Vision and Driving, should have stronger responses than scales that describe other aspects of vision-related quality of life, such as Ocular Pain or Color Vision.7
Although the item content of the NEI-VFQ-25 focuses on distinguishing vision-related problems, we were interested in how the scales were related to other demographic variables. The variables we thought would be associated with scale scores included age, gender, acculturation, English or Spanish questionnaire, income, level of education, and whether the participant had medical insurance. Age is the only continuous variable in the model.
The acculturation index is taken from the Hispanic Health and Nutrition
Examination Survey (HHANES).18
Acculturation, as used in
HHANES, refers to two components of adaptation to an English-speaking
culture in the United States. The first component measures the trend
toward increasing use of English for reading, writing, and conversing;
the second accounts for the number of generations of family born in the
United States. The acculturation score ranges from 1 to 5, where 1 is
low acculturation and 5 is high acculturation. Dividing the score into
low (1–1.5), middle (1.5–2), and high (2+) acculturation levels
creates our acculturation variable. The income variable was created by
dividing income into two levels: family income less than $20,000 (low
income) and family income $20,000 or higher (high income). Similarly,
the education variable was divided into two levels: those who had
completed high school and those who had not. The full model is as
follows:
 |
 |
 |
 |
|
|
|
|
|
|
|
|
 |
 |
 |
 |
 |
 |
 |
In addition to using this model in the full population, we used the model in two subgroups of the population: those with impaired presenting acuity and those with nonimpaired presenting acuity. This subdivision allowed us to explore the possible interaction between presenting acuity and the demographic variables.
We explored the possible nonlinear relationship of scale scores and age using descriptive plots of the data. Plots of scale score against age were plotted, and a smoothed spline was calculated in S-plus, along with 95% confidence intervals.
To assess the comparability of the Spanish and English versions of the questionnaire, we selected responses from interviewers who each gave 60 or more interviews and whose proportion of English interviews was 30% or more. The goal was to obtain a roughly equal amount of participants taking the Spanish and English questionnaires and to remove a possible interviewer effect. We adjusted scores for the demographic variables that were hypothesized to affect the score: age, gender, acculturation, income, education, and medical insurance. If the Spanish and English versions of the interview are equivalent, we expect to see no differences between reporting on the different versions of the questionnaire.
|
Results |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
The number of responding participants and distribution of item responses is shown in Table 1 . The lowest number of responses occurred in Driving (day and night), with a nonresponse (for reasons other than eyesight) rate of approximately 25%. Going to movies had the next highest nonresponse, occurring at 10%. The rest of the items ranged between 0% and 3% nonresponse.
|
Items rated for "amount of difficulty" and "amount of time" were generally skewed, with a high proportion of responses in the "no difficulty" and "not often" categories (ceiling effect). Exceptions occurred in general health and general vision, where responses were more normally distributed, and in "amount of time" worrying about eyesight, which was close to uniformly distributed. "Amount true" items tended to be bimodally distributed, with the lowest proportion of responses in the middle category ("don’t know" response).
Internal consistency statistics are shown in Table 2
. Cronbach scores are highest for the Dependency scale, at 0.86, and
lowest for the Driving scale, at 0.65. The Driving and Ocular Pain
scales, which have < 0.7, consist of only two items. This
paucity of items results in a lower estimate of . Note that although
the Mental Health scale has a high Cronbach coefficient, it also
has a relatively low average inter-item correlation. The high for
the Mental Health scale most likely occurs because it contains four
items, the highest number of any of the scales.
|
|
|
|
A summary of the association of other demographic variables is shown in Table 6 . The variables that have significant and consistent directionality across the NEI-VFQ-25 scales are income, gender, and presenting acuity. Those with high yearly income had better scores in all scales. Women had lower scores than men, except in Color Vision. Participants with middle acculturation generally had better scores than both those with low acculturation and those with high acculturation, except in Ocular Pain. There was no association between age and scores in General Vision and Mental Health. Near Vision and Ocular Pain showed a positive association with age. Those with medical insurance tended to have lower scores in General Health and Ocular Pain scales.
|
When analyzing the impaired and nonimpaired subgroups separately, we found that parameter estimates that were significantly associated with scale scores from the total group continued to be significantly associated in the nonimpaired group. In the impaired subgroup, however, some of the associations became nonsignificant because of the few number of participants in this group (only 8.1% with worse than 20/40 presenting acuity). However, we can evaluate the influence of the demographic variables on the basis of the magnitude and direction of the estimates.
In a majority of the scales (Distance Vision, Driving, Role Function, Dependency, Social Function, Mental Health, Color Vision, and Peripheral Vision) the directionality of the parameter estimates in the impaired group was similar to estimates in the nonimpaired group. In these scales, the magnitude of the estimates from the impaired group was equal to the corresponding estimates in the nonimpaired group. In other scales (General Health, General Vision, Near Vision, Ocular Pain, and Peripheral Vision) the magnitude of the estimates from the impaired group was equal to the corresponding estimates in the nonimpaired group but with opposite directionality. Thus, the influence of the demographic variables is not diminished in the group with visual impairment and is important in determining selfreported function.
The behavior of age and the different NEI-VFQ-25 scales varies, with evidence of a threshold effect in some scales and a general linear decline in others. Figure 1 shows smoothed plots of some scale scores against age: we observe some nonlinear trends as age changes. Scales such as Driving, Distant Vision, and Role Difficulties are flat and then decline after a certain age. In the Near Vision scale, the scores start high at age 40 years, slope downward at about age 50 years, and then climb back by age 70 years. This nonlinear association may better explain the relationship of age and near vision than a linear association and present a more understandable picture.
|
|
Discussion |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
Comparing scores from participants with nonimpaired and impaired presenting acuity showed evidence for validity of the questionnaire. Impaired presenting acuity was significantly associated with decrements in all scales in the NEI-VFQ-25, accounting for age and gender. Ideally, this instrument should be sensitive only to different levels and types of visual impairment. The value of testing the NEI-VFQ-25 in a population-based study is to determine non-vision related variables that may influence item responses. We found that among both visually impaired and non–visually impaired groups, demographic variables such as age, gender, and income play an important role in selfreported function. However, we also found a limitation to its use in this type of study: we were not able to capture as wide a distribution of selfreported values as might exist in visually impaired groups with large numbers of persons with severe impairments. This truncation in the distribution can be seen in the high ceiling effects generated by this generally healthy population.
We also found some limitations in the scale properties of the NEI-VFQ-25. First, we could not determine the internal reliability of several scales, because the scales consisted of only one item. These items include Peripheral Vision, Color Vision, General Health, and General Vision. Although we may explore the validity of these scales, there is no way to determine internal consistency, because there are no other items measuring a similar variable with which to compare. These items should probably not be classified as scales.
Second, the nature of the response to items differed, with some responses being "amount of difficulty," some being "amount of truth," and others being "amount of time." Combining such disparate responses into an overall score assumes some underlying rationale for equivalencies of these responses, which is unknown and likely not valid.19 It is for this reason that we chose to work only within the scales and across the scales without presuming a summary score of all the scales (overall score). This issue also presents a problem in the mental health scale, where one of the four items is a "how often" question, whereas the other items were "amount true" items.
Third, some of the item–scale scores correlated better with other scales than the scales to which they were assigned. This was especially true for Social Function scale, which has a higher correlation with "going to see movies, etc." than with items in the scale. The Mental Health scale was highly correlated with items from the Dependency and Role Function scales, suggesting that the Mental Health scale might be absorbed, rather than be a scale unto itself. The high correlation may also be attributed to the scale construct: both Mental Health and Dependency items use "amount true" type response constructs.
Differences in response between the Spanish and English translations of the home questionnaire are not likely due to differences in translation, given our rigorous methods of translating and back-translating the questions. Also, we found that the majority of scales showed no differences in reporting between the Spanish and English versions in the subset of participants who had interviewers administering large numbers of both versions. A possible explanation for the differences we did find is that those who selected the Spanish version were inherently different from those who selected the English version, even after accounting for a measurement of "acculturation." Because most of those who exclusively chose the Spanish version resided in Nogales, compared with Tucson, we cannot rule out an effect of living in a border community with strong cultural ties to Mexico.
It is important to note that we only performed certain analyses of the psychometric properties of this instrument. Reliability was only assessed as internal consistency and item–scale correlation. Also, we examined the validity of the instrument in relation to visual acuity loss only and to no other measure of visual loss. Clearly, visual loss may have multiple dimensions, such as loss of contrast sensitivity, peripheral field, stereo acuity, and other measures, some of which may be more closely correlated with decrements in the selfreported scores. However, our intent was to show general validity, which was demonstrated with the decrements in acuity.
We collected information on the existence of diabetes and high blood pressure, which we found were associated with scale score, diabetes more so than high blood pressure. However, other co-morbidity information was not collected and was most likely absorbed in the age and possibly income variables. Therefore, further refinement by consideration of other co-morbid conditions is warranted. Finally, this study presents results of the NEI-VFQ-25 performance in a population-based study consisting of Mexican Americans; we realize that different responses and interpretations of the questions may occur in other population groups with a different cultural framework. It is important that such studies be carried out to show the utility of the NEI-VFQ in heterogeneous populations.
In summary, in this study of Hispanic people age 40 years or older, the NEI-VFQ-25 was sensitive to presenting acuity and to other demographic variables, such as age, gender, and income. The findings from this psychometric analysis provide evidence of the reliability and validity of some of the scales in the 25-item NEI-VFQ when used among people with a range of visual acuity levels. Because demographic properties of a population are associated with response to scale items, explanatory variables other than visual function should be considered when creating hypotheses about the scales. A summary score across all scales is not justified for this instrument.
|
Appendix 1 |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
 |
j is the random error in item j.
We make the traditional assumptions about j:
cov(T, j) = 0 and
cov(j,
k) = 0, for j 
k. In addition, we assume that var(T) =
2 and var(j) =
2, independent of j. Then, we
obtain the following:
 |
 |
 |
The correlation between item 1 and the average of items 2 and 3 is
then
 |
 |
 |
 |
when 
= 0, to 1 when = 1. We can estimate
with the average inter-item correlation for a given scale. |
Acknowledgements |
|---|
|
Footnotes |
|---|
Submitted for publication June 30, 2000; revised October 18, 2000; accepted November 2, 2000.
Commercial relationships policy: N.
Corresponding author: Sheila K. West, Wilmer Room 129, Johns Hopkins Hospital, 600 N. Wolfe Street, Baltimore, MD 21287. swest{at}dcpom.med.jhu.edu
|
References |
|---|
|
TopAbstractIntroductionMethodsResultsDiscussionAppendix 1References |
|---|
This article has been cited by other articles:
|
|
 |
|
  K W Muir, C Santiago-Turla, S S Stinnett, L W Herndon, R R Allingham, P Challa, and P P Lee Health literacy and vision-related quality of life Br. J. Ophthalmol., June 1, 2008; 92(6): 779 - 782. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 A. T. Broman, G. Hafiz, B. Munoz, J. Rodriguez, R. Snyder, R. Klein, and S. K. West Cataract and Barriers to Cataract Surgery in a US Hispanic Population: Proyecto VER Arch Ophthalmol, September 1, 2005; 123(9): 1231 - 1236. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. K. West, B. Munoz, G. S. Rubin, K. Bandeen-Roche, A. T. Broman, and K. A. Turano Compensatory Strategy Use Identifies Risk of Incident Disability for the Visually Impaired Arch Ophthalmol, September 1, 2005; 123(9): 1242 - 1247. [Abstract] [Full Text] [PDF]
|
|
|
|
 |
|
 D. Globe, R. Varma, S. P. Azen, S. Paz, E. Yu, S. Preston-Martin, and the Los Angeles Latino Eye Study Group Psychometric Performance of the NEI VFQ-25 in Visually Normal Latinos: The Los Angeles Latino Eye Study Invest. Ophthalmol. Vis. Sci., April 1, 2003; 44(4): 1470 - 1478. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
The Submacular Surgery Trials Research Group Responsiveness of the National Eye Institute Visual Function Questionnaire to Changes in Visual Acuity: Findings in Patients With Subfoveal Choroidal Neovascularization--SST Report No. 1 Arch Ophthalmol, April 1, 2003; 121(4): 531 - 539. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
A. T. Broman, B. Munoz, J. Rodriguez, R. Sanchez, H. A. Quigley, R. Klein, R. Snyder, and S. K. West The Impact of Visual Impairment and Eye Disease on Vision-Related Quality of Life in a Mexican-American Population: Proyecto VER Invest. Ophthalmol. Vis. Sci., November 1, 2002; 43(11): 3393 - 3398. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
S. P. Azen, R. Varma, S. Preston-Martin, M. Ying-Lai, D. Globe, and S. Hahn Binocular Visual Acuity Summation and Inhibition in an Ocular Epidemiological Study: The Los Angeles Latino Eye Study Invest. Ophthalmol. Vis. Sci., June 1, 2002; 43(6): 1742 - 1748. [Abstract] [Full Text] [PDF]
|
|
|
|
|
|
M. M Brown and G. C Brown Outcome of corneal transplantation Br. J. Ophthalmol., January 1, 2002; 86(1): 2 - 3. [Full Text] [PDF]
|
|
|
|
|
|
K. D. Frick, B. M. Melia, R. R. Buhrmann, and S. K. West Trichiasis and Disability in a Trachoma-Endemic Area of Tanzania Arch Ophthalmol, December 1, 2001; 119(12): 1839 - 1844. [Abstract] [Full Text] [PDF]
|
|
| ||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||||
| HOME | HELP | FEEDBACK | SUBSCRIPTIONS | ARCHIVE | SEARCH | TABLE OF CONTENTS |
