Vision and Quality of Life Index: Validation of the Indian Version Using Rasch Analysis

Abstract

Purpose.: A multi-attribute utility instrument (MAUI) consists of a descriptive system in which the items and responses seek information about a concept of the universe of health-related quality of life (QoL), and responses to these items then are weighted and combined to produce the index. To our knowledge, the 6-item Vision and Quality of Life Index (VisQoL) is the only available vision-related MAUI, developed and validated in Australia, specifically for visually impaired (VI) populations. To our knowledge, the psychometric properties of the VisQoL have not yet been investigated in an Indian VI sample; this was the aim of our study.

Methods.: The Indian VisQoL was administered to 349 VI adults face-to-face by a trained interviewer at the Vision Rehabilitation Centres of a tertiary eye care facility, South India. Rasch analysis was used to assess the psychometric properties.

Results.: Rescoring was necessary for all except one item before ordered thresholds were obtained. All items fit the Rasch model and unidimensionality was confirmed. Person separation was acceptable (2.01), indicating that the instrument can discriminate among three strata of participants” vision-related QoL (VRQoL). The VisQoL items were targeted substantially to the participants” VRQoL (−0.69 logits). One item (“ability to have friendships”) demonstrated large differential item functioning by work status; working participants reported the item to be more difficult (−1.13 logits) relative to other items when compared to the nonworking participants.

Conclusions.: The 6-item Indian VisQoL satisfies unidimensional Rasch model expectations in VI patients. Disordering of response categories was evident; replication is required before a common rescoring option should be considered.

Introduction

More than 90% of the world's visually impaired population reside in developing countries, such as India. ^1,2 The prevalence of vision impairment (VI) is higher in the relatively older than younger age groups. ³ Over the years, a marked increase in life expectancy witnessed by developing countries suggests that the elderly would constitute almost three-quarters of the population in these regions, similar to that of developed countries. ^4,5 Given the increase in VI with age coupled with changing demographics, the prevalence of blindness and VI in India is predicted to rise with population ageing. ⁶ Vision impairment has detrimental effects, as it reduces a person's ability to perform activities of daily living; increases risk of depression, falls, and hip fractures; and impairs his/her quality of life (QoL). ^7–11 In addition to medical and surgical management of ocular conditions that cause VI, another goal of eye care services is to improve QoL of the affected people, since it is an outcome derived from patient perception. 

Assessment of QoL provides a holistic view of the impact of the disease process on a patient's life from the patient's perspective. Additionally, it helps in micro and macro level health care decision-making. ¹² One method of quantifying the impact on QoL associated with a particular state of health is with utility assessment, which forms an integral component in the estimation of quality-adjusted life years (QALYs) used in economic evaluation of health care interventions. In a general sense, utility refers to the preference an individual or a society may have for any particular set of health outcomes, and the use of multi-attribute utility instruments (MAUI) represents one of the methods of utility assessment. Using MAUI, the researcher either can obtain health-related utilities or disease-specific utilities (e.g. ophthalmic/vision-related) depending upon whether a generic or a disease-specific MAUI is used. Generic measures cover a range of dimensions of health status and are applicable to virtually all adult populations (they are not restricted to a particular area of health), so they facilitate comparison across different health conditions. This facilitates broad comparisons, and makes these measures especially well suited for cost-effectiveness analysis (CEA) and cost-utility analysis (CAU), including those that go across diseases to aid broad decisions regarding health service resource allocation. Some of the popular generic MAUI include the EuroQoL Health questionnaire (EQ-5D) ¹³ and the Health Utilities Index (HUI). ¹⁴  

Although generic health-related MAUI have an important advantage that they enable comparison of health-related QoL across different health states, they may, however, lack the sensitivity or responsiveness to particular diseases or health problems, for example, VI, thereby limiting their usefulness in the evaluation of vision-related programs. ^15,16 Disease-specific MAUI may ameliorate some of these potential limitations. The Vision and Quality of Life Index (VisQoL) was developed as a means to represent the vision-related MAUI and remains the only one to date to our knowledge. ¹⁷ Developed originally in English for a vision impaired Australian population, the developers have provided the utility scoring algorithm, too. ^18,19 The VisQoL has been shown to possess good psychometric properties using a combination of item response theory (IRT), exploratory factor analysis, and structural equation modeling in the Australian population. ¹⁷  

For the VisQoL to be used in multinational projects, further cross-national psychometric testing of the instrument is required. For economic reasons, it is more expensive to develop a questionnaire from scratch rather than to adapt an existing one. Thus, development of appropriate linguistic versions of the VisQoL for use in non-English speaking populations seems a more economic, efficient, and practical alternative. Such a rigorous validation process is critical in QoL research because the ramifications of utilizing linguistically or culturally inappropriate health outcome measures across various populations and cultures are far-reaching, not only in terms of decisions made on effective care, but also in terms of health policies that may be developed from research findings. Thus, accurate measurement across cultures is dependent on the proper linguistic and cultural adaptation, and application of an outcome measure for a specific population. ^20–22  

Translation of questionnaires is the most frequently chosen route to implementing “equivalent” questionnaires in cross-national and cross-lingual survey research. Two major reasons for using the same (translated) questionnaire in different countries include: (1) feasibility of analysis of results and a common international interpretation only if the data come from the same questionnaire, and (2) all new data acquired about a questionnaire contribute to validation and reputation of the questionnaire (especially relevant in the context of much-used questionnaires). ²³ Recently, the VisQoL was translated into German and was validated using Rasch analysis in the German population consisting of vision impaired and normally sighted patients. ²⁴ However, the German VisQoL required revisions to achieve an optimal fit to the Rasch measurement model. ²⁴ Rasch analysis, one of a number of psychometric techniques developed under IRT, is more useful because it provides greater insight into the measurement characteristics of an instrument (such as comprehensive understanding of the underlying latent structure) and is less sample dependent. ^25,26  

To our knowledge, an Indian version of the VisQoL has not been validated to date, and this is significant given that India is home to approximately one-fifth of the worlds' visually impaired population. ² Therefore, the aim of our study was to investigate the psychometric properties of the Indian VisQoL using Rasch analysis in a visually impaired sample in India. 

Methods

Participants

Participants were adults with VI referred to the Vision Rehabilitation Centres (VRC), L V Prasad Eye Institute (LVPEI), Hyderabad, India between December 2008 and September 2009. Participants were administered the VisQoL (described later in this section) at VRC in face-to-face interview in one of the three languages (English, Telugu, or Hindi as per the participant's convenience) by a trained researcher before any low vision examination, so as to ensure that the participant's responses were unbiased. The VisQoL was administered in a quiet room free of any distractions. 

For eligibility, participants had to be 18 years or older, English-speaking or speak one of the two local Indian languages (Hindi/Telugu), have no additional disabilities (such as physical, cognitive, or hearing), and have a duration of vision loss (self-reported) of ≥6 months. This duration of vision loss was chosen as a cut-off to ensure that all the participants had some experience of having lived with VI. Sociodemographic data were extracted from the clinical records. Data regarding patients' other demographic details, such as age; occupation; highest level of education attained; duration of VI; presence/absence of systemic comorbidities, such as diabetes, hypertension, arthritis, asthma, and coronary artery disease, were collected using a demographic data sheet during the interview. The cause of VI was obtained from the patients' clinical records. The study received approval from the Ethics committee of the LVPEI, and the research adhered to the tenets of the Declaration of Helsinki. Informed consent was obtained from the participants after a detailed explanation of the study. 

Vision and Quality of Life Index (VisQoL)

The VisQoL measures vision-related QoL (VRQoL), which can be translated into health states defined by the VisQoL responses. ¹⁸ The VisQoL originally was developed as part of the Assessment of Quality of life (AQoL instrument 7-D), as its seventh dimension and can be used either in conjunction with the AQoL or as a stand-alone instrument. ²⁷ The raw data of the VisQoL can be converted into vision-related utilities using the suggested SPSS algorithm provided by the developers on the website, available in the public domain at http://www.aqol.com.au/scoring-algorithms/138.html. However, we did not convert the raw VisQoL scores to utilities as our study was concerned solely with assessing the psychometric properties of the VisQoL. 

The VisQoL comprises six items from six domains: physical well-being, social well-being, emotional well-being, level of independence, self-actualization, and planning and organization (Table 1). ¹⁷ There are 5 to 6 response categories for each of these items and the response categories vary across these items; two items (item 2 and 5) have similar response categories (six). In addition, a “not applicable” option exists for items 3 and 4, which was considered as missing data for purposes of Rasch analysis. Higher VisQoL scores indicate worst possible VRQoL and lower scores indicate best possible VRQoL. 

Table 1

 

Item Content of the VisQoL

Item No.	Description of Health State Along With Response Categories
Q 1	Does my vision make it likely I will injure myself (i.e., when moving around the house, yard, neighborhood, or workplace)?
	(1) It is most unlikely I will injure myself because of my vision
	(2) There is a small chance
	(3) There is a good chance
	(4) It is very likely
	(5) Almost certainly my vision will cause me to injure myself
Q 2	Does my vision make it difficult to cope with the demands in my life?
	My vision:
	(1) Has no effect on my ability to cope with the demands in my life
	(2) Does not make it difficult at all to cope with the demands in my life
	(3) Makes it a little difficult to cope
	(4) Makes it moderately difficult to cope
	(5) Makes it very difficult to cope
	(6) Makes me unable to cope at all
Q 3	Does my vision affect my ability to have friendships?
	My vision:
	(1) Makes having friendships easier
	(2) Has no effect on my friendships
	(3) Makes friendships more difficult
	(4) Makes friendships a lot more difficult
	(5) Makes friendships extremely difficult
	(6) Makes me unable to have friendships
	(7) Not applicable; I have no friendships
Q 4	Do I have difficulty organizing any assistance I may need?
	(1) I have no difficulty organizing any assistance I may need
	(2) I have a little difficulty organizing assistance
	(3) I have moderate difficulty organizing assistance
	(4) I have a lot of difficulty organizing assistance
	(5) I am unable to organize assistance at all
	(6) Not applicable; I never need to organize assistance
Q 5	Does my vision make it difficult to fulfill the roles I would like to fulfill in life (e.g., family roles, work roles, community roles)?
	My vision:
	(1) Has no effect on my ability to fulfill these roles
	(2) Does not make it difficult to fulfill these roles
	(3) Makes it a little difficult to fulfill these roles
	(4) Makes it moderately difficult to fulfill these roles
	(5) Makes it very difficult to fulfill these roles
	(6) Means I am unable to fulfill these roles
Q 6	Does my vision affect my confidence to join in everyday activities?
	My vision:
	(1) Makes me more confident to join in everyday activities
	(2) Has no effect on my confidence to join in everyday activities
	(3) Makes me feel a little less confident
	(4) Makes me feel moderately less confident
	(5) Makes me feel a lot less confident
	(6) Makes me not confident at all

Translation of the VisQoL

The VisQoL translations followed the published guidelines. ^28,29 Rather than adapt and translate the VisQoL verbatim, we aimed for conceptual equivalence. Briefly, forward translations from the original English version into two local Indian languages (Telugu and Hindi) were performed by two bilingual experts. The reconciled versions then were compared to the original one for conceptual equivalence for each of the languages. The back translations were performed by another two bilingual experts. Finally, we conducted cognitive interviewing using a verbal probing technique to ensure patient comprehension of the new translated VisQoL. ³⁰  

Clinical Assessment

Habitual monocular visual acuity (VA) for each eye was measured on all participants using a high-contrast letter chart based on logMAR principles. Visual impairment (this includes low vision and blindness that have been described later) was defined as presenting VA in better eye <20/60 or a visual field (VF) <20° from the point of fixation in the better eye on automated perimetry (using Humphrey Automated Field analyzer, 24-2 Swedish Interactive Threshold Algorithm – Standard; Carl Zeiss Meditec, Inc., Dublin, CA). 

Statistical Analysis

All non–Rasch-related statistical analyses were performed using the Statistical Package for Social Sciences for Windows version 16.0 (SPSS, Inc., Chicago, IL). Characteristics of the study population were examined using proportions, means, medians, and standard deviations. Independent t-test was used to examine between-group differences in the Indian VisQoL score across various sociodemographic variables and level of VI. Sociodemographic variables included age (categorized by median), sex, number of years of education (categorized as median number of years <10 years and ≥10 years), work status (working, not working), and systemic comorbidity (present, absent). We classified the patients into two VI groups, low vision and blind, based on their presenting visual acuity (PVA). Low vision was defined as PVA <20/60 to 20/400 in the better eye and/or VF <20° from the point of fixation, and blindness was defined as PVA <20/400 in the better eye and/or a VF <10° from the point of fixation. A P value less than 0.05 was considered to be statistically significant. 

Rasch Analysis

Rasch analysis was used to determine whether the total VisQoL raw score was valid and reliable, and possessed the required measurement characteristics. If we found flaws in the VisQoL, we attempted to create a re-engineered version. Rasch analysis ³¹ was performed with Winsteps software (version 3.68.0; Winsteps, Chicago, IL) ³² using the Andrich rating scale model for polytomous data. ³³ A group analysis was used with one rating scale model per question format (five). This approach has been described previously. ²⁵ The unit of a Rasch-calibrated scale is the log-odds unit or logits. The logit is the natural logarithm of the odds of a participant being successful at a specific task or an item being performed successfully. For the participants, logit measures indicated whether one person is more able than the other (e.g., does one person have better VRQoL than another?); for items, logit measures indicated how easy or difficult it is to endorse an item compared to another (e.g., is it more difficult to fulfill roles than organize for assistance?); and for rating scale categories, logit measures indicated whether one rating scale category is greater or less than another (e.g., does a rating of 2 [does not/no effect/not difficult] consistently represent higher Qol than a rating of 3 [little difficult/less] in the VisQoL?). With respect to the VisQoL, positive item logit values represented those items that are less difficult to endorse, and positive logit values for participants represented those with lower VRQoL. On the other hand, negative item logit values represented those items that are more difficult to endorse, and negative logit values for participants represented those with higher VRQoL. The Rasch measurement model has been described elegantly by Massof. ³⁴ Various steps of Rasch analysis to assess instrument quality have been described in detail previously, ²⁵ so we provide the Rasch procedures only in brief here. 

We began with an assessment of the rating scale structure. This involved investigation of the category probability curves, specifically the category thresholds or the point on the logit scale of the QoL at which a participant is equally likely to choose between two adjacent categories, should be ordered. If thresholds were disordered, categories were combined to obtain ordered thresholds and this maneuver often can improve overall fit to the model. 

Second, the ability of the VisQoL to discriminate among participants was assessed using person separation statistics and the related reliability. ^35,36 Person separation indicated the number of distinct strata of participants that can be discerned by the VisQoL. Strata can be calculated using the formula: Number of Strata = (4 × person separation + 1)/3. There should be good overall separation, as measured in logits: the greater the person separation, the more distinct strata (i.e., groups) are identified. Reliability is the ratio of the “true” measure variance to the “observed” measure variance. A person separation of 2.0 was considered minimum acceptable (reliability value of ≥0.80) and corresponded to the ability to differentiate among three strata of a trait. ³⁷  

Third, item fit mean square statistics (infit MnSq) was used to determine if individual items contributed towards the measurement of the underlying trait. When the data fit well they indicated that the items contributed to a single underlying construct (unidimensionality). Infit MnSq is a measure of deviation from expectation and represents the observed variance divided by expected variance; therefore, the desired value of MnSq is 1.0. Values outside the range of 0.7 to 1.3 were used to identify misfitting items, and such items must be examined and revised or eliminated. ³⁸ Unidimensionality was assessed further using principal components analysis (PCA) of the residuals. By examining the structure of the residuals in the PCA, one can examine whether there is evidence of additional component captured by the items. ³⁹ In the PCA, we used the following two rules of thumb to confirm unidimensionality: a cutoff of 60% of the variance explained by the Rasch factor ³⁵ and unexplained variance explained by the first contrast with an eigenvalues smaller than 2. ⁴⁰  

Fourth, we inspected the person-item map to determine if the item difficulty matched with the participant's QoL (i.e., targeting). For a well-targeted instrument, the mean item difficulty (this usually is set at zero) would match mean person ability of the population; a difference between the means of >1.0 logits indicates notable mistargeting. ⁴¹  

Finally, we assessed differential item functioning or DIF (indicator of the presence of any systematic differences between the various item difficulty estimates across subgroups of participants). ⁴² We selected the variables for DIF analysis a priori and these included age (<43 vs. ≥43 years; median age, 43), sex, number of years of education (<10 vs. ≥10 years), work status (working versus not working), systemic comorbidity (present versus absent), and VI group (low vision versus blind). Testing for DIF can occur based on either significance or magnitude. Because significance testing is highly sample size–dependent, we prefer testing for DIF magnitude. ²⁷ Therefore, in our study, we defined DIF based on magnitude and considered it relevant if it was large (>1.00 logit). ²⁸  

Given that we used three language versions of the VisQoL, we examined the cross-cultural/linguistic validation of the instrument through an assessment of DIF, and local response dependency (dependent items will form their own “dimension”) in our population. Each participant was administered the instrument in only one language, so we used the nested study design that provides DIF for item effects within language group (and not across language group) in DIF testing. Overall differences in language difficulty will be shown by differences in average ability for the participants being tested in each language and we used the independent sample t-test to investigate this difference. Statistical significance was set at P < 0.05. 

Results

Sociodemographic and Clinical Characteristics

All 349 patients with VI enrolled responded to the VisQoL. A little over half of the participants (53%) were administered the VisQoL in Telugu, 32% in Hindi, and 15% in English. Sociodemographic and clinical characteristics of the 349 participants are shown in Table 2. The mean (±SD) age was 43.4 ± 17.8 years (range, 18–84 years) and 77% were male. Most of the participants were working (68%). The mean ± SD presenting distance VA was 0.83 (Snellen equivalent, 20/125⁻²) ± 0.43 logMAR (range, 20/20 to light perception). The majority of the participants had low vision (76%); 24 (6.8%) had VF loss <20° in the better eye (with PVA ≥20/60 in the better eye). 

Table 2

 

Sociodemographic and Clinical Characteristics of the Study Participants (n = 349) Who Responded to the VisQoL

Participant Characteristic	n (%)	Vision and Quality of Life Index Overall Score,* Mean ± SD	P
All participants	349 (100)	−0.69 ± 1.53	–
Mean Age, y
<43	184 (53)	−0.79 ± 1.63	0.71
≥43	165 (47)	−0.86 ± 1.83
Sex	267 (77)
Male		−0.79 ± 1.77	0.46
Female	82 (23)	−0.95 ± 1.57
Marital status
Presently married	227 (65)	−0.90 ± 1.78	0.25
Never married/widowed	122 (35)	−0.68 ± 1.62
Location
Urban	242 (69)	−0.92 ± 1.77	0.30
Rural	107 (31)	−0.67 ± 1.76
Median y of education†
<10	115 (33)	−0.79 ± 1.74	0.84
≥10	216 (62)	−0.83 ± 1.71
Systemic co-morbidity‡
Absent	237 (68)	−0.72 ± 1.69	0.09
Present	112 (32)	−1.06 ± 1.79
Work status
Working	236 (68)	−0.86 ± 1.62	0.60
Not working	113 (32)	−0.75 ± 1.97
Visual impairment group§
Low vision	264 (76)	−1.11 ± 1.60	<0.0001¶
Blind	85 (24)	−0.06 ± 1.83

*  VisQoL overall score in logits (log-odd units; see text for details). Higher negative score indicates a better vision-related QoL.

†  Data incomplete.

‡  Includes diabetes mellitus, hypertension, asthma, coronary artery disease, arthritis.

§  Low vision defined as presenting VA <20/60 to 20/400 in the better eye and/or visual field <20° from the point of fixation, and blindness defined as presenting VA <20/400 in the better eye and/or a visual field <10° from the point of fixation.

¶  Value of P indicates statistical significance.

Psychometric Validation of the Indian VisQoL

Except for one rating scale (No. 1), the 6-item Indian VisQoL displayed ordered thresholds for the remaining (four) rating scales, necessitating category reorganization. 

The response categories were intended to cover a range of VRQoL, whereby each category should be the most likely to be chosen for part of this range representing stepwise increase in severity. Nonetheless, this was not the case. As an illustrative example, Figure 1 depicts the category probability curves for rating scale 2 that shows the range of VRQoL for which each of the six response categories were most likely to be chosen. Category 4, “moderately difficult,” is not the most likely category to be endorsed at any level of VRQoL. This is described as disordered thresholds (Fig. 1A) because the threshold between categories 3 and 4 lies to the right (instead of the left) of the threshold between categories 4 and 5. Therefore, we combined category 4 “moderately difficult” with category 5 “very difficult” (123445), resulting in a new category (“somewhat difficult”) and following this, fairly equal widths over which each category was the most likely response, as is desirable, was evident, resulting in ordered thresholds (Fig. 1B). Likewise category reorganization was done for the rating scales 3 to 5. For rating scale 3, a five-category response structure in which the categories “more difficult” and “a lot more difficult” were combined, which resulted in ordered thresholds (123345). For rating scale 4, a five-category response structure in which the categories “a lot of difficulty” and “unable to organize assistance at all” were combined initially, but disordering persisted. Subsequently, a four-category response structure in which the categories “no difficulty” and “a little difficulty” were combined resulted in ordered thresholds (11223). For rating scale 5, a five-category response structure in which the categories “a little less” and “moderately less” were combined, which resulted in ordered thresholds (123345). 

Figure 1

 

Rasch model category probability curves for the original six-category structure of rating scale two (items 2 and 5) in the 6-item VisQoL showing the likelihood that a participant with a particular level of VRQoL will select a category. The scale (x-axis) symbolizes the latent trait of VRQoL. The y-axis represents the probability of category being selected. Response categories: 1, “no effect;” 2, “does not make it difficult;” 3, “little difficult;” 4, “moderately difficult;” 5, “very difficult;” and 6, “unable.” For any given point along this scale, the category most likely to be chosen by a participant is shown by the category curve with the highest probability. At no point was category 4 the most likely to be chosen, resulting in disordered thresholds (A). However, after combining category 4 with 5 (4 + 5 resulting in a new category that we refer to as “somewhat difficult'), the revised five-category structure shows ordered thresholds (B).

The person separation index was 2.01 (person separation reliability [PSR] = 0.80), which indicates that the 6-item Indian VisQoL can discriminate adequately three strata of participants' VRQoL. All items fit the model well (infit MnSq, 0.86–1.26), and PCA of the residuals showed that the variance explained by the measures was comparable for the empirical calculation (59.6%) and by the model (57.9%). The unexplained variance by the first contrast was 1.6 eigenvalue units. Furthermore, there was no evidence of a second dimension (i.e., there were no dependent items that formed their own “dimension”) so local response dependency was absent. Taken together, these findings suggested that the VisQoL was unidimensional and that it was linguistically valid across the three languages. Targeting was −0.69 logits, indicating that the item difficulty matched well with the participants' VRQoL, and this is evident by the close proximity of the item and person means (distance < 1 logit) in the person-item map (Fig. 2). The overall mean participant score of the VisQoL was −0.69 ± 1.53 logits (range, −4.37–4.35 logits). 

Figure 2

 

Person-item map for the 6-item Indian VisQoL in visually impaired patients. The vertical line represents the measure of the VRQoL in logit units. Participants are located on the left of the vertical line, and participants with higher VRQoL are located at the bottom of the map. Items are on the right of the vertical line, with more difficult-to-endorse items located at the bottom of the map. Item names have been abbreviated to fit the space and the correct description of items can be found in Table 1. Each “x” represents 3 participants and each “.” represents one to three participants. By convention, the mean item endorsability is set at 0 logits (indicated with M'). Accordingly, mean VRQoL of participants is indicated with M. M, mean; S, 1 SD from the mean; T, 2 SD from the mean.

Only one item displayed large DIF by work status. Working participants reported item No. 3 (“ability to have friendships”) to be more difficult (−1.13 logits) relative to other items when compared to the nonworking participants. The mean participant abilities of those who were administered the Telugu version of the VisQoL was −0.76 ± 1.70 vs. −0.90 ± 1.75 logits for English/Hindi versions; this difference was not statistically significant (independent sample t-test, t = 0.76, P = 0.45), implying that there was no mean difference in person abilities and item difficulties across the languages. 

VisQoL Score and Participants' Sociodemographic and Clinical Characteristics

Except for VI group, the VisQoL score was not significantly different across subgroups stratified by various sociodemographic variables (Table 2). Those with low vision had statistically significant better VisQoL scores as compared to those blind (−1.11 ± 1.60 vs. 0.06 ± 1.83 logits, P < 0.0001). 

Raw Score to Rasch Measure Conversion

While it will be ideal for users of the Indian VisQoL to perform Rasch analysis of their own data, clinicians/researchers unfamiliar with Rasch analysis still may wish to use its scoring benefits. We have provided ready-to-use spread sheets that convert raw scores entered to Rasch-scaled scores for the Indian VisQoL. These sheets can be downloaded directly from the journal's website or obtained by contacting the corresponding author. 

Discussion

Rasch analysis of the Indian VisQoL in our visually impaired population confirmed the good psychometric properties, including the critical property of unidimensionality, of the original VisQoL. Two important consequences of this are that firstly, the use of a summary or total VisQoL score is justified and secondly, raw VisQoL scores can be transformed easily into interval-scale estimates. Consequently, researchers intending to calculate change scores or use parametric statistics can be confident that the transformed data satisfy this criterion. ⁴³ Clinicians and researchers will be able to use the transformed scores to specify not only that a patient's VRQoL has improved (or worsened), but by how much their VRQoL has changed on a properly constructed interval scale. 

Although the Indian VisQoL met the requirements of the Rasch model, it is interesting to note that the rating scale structure required revisions. The response categories proposed for the original VisQoL (English version) were found unsuitable for use in the Indian visually impaired population. Nevertheless, disordered thresholds of the Indian VisQoL categories have been reported for the German VisQoL too, ²⁴ indicating that this finding is not a peculiarity of our study. Our participants found the distinction between “a little” and “moderate,” and “more difficult” and “a lot more difficult” ambiguous. Consequently, Rasch analysis demonstrated the need for post hoc category reorganization for five of the six Indian VisQoL items to achieve optimal functioning of the rating scale. Issues with response categories can occur when the labeling of the response options is ambiguous or too many response options have been included. Given that due vigilance, such as consideration of patients' perspectives, and use of a mixture of classical and modern test theory, was employed in ensuring the appropriateness of the content during the developmental phases of the VisQoL, it is unlikely that the six category response options are ambiguous; rather, it is possible that the VisQoL has too many response options. As noted in the Methods section, most of the items had six response categories, but it has been found that participants typically tend to use only four or five categories. ^44,45 In line with this observation, we found a four-to-five category revised rating scale structure to be adequate across all the items for the Indian VisQoL as compared to a three-to-four category structure reported to be optimal in the German VisQoL. ²⁴ Our findings of category reorganization may be a function of the distinct study population and not the measure itself. The need for such post hoc category revision has been reported for other instruments that have been revalidated using Rasch analysis, for example, visual disability instruments, which have found that the participants are not always able to distinguish between finer increments (e.g., between “more difficult” and “a lot more difficult”) in response options. ^25,46 Although category reorganization reduces the comparability with the original and other language versions of the VisQoL data, it may, however, be an appropriate procedure in patients expected to have lower VRQoL (e.g., those with severe visual impairment/blindness). Nonetheless, we still would caution that the revised rating scale structure of the Indian VisQoL should be tested in a new sample before being acceptable for use in future studies. 

Among the desirable features of an optimally functioning instrument include an ability to discriminate as many strata or groups of participant QoL as possible, simulating the gradations on a ruler; the finer the gradations, the better the measurement properties. ^30,31 The Indian VisQoL possessed adequate measurement precision in that it was able to discriminate reliably among three distinct strata of participants' VRQoL, that is, good, intermediate, and poor. For a valid measurement, in addition to acceptable measurement precision, it must possess certain attributes, including adequate spread along the dimension of measurement, and negligible floor and ceiling effects. ⁴⁷ The Indian VisQoL demonstrated good targeting (−0.69 logits) of the item endorsability to the participants' VRQoL in our patient population. Given that the instrument was developed for the visually impaired population (54% had VA <20/60 in better eye in the development study), albeit Australian, ¹⁷ it can be expected that the item content would suit the population under consideration. By comparison, however, the German VisQoL demonstrated suboptimal targeting of person ability to item difficulty, which possibly can be explained by the fact that the investigators included normally sighted (80%) and visually impaired patients (20%). ²⁴ Given that targeting is sample dependent, it is important to consider differences in the participants when assessing the targeting of instruments. 

Interestingly, we did not find any item to misfit in the Indian VisQoL. This finding is at variance with that reported for the German VisQoL in which one item (item No. 4 [“difficulty organizing any assistance that I may need”]) demonstrated considerable misfit (infit MnSq 2.21) in the German visually impaired population. The investigators speculated that the misfit was perhaps due to the ambiguous phrasing with the result that participants were unclear if the task pertains to organizing per se, the need for assistance or the level of assistance required. Given that the investigators conducted pilot testing of the final German translated VisQoL in focus group of patients resulting in several changes of the wording of the items, the need for further revisions in the wording of item No. 4 as has been suggested by them is surprising. Although we also adopted standard procedures for translation of the VisQoL into Indian languages, we did not encounter similar problems for the Indian VisQoL. While there appears to be no obvious explanation for this difference, we perhaps can speculate on one reason as follows. Most of our patients (70%) reported that they had “no difficulty” with organizing assistance and one of the reasons for such low/no difficulty may be related to ease with which assistance is available, if required, especially when someone in the immediate family has a disability, in India. Therefore, the patients perhaps could relate easily to this item and respond appropriately. By comparison, family support systems may not be so readily available in the Western countries, such as Germany. Given this, the German patients may have had difficulty relating to this item and may have required further explanation/clarifications to respond to it. 

We found some evidence for the presence of DIF, albeit for a single item (item No. 3). This item (“ability to have friendships”) demonstrated large DIF (>1.0 logit; −1.13 logits), although marginal, by work status. Participants who were “not working” reported greater difficulty with this item relative to other items. While the comparison of scores for this subgroup of participants in the Indian VisQoL may not be appropriate under such circumstances, the marginally large DIF in just one item is unlikely to affect the measurements materially. While the reason for DIF of item No. 3 is not entirely clear, one explanation for DIF by work status may be that the response of those “not working” to this item was influenced perhaps by the confinement to their homes because of their inability to socialize (difficulty with skills, such as recognize familiar faces while walking, use mobile phone to keep in touch with friends) given their VI. Therefore, it was likely more difficult for them to “have friendships” compared to those “working.” While DIF can be expected on account of change in the language and the presence of cultural differences in translated instruments, ⁴² we did not find a mean difference in the person abilities and item difficulties across languages, implying that there was no overall difference in language difficulty for the VisQoL. However, it is possible that person abilities and item difficulties are lower in one language but we would need another “crossed” study (where each participant is administered the instrument in the three languages) to eliminate this possibility. 

Not surprisingly, patients with low vision in our study had significantly higher Indian VisQoL scores (better QoL) when compared to those with blindness given that consistent decline of vision-related QoL with worsening VI has been reported in a similar population ⁴⁸ and Western cohorts. ^49–51  

Quality of life has been known to vary across countries. ^52–54 Some of the reasons for this variation include differences in lifestyles, economic status, health care system, cultural value system, and health policy. Cross-cultural or cross-national comparisons would help to understand these factors, but to our knowledge there are no data concerning the possible effect of cross-cultural differences on the VRQoL among Western and Asian VI populations. Since cross-cultural differences in QoL might influence the outcome of CEA, it is important to evaluate whether the association of VRQoL and VI is confirmed in the Asian (Indian) context. For cross-cultural comparisons, it would be straightforward to use a single questionnaire that transcends cultural and linguistic barriers; however, researchers who are confronted with a linguistically diverse population (as in our study) seem to have little choice but to accept the cost and inconvenience of questionnaire translation. Additionally, it may not be practical to develop a version of the questionnaire with universal appeal. Nonetheless, the measurement problems arising from cross-cultural adaptation and validation of questionnaires that may result in slightly different versions in respective populations, as is the case with the VisQoL (German and Indian), may be overcome by equating the versions (i.e., scale equivalence) as has been done recently by us for the National Eye Institute – Vision Function Questionnaire and the Indian Vision Function Questionnaire. ⁵⁵ The availability of English, German, and Indian versions of the VisQoL provides an opportunity to compare the VRQoL between Western and Asian countries that are significantly different in culture and socioeconomic environment. ⁵⁶ Making cross-cultural comparisons from separate studies that have been conducted in single countries (such as in Germany and India) is complicated due to variation in participant groups studied, and differences in research methodology and procedures used. However, cross-cultural comparisons of the VRQoL in the same participant group and using the same methods help to overcome these difficulties. This is possible through international collaborative efforts, and we intend to pursue this in our future research. 

As stated earlier, a major application of MAUI is a quality-adjustment weights in various types of analyses designed to assist in the formulation of policy decisions concerning resource allocations. Generic MAUIs, such as HUI3 and EQ-5D, have been included in major United States, Canadian, and United Kingdom population health surveys, providing abundant data on population norms. All the MAUIs have their benefits and challenges in different settings, and they enjoy varying degrees of acceptability by health authorities. However, there is considerable controversy about the merits of using a disease-specific, such as vision-specific, as opposed to a generic MAUI. For example, instrument recommendations made by the government agencies, such as the National Institute for Health and Clinical Excellence (NICE) in the United Kingdom prefers a generic MAUI - the EQ-5D, so as to enable comparability of interventions across different technologies and patient groups. ⁵⁷ NICE will not consider disease-specific MAUIs when setting reimbursement policy, because they do not allow easy comparisons of the health economics across diseases. However, NICE has acknowledged that the EQ-5D may not be an appropriate measure of health-related utility in all circumstances. ⁵⁷ Given the coarseness of the generic MAUIs, there is reason to suspect that important aspects of HRQoL may not be captured by these measures. Therefore, a number of guidelines for the assessment of healthcare technologies recommend the concurrent use of disease-specific MAUI alongside generic MAUIs, such that more targeted coverage of HRQoL aspects important in that context is ensured. ^58,59  

The main strengths of our study include a relatively large sample size, the use of standard accepted procedures in translation of the VisQoL for use in India, and use of a modern psychometric approach, Rasch analysis, to validate the Indian VisQoL. 

However, there are some limitations that must be acknowledged in the context of this study. First, the present research was confined to a single sample from a specialist rehabilitation service in South India dealing mostly with the moderate to severely visually impaired. Thus, the findings may not be generalizable to populations from other settings, such as the community setting. It would be useful to apply Rasch analysis to Indian VisQoL responses from other visually impaired samples perhaps representing more varied levels of VI. It would be particularly informative to see whether disordered thresholds are a common problem across these samples. If that were the case, then an argument can be made for rescoring the Indian VisQoL for visually impaired patients as reported in our study. In the meantime, users of the Indian VisQoL should continue to use the original scoring system. Second, there was a male preponderance in our study, which is in accordance with the slightly skewed distribution among sex presenting to our center. ⁶⁰ However, this interferes with the ability to reach conclusions confidently about the QoL of women with VI. Finally, given that we translated the VisQoL into two local languages, the validity of VisQoL in other major Indian languages remains unknown. While in most developed countries, the cross-cultural adaptation and validation of an outcome measure typically is reported for one predominant language, the implementation of an outcome measure in one language is not feasible in a country like India, where there is vast linguistic and cultural diversity, ⁶¹ A total of 22 official languages are spoken throughout India, with the dominance of each language varying between different parts of the country. For the present study, we developed two language versions of the VisQoL—Hindi and Telugu. Hindi is arguably the most widespread language of India, and is spoken by approximately 50% of the population, so we believe that the Hindi version of the VisQoL will be widely applicable in India. However, the Telugu versions were developed as it is the native language of the state of Andhra Pradesh (population of approximately 90 million) and 53% of our participants used this version. Although the validity of the VisQoL could be tested in other Indian languages, the process could be rendered less cumbersome by using adjusted linguistic validation process (the forward and backward translation steps are replaced by an adaptation step, where the work is based on a version considered as the “mother language” version) for languages that are almost similar. The cross cultural adaptation of existing instruments, although a reasonably quick and practical means of obtaining instruments in other languages, has several drawbacks when it comes to obtaining cross culturally equivalent measures. ⁶² To avoid these problems, several investigators have recommended the simultaneous development of instruments in different countries. ⁶³  

In conclusion, to our knowledge our study is the first to examine the psychometric properties of the Indian VisQoL in a large Indian visually impaired sample by using Rasch analysis. We confirmed that the Indian language versions of the instrument satisfy the requirements of Rasch measurement model in its current 6-item format. However, some disordering of response categories was evident, but this evidence requires replication before any common rescoring option of the Indian VisQoL should be considered. Further studies should include an analysis of the actual performance of the new response structure, because collapsing 6 categories to 4 or 5 is not the same as presenting 4 or 5 categories to the participants. 

Acknowledgments

Supported in part by the Hyderabad Eye Research Foundation, Hyderabad, India. The authors alone are responsible for the content and writing of the paper. 

Disclosure: V.K. Gothwal, None; D.K. Bagga, None