Assessment of the Impact of Keratoconus on Vision-Related Quality of Life

Abstract

Purpose.: We determined if the Impact of Vision Impairment (IVI) is a valid questionnaire to measure the vision-related quality of life (VRQoL) in keratoconus patients, and investigated if the VRQoL varied with disease severity in this sample.

Methods.: We recruited 160 consecutive bilateral keratoconus patients (mean age 23.3 years, 63% male) from an Indian tertiary eye care center. Participants underwent a clinical examination and completed the IVI. Participants were divided into three groups based on the average of the steep keratometry (K) readings: mild (average Sim K < 45 diopters [D]), moderate (average Sim K 45–52 D), and severe (average Sim K > 52 D). Rasch analysis was used to validate the IVI and the VRQoL scores thus obtained were compared across the disease groups.

Results.: The majority (63%) of patients had severe, while the remainder (37%) had moderate keratoconus. Rasch analysis demonstrated the validity of the IVI to assess VRQoL through two subscales: vision-specific functioning (VF) and emotional well-being (EWB). There was no significant difference in VF (mean change −0.16, P = 0.55) and EWB scores (mean change −0.32, P = 0.23) between moderate and severe keratoconus groups.

Conclusions.: The revised IVI subscales have interval-level measurement properties, which support their suitability to measure VRQoL in this keratoconus sample. Patients with moderate or severe keratoconus had similar, but higher VRQoL scores as assessed by the revised IVI subscales, indicating lack of impact of the disease on their VRQoL. However, this does not exclude the possibility of finding an impact in other populations.

Introduction

Keratoconus is a noninflammatory ectatic corneal disorder characterized by anterior corneal protrusion, thinning of the central stroma, and irregular astigmatism. ¹ Although typically bilateral (B/L), the affliction is markedly asymmetric. ^2,3 It often occurs at puberty ^1,4,5 and progresses relentlessly before stabilizing by 30 to 40 years of age. ⁴ With progression, patients experience increased blurring and visual distortions resulting from higher order aberrations. ⁶ In moderate keratoconus, such visual disturbance can interfere with daily tasks, such as recognizing faces across a room, driving, watching TV, reading small print and labels or instructions on medicine, reading street signs, outdoor mobility in daytime (due to glare), and so forth. Additionally, affected patients must live with the fear and anxiety of having to undergo surgery, such as keratoplasty, at some stage in their life, ⁴ and concern that it may progress to a severe stage and possibly lead to visual impairment (VI). ⁷ In severe cases, VI not only impacts performance of daily activities, but also adversely affects emotional well-being (EWB ^{8 ,9} ) leading to depression, ¹⁰ and poses role difficulties. Taken together, this suggests that, although not a blinding disorder in the classical sense, ^11,12 the unpredictable nature of visual sequelae and visual disturbance from keratoconus can result in a disproportionate impact on visual function and on EWB, which are not likely reflected by traditional clinical measures of vision, such as visual acuity (VA). ^12,13  

Increasingly, patient-reported outcomes are being used to assess the impact of ocular conditions on vision-related quality of life (VRQoL). ¹⁴ The construct of VRQoL encompasses vision functioning (VF), symptoms, EWB, social relationships, concerns, and convenience as they are affected by vision. ¹⁵ Given that it is based on patient's perspective, it enables more meaningful and detailed information to be obtained. Available literature, albeit from the developed countries including the report by the Collaborative Longitudinal Evaluation of Keratoconus (CLEK) study group, ¹² suggests keratoconus to have a detrimental impact on VRQoL that worsened over time in a significant number of patients. ^12,16,17 While there are several cohort studies of keratoconus from developing countries, such as India, ^18,19 there is a paucity of studies on the VRQoL of these patients. It seems reasonable to determine their VRQOL, since the characteristics of keratoconus in Asians varies from those of other ethnic groups. For example, the incidence of keratoconus patients presenting to a hospital in the United Kingdom was 25/100,000 Asians per year, compared to 3.3/100,000 white patients, a ratio of 7.5:1. ²⁰ Studies also have shown that Asian (i.e., India^18,19) or Middle-Eastern keratoconus patients ²¹ tend to be younger at diagnosis, and present with severe forms of the disease. Taken together, these indicate that Asians from the Indian subcontinent may have a higher incidence, are affected at an earlier age, and suffer more advanced disease when compared to white populations. ^20,22–24 Given these ethnic differences, the results of VRQoL studies from developed countries cannot be extrapolated to Indian keratoconus patients. Therefore, it is important to examine independently the VRQoL of keratoconus patients in India, and identify the factors affecting their VRQOL, since these findings may influence treatment outcomes. Furthermore, such information may prove beneficial to clinicians given that it provides the patients' perspectives, which can be integrated into the decision-making process for any treatment being planned. 

The Impact of Vision Impairment Questionnaire (IVI) is a vision-specific QoL instrument developed to measure participation in daily living in patients with VI. ^25–29 It has been validated extensively using modern psychometric theory ^25–28 and is recommended by the American Academy of Ophthalmology to assess the impact of VI on QoL. ³⁰ Therefore, we used the IVI to investigate the VRQoL of keratoconus patients in India. Specifically, we first determined the psychometric properties of the IVI in Indian keratoconus patients using Rasch analysis. Second, we investigated if the VRQoL differed with disease severity. 

Methods

Participants

Participants were B/L keratoconus patients (both newly-diagnosed and follow-up) of the L V Prasad Eye Institute (LVPEI), Hyderabad, India. Few potential participants (N = 6) refused to participate for lack of time. Although this number was too small, a formal analysis revealed that their demographic characteristics did not differ substantially from those included. 

Included participants were aged ≥18 years, spoke English, Telugu, or Hindi, and had no severe cognitive impairment. Patients with unilateral or suspected keratoconus, and also those with concomitant ocular diseases that could affect vision, or a history of recent graft rejection (<6 months) following penetrating keratoplasty (PK) were excluded. Participants were provided verbal instructions by research assistants before filling out the questionnaire. While a majority (N = 113, 71%) self-administered the questionnaire, trained interviewers administered it face-to-face in a quiet room to the remaining participants. Interviewers attended a half-day training workshop conducted by experienced interviewers from LVPEI during which they were familiarized with the principles of questionnaire administration and supervised in the conduct of practice interviews before the start of the study. 

Included patients wore spectacles, contact lenses (CLs), or CLs plus spectacles (i.e., residual correction in the form of spectacles over CLs). Additionally, some patients had undergone PK > 1 year ago in one or both eyes for keratoconus. Participants were divided into three groups based on the average of the steep keratometry (K) readings: mild (average Sim K < 45 diopters [D]), moderate (average Sim K 45–52 D), and severe (average Sim K > 52 D). ^5,31  

Ethical approval was obtained, and all patients who agreed to participate signed a consent form. The study was conducted in accordance with the tenets of the Declaration of Helsinki. 

Clinical Assessments

Clinical assessments consisted of VA, anterior segment biomicroscopy, refraction, and corneal topography. Presenting VA (PVA) was recorded monocularly and binocularly using high-contrast letter charts based on LogMAR principles with chart luminance of 150 cd/m². Binocular PVA was recorded as this is considered to be representative of real world ability. ^32,33 Corneal topographic measurements were obtained using Orbscan IIz (Bausch and Lomb Surgical, Orbtek Inc., Salt Lake City, UT). Demographic data were recorded from the patient's medical records. 

IVI Questionnaire

The IVI was designed originally to consist of 32 items and was validated using classical test theory. ²⁹ Subsequently, it was revalidated using Rasch analysis to result in a 28-item version, a more robust and psychometrically sound questionnaire. ²⁷  

We chose to use the 32-item IVI (Table 1) for a few reasons: The 32-item IVI was revalidated using Rasch analysis in patients with VI. Firstly, 4 items were deleted but these (“paid or voluntary work,” “favorite pastimes or hobbies,” “going out to sports/events/movies/plays,” and “reading a sign across the street”) items may be relevant to a relatively younger and less VI population, such as keratoconus. However, this has not been investigated as yet to our knowledge. Secondly, the rating scale was shortened from six to four following Rasch analysis in the 28-item IVI. Specifically, one of the lower response categories (“hardly at all”) was eliminated, and this is not surprising given that those with VI would be expected to choose the higher end of the categories more often than the lower end. It can be hypothesized that the use of categories may follow a different pattern in a less VI sample, such as keratoconus, but this hasn't been tested as yet. Therefore, it seems logical to use the original 32-item IVI compared to the revised 28-item version to test these hypotheses. 

Table 1. 

 

Item Content and Scales of the 32-Item IVI Questionnaire

Items—Native Version	Summary Scales—New Version
01. Paid or voluntary work	Vision-specific functioning scale
02. Favorite pastimes or hobbies
03. Your ability to see and enjoy TV?
04. Taking part in recreational activities, such as bowling, walking, or golf?
05. Going out to sports events, movies, or plays?
06. Shopping? (finding what you want and paying for it)
07. Reading ordinary size print? (e.g., newspapers)
08. Visiting friends or family?
09. Recognizing or meeting people?
10. Getting information that you need?
11. Generally looking after your appearance? (face, hair, clothing, and so forth)
12. Opening packaging? (e.g., around food, medicines)
13. Reading labels or instructions on medicines?
14. Operating household appliances and the telephone?
15. Reading a sign across the street?
16. Getting about outdoors? (on the pavement or crossing the street)
17. In the past month, how often has your eyesight made you go carefully to avoid falling or tripping?
18. In general, how much has your eyesight interfered with traveling or using transport? (bus & train)
19. Going down steps, stairs, or curbs?
20. Your general safety at home?	Emotional well-being scale
21. Spilling or breaking things?
22. Your general safety when out of your home?
23. In the past month, how often has your eyesight stopped you doing the things you want to do?
24. In the past month, how often have you needed help from other people because of your eyesight?
25. Have you felt embarrassed because of your eyesight?
26. Have you felt frustrated or annoyed because of your eyesight?
27. Have you felt lonely or isolated because of your eyesight?
28. Have you felt sad or low because of your eyesight?
29. In the past month, how often you worried about your eyesight getting worse?
30. In the past month, how much has your eyesight made you concerned or worried about coping with everyday life?
31. Have you felt like a nuisance or a burden because of your eyesight?
32. In the past month, how much has your eyesight interfered with your life in general?

The questionnaire responses were scored as recommended by the developers. ²⁹ It uses two rating scales: one for items 1 to 19 and another for items 20 to 32. Using standard translation procedures, local language versions were obtained and participants were administered the questionnaire in one of 3 languages convenient to them. Higher IVI scores indicate worse VRQoL. 

Rasch Analysis

Rasch analysis ³⁴ was conducted using the Andrich rating scale model ³⁵ with Winsteps software (version 3.68, available in the public domain at http://www.winsteps.com/index.htm). ³⁶ Given that the 32-item IVI uses two different ratings scales, we used the two Andrich rating scale model (one for items 1–19 and another for items 20–32). This approach has been described previously. ^37,38 Rasch analysis is an iterative procedure that estimates interval measurement from ordinal data and the unit is logits (log-odd units). ^38–40 For our study, a negative item logit indicates a more difficult item and a negative logit value for a participant indicates that the he/she possesses a higher level of the assessed latent construct (VRQoL); that is, better VRQoL. The Rasch procedures have been described in detail previously in this journal. ^41,42 We used six fundamental indicators to assess the validity of the IVI ⁴³ : (1) Behavior of the rating scale was defined specifically as an examination of category thresholds (threshold is the midpoint between adjacent response categories and indicates the point where the likelihood of choosing either response category is equal). (2) Item fit (extent to which use of a particular item is consistent with the way the participants have responded to other items) to the Rasch model was assessed using the weighted mean square (MnSq) or infit statistic. The infit MnSq is less sensitive to distortion from outliers, so is considered the more informative fit statistic, ⁴⁴ and is the ratio of the observed variance of the residuals to the variance explained by the Rasch model. It has an expected value of 1.0 (range from 0 to infinity). Deviations in excess of the expected value may be interpreted as “noise” or lack of fit between the items and the model. We reported fit statistics as mean square standardized residuals (MNSQ) and used a criterion of 0.7 to 1.3 for Infit MnSq to diagnose misfitting items. ⁴⁵ Any misfitting item (fit > 1.30 indicated 30% more variance than expected and, thus, suggested that the item measures a construct different from the overall scale) was removed and Rasch analysis rerun, and this iterative process was continued until no further misfit was observed. (3) Measurement precision was represented by person separation index (PSI, minimum acceptable value of 2.0) and associated reliability; that is, person separation reliability (PSR, minimum acceptable value of 0.8). (4) Targeting was the extent to which the items match participant's VRQoL, and was inspected using the person-item map; >1.0 logits indicates notable mistargeting. ³⁷ (5) Unidimensionality was defined as the extent to which all the items measure a single underlying construct (VRQoL in the case of IVI) assessed by principal components analysis (PCA). The rationale for this is that after the “Rasch factor” has been extracted (in an attempt to account for all variation in the data) only standardized residuals equivalent to random noise should remain. A high level of variance accounted for by the principal component suggests a lower chance of finding additional components and a variance of ≥60% is considered to be good. Also, if the variance explained by the principal component for empirical data (variance components for observed data) and for the Rasch model (variance that would be explained if data complied with Rasch definition of unidimensionality) are comparable, then the chance of finding additional constructs is low. ⁴⁴ The first contrast in the residuals reports whether there are any patterns within variance unexplained by the principal component to suggest that a second construct is being measured. We used the criterion that the contrast should have the strength of at least 3 items (as measured by an eigenvalue > 3.0; eigenvalue provides an indication of the proportion of the total variance explained by an individual factor) to be considered evidence of a second construct as this indicates that the potential second dimension has only marginal explanatory power, and this result allows for ignoring further components. ⁴² The loading of items onto the contrasts allows identification of which items tap different constructs; we used a minimum loading of 0.4 to identify contrasting items. (6) Lastly, we assessed differential item functioning (DIF; form of bias in which one subgroup, e.g., women, with given levels of VRQoL respond differently to an item compared to another subgroup, e.g., men, with similar levels of VRQoL) for age (<23/>23 years), sex, and keratoconus subgroups as defined in the study. We considered DIF to be insignificant if it was <0.50 logits, mild (but probably inconsequential) if it was 0.50 to 1.00 logit, and notable if >1.00 logit. ^46,47 The interval-level IVI scores generated by the Winsteps software (version 3.68, available in the public domain at http://www.winsteps.com/index.htm), after the data fit the Rasch model, were used for analyses. 

Statistical Analysis

Descriptive analyses were performed using SPSS software (version 16.0; SPSS, Inc., Chicago, IL) to characterize the participants' sociodemographic, clinical, and IVI data using univariate analyses of variance. Normality of the distribution of the scores was examined using the Kolmogorov-Smirnov (K-S) test. We compared the VRQoL scores between the different groups of disease severity. Statistical significance was set at P < 0.05. 

Results

Sociodemographic and Clinical Characteristics of the Participants

A total of 160 patients responded to IVI (response rate 96%), and English was the preferred language by the majority (N = 115, 72%). Given the time constraints in the clinic, we could not alternate between languages or the mode of administration (self versus interviewer administered) to determine if responses differed by translation and language. 

The mean age (±SD) of the participants was 23.3 ± 5.8 years (range, 18–53) and there was a male (63%) preponderance (Table 2). The majority of the participants (79%) had PVA in the better eye and binocular PVA of ≥20/40. However, 46% of the participants had a PVA in the better eye and binocular VA of ≥20/20. Half (51%) of the participants wore CLs and 7% had undergone PK in both eyes. Approximately two-thirds of the participants had severe (63%) and the remainder had moderate (37%) keratoconus. None of our participants had mild keratoconus. The duration of keratoconus (self-reported) ranged from <1 to 27 years, with a mean (±SD) of 4.5 ± 4.9 years (median 3 years). 

Table 2. 

 

19-Item Vision-Specific Functioning Scale and 9-Item Emotional Well-Being Scale Scores According to Participants' Clinical and Sociodemographic Characteristics (N = 160)

Participant Characteristic	N (%)	Vision-Specific Functioning Score*, Mean ± SD	Emotional Well-Being Score*, Mean ± SD
Age, mean, y
<23	84 (53%)	−1.69 ± 1.85	−0.94 ± 1.62
≥23	76 (47%)	−1.58 ± 1.48	−0.71 ± 1.63
Sex
Male	101 (63%)	−1.44 ± 1.53	−0.77 ± 1.53
Female	59 (37%)	−1.97 ± 1.86	−0.95 ± 1.76
Employment status
Working	49 (31%)	−1.69 ± 1.56	−0.86 ± 1.42
Not working	111 (69%)	−1.62 ± 1.73	−0.82 ± 1.71
Binocular presenting visual acuity
≥20/20 (0.00 logMAR)	73 (46%)	−1.86 ± 1.87	−1.00 ± 1.67
<20/20 (0.00 logMAR)	87 (54%)	−1.45 ± 1.48	−0.69 ± 1.57
Duration of keratoconus diagnosis, median, y
≤3	72 (45%)	−1.73 ± 1.47	−1.04 ± 1.37
>3	88 (55%)	−1.56 ± 1.83	−0.66 ± 1.79
Severity of keratoconus†
Moderate	60 (37%)	−1.74 ± 1.56	−1.03 ± 1.45
Severe	100 (63%)	−1.58 ± 1.75	−0.72 ± 1.71
Mode of correction
Spectacles	34 (21%)	−1.78 ± 1.51	−0.99 ± 1.49
CLs	81 (51%)	−1.80 ± 1.82	−0.94 ± 1.67
Spectacles + CLs	33 (21%)	−1.17 ± 1.55	−0.61 ± 1.47
Penetrating keratoplasty ± CLs/spectacles	12 (07%)	−1.45 ± 1.35	−0.30 ± 2.03

*  Higher negative logits for visual function and emotional well-being scores indicate a better visual function and emotional well-being status, respectively.

†  Categories based on the eye with the average steep keratometry reading (moderate 45–52 D, severe > 52 D).

Psychometric Validation of the IVI Questionnaire

The data of the 32-item IVI were fitted to the Rasch model and several fit indices were explored (Table 3). Disordering of thresholds was observed for both rating scale formats. Such disordering can occur due to many reasons, including underutilization of a category, unclear description, or if number of categories exceeds the levels the participants can distinguish. ⁴⁸ Disordered thresholds may affect fit, so categories should be reorganized and Rasch model reapplied to the data. Therefore, we performed category reorganization as a first step, until thresholds were ordered, before proceeding further. 

Table 3. 

 

Fit Parameters of the Native IVI, Revised Vision-Specific Functioning, and Emotional Well-Being Scales Compared to Rasch Model Requirements

Parameters	Rasch Model	Combined Native IVI	Revised IVI Versions
			Vision-Specific Functioning Scale	Emotional Well-Being Scale
N of items	–	32	19	13
N of misfitting items	0	3	0	4
PSR	≥0.80	0.93	0.90	0.88
PSI	≥2.0	3.62	3.07	2.71
Item mean, logits	0	0	0	0
Person mean, logits	0	−1.19	−1.49	−0.69
Difference between item and person mean (targeting)	<1.0	1.19	1.49	0.69
PCA, eigenvalue* for first contrast	<3.0	4.2	2.2	2.0
DIF	<1.0	–	None	None

*  Eigenvalue provides an indication of the proportion of the total variance explained by an individual factor; see text for detailed explanation of terms.

For the first rating scale format, disordering involved categories 3 and 4 (“a fair amount” and “a lot”) so these categories were combined. Consequently, the revised scale consisted of five categories. Disordering persisted, and now it occurred between categories 1 and 2 (“hardly at all” and “a little”) that were combined, following which a four-category scale emerged. This category reorganization resulted in an improvement in the overall model fit and ordered thresholds for the first rating scale. Similar disordering was observed for the second rating scale between categories 3 and 4 (“a fair amount of the time” and “a lot of the time”), which then were combined to result in five categories and thresholds were ordered. 

Following category reorganization, the PSI and PSR values were 3.62 and 0.93 respectively, indicating excellent measurement precision. The PSI gives an estimate of the spread or separation of persons along the measurement construct. ⁴⁴ This index reflects the number of “strata” of measures that are discernible statistically and the number of distinct strata = (4G + 1)/3. ⁴⁹ A PSI of 3.62 and PSR of 0.93 for the IVI imply that it was able to distinguish reliably among five strata of participant's VRQoL. 

Targeting was −1.19 logits indicative of mistargeting, albeit slight. The negative logit sign indicates that the participants' VRQoL was relatively higher than what could be captured by the items in the IVI. There was an evidence of multidimensionality as was evidenced by PCA of residuals. This analysis showed that the proportions of variance explained by the principal component for empirical data and for the Rasch model were comparable (47.3% and 49.5%, respectively), but was <60%. In addition, 7.0% of the unexplained variance was explained by the first contrast, with an eigenvalue of 4.2 (exceeding our criterion of >3.0), suggesting a second dimension. There were no additional contrasts. Analysis of the standardized residual loadings for items in the PCA revealed a set of items loading together (>0.4) pertaining to EWB. Moreover, three items (items 20, 28, 31) demonstrated substantial misfit (infit MnSq > 1.3). These misfitting items were deleted and Rasch analysis was rerun, but items continued to misfit (items 8, 6, 10, 11, 12, 14) in further iterations. Furthermore, item deletion did not improve other fit statistics. 

Results of initial Rasch analysis suggested that the 32-item IVI is not unidimensional. Our objective was to determine whether unidimensionality could be achieved by altering the IVI, although with some modifications. We proceeded by splitting the 32-item IVI into two scales: a 19-item vision-specific functioning scale (VF scale, items 1–19) and a 13-item EWB scale (items 20–32). This strategy resulted in both scales fitting the Rasch model (Table 1). The results presented below pertain to these two scales, separately. 

For the VF scale, the PSI and PSR were 3.07 and 0.90, respectively, indicating good measurement precision (Table 3). Targeting, however, was slightly worse (−1.49 logits) and continued to remain suboptimal. One item (item 4) showed marginal misfit (Infit MnSq 1.34); however, it was retained as it captures an important VF information, that is, “taking part in recreational activities.” PCA of residuals showed that the proportion of variance explained by the Rasch measure was comparable for the empirical calculation and by the Rasch model (49.4% and 49.1%, respectively, although <60%). In addition, 5.9% of the unexplained variance was explained by the first contrast with an eigenvalue of 2.2, thereby satisfying the requirements for unidimensionality. None of the items displayed notable DIF (Table 3). 

The 13-item EWB scale possessed adequate measurement precision (PSI 2.67 and PSR 0.88), but two items misfit (items 20, 21; Infit MnSq values 1.63 and 1.40, respectively). Consequently, these two items were removed iteratively, following which an additional item misfit, albeit marginally (item 22, “your general safety when out of your home,” infit MnSq 1.33). This item also was deleted, following which another item displayed misfit, albeit marginally (item 24, “how often have you needed help from other people,” infit MnSq 1.31). We decided to delete these two marginally misfitting items because they did not appear to be assessing EWB. Following their deletion, all the remaining 9 items (9-item EWB) had infit MnSq between 0.7 and 1.3, and had good measurement precision as evidenced by PSI of 2.71 and PSR of 0.88. Targeting was −0.69 logits, suggesting that the scale was well targeted to the patients' EWB status and this can be seen in the person-item map (see Figure). PCA of the residuals showed that the proportion of variance explained by the Rasch measure was comparable for empirical calculation as well as that explained by the model (61.4% and 60.2%, respectively, which are similar to the recommended variance of 60%). In addition, 7.4% of the unexplained variance was explained by the first contrast with an eigenvalue of 2.0 (which is smaller than the set criterion of 3.0), which satisfies the requirements for unidimensionality. No item displayed notable DIF. Taken together, the above findings indicated that the two scales, VF and EWB, of the 32-item IVI are valid, reliable, and unidimensional, and can be used to assess the impact of keratoconus on VRQoL (Table 3). 

Figure

 

Person-item map for the 9-item Emotional well-being scale (N = 160) in keratoconus patients. The vertical line represents the measure of the emotional well-being variable, in logit units. Participants are located on the left side of the map, while the items are located on the right side of the map. Alongside each item also is indicated its number as in the 32-item original IVI questionnaire. Item names have been abbreviated to fit the space and the correct description of items can be found in Table 1. Each “x” represents 2 participants and each “.” represents one to three participants. By convention, the mean item endorsability is set at 0 logits (indicated with “M”). Accordingly, mean emotional well-being of participants is indicated with “M”. M, mean; S, 1 SD from the mean; T, 2 SD from the mean.

The participants' mean (±SD) VF score was −1.49 ± 1.48 logits and was −0.69 ± 1.47 logits for the EWB scale. Negative logit scores indicated that, overall, the participants had VF and EWB that were higher than what could be captured by the items in these scales. The scores from both scales had interval level properties, implying that it is appropriate to use them in parametric statistics in further analyses. The VF and EWB scale scores were not significantly different from a normal distribution (P = 0.31, K-S test for VF; P = 0.09, K-S test for EWB). 

Differences in Vision-Related Quality of Life Scores by Sociodemographic Characteristics, Severity of Keratoconus, and Mode of Correction

Univariate analyses (Table 2) demonstrated that there was no statistically significant difference in the VF (mean change −0.16, P = 0.55) and EWB (mean change −0.29, P = 0.25) scores between moderate and severe keratoconus groups. Also, the VF and EWB scores did not vary significantly by mode of correction (P > 0.05). 

Discussion

In its native form, the 32-item IVI questionnaire was not unidimensional in our sample as was evidenced by large numbers of misfitting items, and further supported by PCA of the residuals. This finding is consistent with the original IVI validation study in an Australian VI population, ²⁷ which also reported the IVI to lack unidimensionality (albeit borderline) due to EWB items. Similar to the 32-item IVI, the 28-item version also was demonstrated to lack unidimensionality, albeit in a German VI population. ⁵⁰ Taken together, these studies bear out that optimal functioning in another population should not be assumed. Rather, questionnaires should be revalidated in the population in which they are intended to be used. 

Lack of unidimensionality is problematic, in that if more than one latent construct is being assessed by a questionnaire it becomes impossible to interpret a single score from that questionnaire as a measure of any one construct. Recent observations that even minor departure from unidimensionality can affect person estimates emphasize the significance of unidimensionality as a fundamental measurement property of a questionnaire. ⁵¹ Furthermore, the lack of unidimensionality invalidates the reporting of a total questionnaire score derived from all items, because such a score does not inherently represent a single concept. Given this fundamental flaw associated with use of a total questionnaire score, there is a growing awareness to move away from such a trend. In the case of the IVI, our analyses revealed that the total score is essentially a combination of VF and EWB constructs, and while the two constructs are related, they are too different to be combined. One could argue that this represents VRQoL. Even if this were so, it still is no justification for reporting a total IVI score, which is a measure of VF and EWB. Rather than deleting EWB items systematically from the 32-item IVI so as to obtain a purely unidimensional overall score, we adopted a minimalist approach, such that the integrity of the original IVI was maintained along with its ability to measure VRQoL, as opposed to only assessing VF. Similar to the German study, the IVI in our sample could be repaired by splitting it into two subscales: VF and EWB. Both new subscales conform to the requirements of the Rasch model, namely unidimensionality, and have well fitting items and ordered thresholds. The revised VF and EWB scales provide interval level scores that now can be used for studies with keratoconus patients in India. One now can measure reliably VF and EWB separately with the IVI and, by reporting scores from these two domains, one then can claim that VRQoL is measured. However, as noted in our Introduction, VRQoL is a complex, multidimensional construct and, while single domains, such as EWB or VF, may be components of VRQoL, they are, by themselves, insufficient to constitute a complete VRQoL assessment. 

Our second aim was to determine whether the VRQoL scores varied by disease severity. While we did not have any patient with mild keratoconus, there was no significant difference in VRQoL between moderate and severe keratoconus. This indicates that the VRQoL does not vary as a function of disease severity in our patients. However, the CLEK investigators found poorer VRQoL scores (including mental health subscale) for those with severe keratoconus (>52D) as opposed to those with corneal curvature < 52 D. Although the CLEK study and ours used average of steep keratometric reading of both eyes for classification of disease severity, we did not find a difference in the VRQoL between moderate and severe keratoconus. Although it is difficult to assign a specific cause for this finding, we speculate a couple of reasons. Firstly, the use of the 32-item IVI as compared to the 28-item version; however, this was not investigated in our study. Secondly, there were almost twice the numbers of patients with severe compared to moderate keratoconus (63% vs. 37%). 

While other studies from Western ¹² and Asian ¹⁷ populations have reported the VRQoL to be impacted in keratoconus patients, we failed to find such an impact in our sample. There could a few reasons for such a deviation. Firstly, the use of different questionnaires across studies could be a reason. For example, CLEK investigators used the National Eye Institute-Visual Function Questionnaire (NEI-VFQ) and reported the VRQoL scores to be compromised significantly for keratoconus patients compared to age-matched controls of rigid contact lens wearers. ¹² Secondly, another cause of deviation could be the inclusion of a control group by other investigators, which we did not report. Tatematsu-Ogawa et al. also used the NEI-VFQ ²⁵ in a Japanese population and found lower scores for keratoconus patients when compared to age-matched controls. ¹⁷ Thirdly, the use of raw scores (and these are not measures) rather than Rasch-scaled scores of questionnaires may have led to the discrepancy. Although our patients had higher VRQoL scores (better for VF than for EWB, higher negative IVI scores indicate higher VRQoL in our study) this doesn't necessarily imply that their VRQoL was unaffected, and given that our patients had moderate or severe keratoconus, and one would expect to find some VF limitations and EWB issues in this group. It is likely that our patients may have experienced VF limitations and EWB issues that are not addressed by the IVI given that it was developed on visually impaired populations, so the content is better suited to them. Nonetheless, the relative lack of impact on VRQoL in our sample does not exclude the possibility of finding one in other populations. 

All the previous studies have used Likert or summary scoring to assess the impact of keratoconus on VRQoL. ^12,16,17,52 As noted earlier, the use of total questionnaire score is flawed, and given this, there is a growing trend to move away from reporting such scores. Furthermore, assigning ordinal numbers to response categories is erroneous because it assumes that such a method represents fundamental measurement and that all items in the questionnaire are of equal difficulty. Proponents of modern psychometric methods, Rasch analysis, have questioned these hypotheses. ⁵³ However, by showing that our keratoconus data fit the Rasch model, we have demonstrated empirically that IVI scores possess interval-level measurement properties. Such transformation is advantageous in that it reduces measurement noise and improves measurement precision, which legitimizes the use of parametric statistical tests in analysis, for example comparison of scores before and after an intervention. Although the total number of items across the two revised subscales is 28, which is similar to the 28-item Rasch version proposed by Lamoureux et al., we note that there are differences. The IVI functioned as three subscales and as an overall score in the 28-item version, and the item content is slightly different. While re-engineering of the IVI helped restored unidimensionality, it is plausible that such remedial measures may not have been required if we had used the 28-item version. Nonetheless, given the performance of the 28-item IVI in a German population wherein, too, reengineering was performed to restore unidimensionality, such a possibility seems unlikely. 

To our knowledge, ours is the first study to have validated the IVI using Rasch analysis in keratoconus patients, and in India. The use of Rasch analysis for the IVI in this sample so as to produce estimated linear interval of overall measures of VF and EWB is a significant strength of this study. When data satisfy Rasch model requirements, the approach transforms ordinal manifest data to latent interval-level data, which justifies the use of parametric statistics in the calculation of change scores, such as comparison of pre- and postinterventions, for example, in fitting of CLs, PK, for keratoconus patients. Our large sample of keratoconus patients recruited from a single tertiary eye care center and the inclusion of patient feedback in the cognitive debriefing stage of the translation of the IVI in the local languages constitutes another strength of our study. 

Our study has some limitations. Firstly, there was a male preponderance (63%) and 100% of the sample was literate. Secondly, all patients were recruited from a single hospital so our sample may not be fully representative of the Indian population, and the results cannot be generalized to keratoconus patients in other hospitals or in the community. Thirdly, the cross-sectional design of our study has to be considered while interpreting the reduction in VF and EWB across subgroups of participants. The reduced scores are based on comparison between participants categorized into different groups and do not necessarily indicate a longitudinal shift in the VF or EWB with a change in the VA or the severity of keratoconus. Fourthly, the suboptimal targeting achieved in the Rasch-analyzed VF scale, perhaps resulting from the use of the 32-item IVI instead of the 28-item version combined with a lack of difficult items, indicates that that mean level of VF of our keratoconus participants is higher than that required to perform various vision-specific tasks. This suggests that the tasks of the IVI are relatively easy for our population; including more difficult and visually challenging tasks are required. However, adding items lies within the purview of the developers of IVI and is beyond the scope of our study. 

In conclusion, Rasch analysis of the IVI questionnaire demonstrated that it was not unidimensional; however, its revised 19-item VF and 9-item EWB subscales have robust interval-level measurement properties, which support their suitability as a measure of VRQoL in this Indian keratoconus sample. Furthermore, the VRQoL did not vary by disease severity. Although our sample had relatively higher VRQoL (suggesting a lack of impact) using the revised IVI scales, this does not preclude the possibility of finding an impact of keratoconus on VRQoL in other populations. Lastly, our study also has provided evidence that IVI is a validated tool for further possible investigations either within India and/or other Indian populations outside India. The VF and EWB scales of the IVI have the potential for use as outcomes measures in clinical trials and intervention studies in keratoconus patients in India. Our results are important to clinicians and researchers alike, and will be able to offer clinicians a good understanding of VRQoL from the Indian keratoconus patient's perspective. 

Acknowledgments

Supported 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; S.P. Reddy, None; A. Fathima, None; S. Bharani, None; R. Sumalini, None; D.K. Bagga, None; P.M. Sudharman, None