Methods

The study was approved by the Rush University Medical Center institutional review board. All subjects consented to participate in the study. First, the UK version of the LupusQoL was administered to five randomly selected patients with SLE receiving longitudinal care at Rush University Medical Center, to obtain patient feedback on the appropriateness of the wording for the US modification of the instrument. Guidelines for cross-cultural validation of an instrument used in a country other than that in which it was developed may require adaptation if the populations concerned have another culture with similar language. There are sufficiently meaningful differences between the American and British cultures to necessitate modification of some items and validation of the measure in its new setting.27 In this case, adaptation, not translation, is indicated.

Based on subject feedback, minor wording revisions were performed. A team of interdisciplinary reviewers (MJ, ASP, CW, LST, KM), including the developers of the original LupusQoL, worked on the proposed wording changes to preserve the sensibility of the tool. Some examples of the wording changes made included “picking up groceries” was added to item 2 and “slower pace” was replaced by “do things more slowly” for item 7 for experiential equivalence. The expression “so severe” was dropped from item 11 to reduce comprehension error while maintaining the general concept of the item. “Honestly” was replaced by “as best” in the instructions for conceptual equivalence. The modified instrument was pretested in an additional five randomly selected subjects with SLE to check for errors and deviations from the original instrument using a probe technique. Feedback from these five debriefings led to additional minor modifications targeted to keep the readability of the instrument at the fifth grade level (Flesch–Kincaid grade level score 5–6).28 The modified instrument was then administered to another five subjects with SLE, who had no further suggestions for revision (fig 1). Modifications were made in collaboration with the UK LupusQoL developers, to ensure the goals of the measure were preserved.

• Download figure
• Open in new tab
• Download powerpoint

Figure 1

Method for modification of the LupusQoL.

Next, the modified tool, referred to as the LupusQoL-US (see supplementary online material), was administered to consecutively recruited patients with SLE who received rheumatology care at Rush University Medical Center. Demographic data and assessment of HRQoL (SF-36 and EQ5D) and disease status (disease activity and damage) were obtained, as described below.

Demographic data

Demographic information included age, age at disease diagnosis, ethnicity, education and marital status.

The LupusQoL (UK) instrument has 34 items in eight domains: physical health (PH), pain (PN), planning (PL), intimate relationships (IR), burden to others (BU), emotional health (EH), body image (BI) and fatigue (F) (fig 2). The Cronbach α for internal consistency and reliability of the domains was good (physical health α = 0.95, emotional health α = 0.95, body image α = 0.90, pain α = 0.89, planning α = 0.93, fatigue α = 0.94, intimate relationship α = 0.60 and burden to others α = 0.86).The concurrent validity of the four comparable domains was acceptable (r = 0.71 for physical health/physical functioning, r = 0.79 for emotional health/mental health, r = 0.76 for pain/bodily pain and r = 0.72 for fatigue/vitality).26

• Download figure
• Open in new tab
• Download powerpoint

Figure 2

Domain structures of UK and US LupusQoL versions.

Medical Outcomes Study-Short Form 36 (SF-36)

The SF-3613 is a generic HRQoL tool with eight domains: physical function (PF), role physical (RP), vitality (V), bodily pain (BP), social functioning (SF), general health (GH), mental health (MH) and role emotional (RE). Higher scores denote a better HRQoL.

The EQ5D29 is a generic preference-based measure of health. It includes a health state classifier that consists of five dimensions: mobility, self-care, usual activities, pain/discomfort and anxiety/depression. A utility index-based summary score is derived from the self-classifier, where 1 represents full health and 0 represents dead. EQ5D was used in addition to the SF-36 because the SF-36 may not correlate well with disease status in SLE.18 30

Disease characteristics

Organs affected by the disease were determined for each patient with SLE by reviewing medical, laboratory and radiological records. Disease activity was determined using the SLE Disease Activity Index (SELENA-SLEDAI) instrument.31 The scores range from 0 to 105, with higher scores reflecting greater disease activity. Damage was assessed using the SLE damage index (SLICC-ACR) tool.32 The damage index (SDI) score is based on evaluation of 12 organ systems.

Statistical analyses

SPSS software, version 12, was used to analyse data. Descriptive statistics were reported. The continuous variables were tested for normality. A non-parametric test (Mann–Whitney) was used for comparing continuous data. LupusQoL has a five-point Likert response format, where 0 = all of the time, 1 = most of the time, 2 = a good bit of the time, 3 = occasionally and 4 = never. Summary scores for LupusQoL-US (fig 2) were calculated guided by the scoring guidelines for LupusQoL-UK26 (see supplementary online material). Scores for the eight domains of the SF-36 and the physical and mental component summary scores were generated.13

Exploratory factor analysis (principal component analysis with varimax rotation) with eight imposed factor loadings consistent with the UK model was performed to confirm the factor structure of the US version of the instrument. Similarly, confirmatory factor analysis (CFA) was performed using structured equation modelling on AMOS software with the hypothesised eight-factor structure. A separate exploratory factor analysis (EFA) without an imposed number of domains was also performed and the model obtained was then tested by CFA. An acceptable fit for the CFA was considered if the χ²/df was <2. Rasch analysis was performed for all the items using WINSTEPS software using partial credit model. Rasch analysis33 is the current standard for the development of metric quality outcomes in healthcare and has been widely used in the development and validation of a number of outcome measures.33 Using Rasch analyses, fit analysis for each domain item was conducted. The latter refers to the differences between the data observed and that expected by the model for each person and item. Misfit item was defined by mean square outfits >±2 or <0.6. A person and item reliability index was obtained.

Measurement of psychometric properties

Internal consistency reliability

The LupusQoL-US was tested for internal consistency using Cronbach’s α. Internal consistency measures the correlation between answers to different questions about the same concept.34 A measure is considered to be reliable at the group level and have internal consistency if Cronbach’s α is >0.7.35 Test–retest reliability was determined using an intraclass coefficient to assess the stability of the measure. A small sample size of 15 subjects provided a 1-week follow-up LupusQoL-US questionnaire data. It was expected that within a 1-week period, the health status of the subjects would not vary significantly. Moreover, change in health status in the intervening 1-week period was also confirmed by a 15-point health status change scale (−7 to +7).

Content validity

Content validity measures how well the instrument samples the content of the concept of interest. This was assumed present, as the items were generated based on patient feedback.

Discriminant validity

Discriminant validity is aimed at assessing whether the instrument can distinguish between patients of varied disease severity. Disease severity in SLE can be defined by disease activity or damage. In this study both disease activity and damage were used as the external anchors and discriminant construct validity using analysis of variance was assessed.

Convergent validity

Finally, convergent validity, which measures the extent of correlation between observed relationships of the concepts and the hypothesised concepts34 was assessed. A strong correlation was defined as ⩾0.70, moderate to substantial as 0.30–0.70 and weak as <0.30.34

A proposed sample size of 20036 37 would allow approximately 5–6 subject data per item. A two-tailed p value of ⩽0.05 was considered significant in all analyses.

Results

Two hundred and five patients participated. Complete data were available for 186 subjects. Ninety-four per cent of subjects were women; ethnic composition was as follows: 60% African American, 23% Caucasian, 12% Hispanic and 6% Asian. The mean (SD) age was 42.5 (12.9) years. The mean (SD) SLEDAI and SDI were 6.2 (5.8) (median 4, range 0–27) and 2.0 (2.1) (median 1, range 0–10), respectively.

The internal consistency reliability of the LupusQoL-US ranged from 0.85 to 0.94 (table 1). The test–retest reliability ranged from 0.68 to 0.92 (table 1).Content validity was not determined in this study as the items and their integrity were maintained in the modification process of the UK version. The LupusQoL-US domains demonstrated substantial evidence of construct validity when compared with equivalent domains on the SF-36 and EQ5D (table 2).

LupusQoL-US could discriminate between patients with varied disease activity in all domains except pain, intimate relationships, body image and fatigue (table 3). Similarly, it could differentiate among subjects with varied disease damage in all domains except, intimate relationships and fatigue (table 3).

Fit analyses of items within each domain were acceptable (mean square outfits were >0.6 and none was >±2 (table 4). Item reliability index was poor for the pain domain whereas person reliability index was poor for the intimate relationship domain.

Confirmatory factor analysis of the US version of the LupusQoL using the eight domain UK version loadings of the 34 items resulted in a poor fit (χ²/df 2.06, root mean square error of approximation (RMSEA) 0.08, goodness-of-fit index (GFI) 0.76, χ² = 1059 (p = 0.001)). Standardised regression weights for item 29 and 31 with their respective domains of BI and fatigue were <0.6, but for all other items exceeded 0.6. Exploratory factor analysis (principal component analysis) with eight-factor imposition (UK version is an eight-factor model), resulted in three physical health items (nos 5, 7, 8) loading with the pain items (see supplementary online table 5). Planning, intimate relationship and burden items loaded on three separate domains as expected. However, body image item (no 26) and one fatigue item (no 31) loaded on the emotional health factor domain. Five of the eight factors had an eigenvalue of >1 and cumulatively explained 73% of the variance. The first factor had an eigenvalue of 17 and by itself explained 52% of the variance. A scree plot indicated a five-factor loading structure.

Fatigue item (no 31) refers to having difficulty “concentrating for long periods of time”. The content of the item inherently justifies its loading with the emotional health domain. On closer inspection, the modified version of item no 28 captured the same information as item no 29. An exploratory factor analysis without constraining the number of factors, after deletion of body image item no 29 and recategorisation of fatigue item no 31 as emotional health item no 7 was performed. This resulted in a five-factor loading model. The results of the rotated component matrix are shown in supplementary online table 6. Physical health, pain and planning constituted the first factor. The second factor comprised emotional health and body image items. Intimate relationship, burden to others and fatigue had separate factor loadings. The internal consistency reliability (ICR) of the thus modified five domains was as follows: first factor (PH, pain, planning; ICR = 0.96), second factor (EH seven items and BI; ICR = 0. 94), third factor (IR; ICR = 0.91), fourth factor (burden to others; ICR = 0.92) and fifth factor (fatigue three items; ICR = 0.88). Confirmatory factor analysis of the five-factor model showed similar fit statistics (CMIN/DF 2.0, RMSEA 0.07, GFI 0.78, χ² = 901 (p = 0.001)) to those of the eight-factor model.

Discussion

This study presents evidence of the psychometric properties of the LupusQoL-US. To our knowledge, this represents the first SLE-specific HRQoL tool modified and validated for use specifically in US patients. It contains concepts considered significant for HRQoL among subjects with SLE, such as sleep, fatigue, intimacy and body image. Its strength lies in the patient-based item pool generation and the larger validation of the primary instrument in the UK population.

Of note, some concerns important to US patients with SLE may not be represented in this tool. This is relevant because the item generation phase of the measure was conducted in the UK, which included predominantly Caucasian patients and no Hispanic or African-American patients, which may limit its face validity. In general, US patients have greater ethnic heterogeneity and are treated in a completely different medical care system than in the UK. For this reason, there may be greater variance in our data due to ethnic heterogeneity than in the UK cohorts. SLE disease severity and outcomes are known to vary according to ethnicity and socioeconomic status,38 39 and thus the socioeconomic status, ethnicity and disease and medical care services may pose additional concerns among US patients, which may not be relevant to UK patients.

The ICR estimates of the LupusQoL-US were similar to those of the LupusQoL. Our study includes its validity against another well validated tool EQ5D. The convergent and discriminant validity of the tool were good and comparable with those of the parent LupusQoL instrument.26 Patients may be reporting difficulty with tasks in items physical health 5, 7 and 8 owing to pain. Body image item 1 and fatigue item 1 seem to capture the emotional health construct. A major differences noted between the LupusQoL-US and the original LupusQoL related to the factor structure of the modified instrument, though the fit of both models was similar. Additional larger studies are required to assess the optimal factor structure or resolve the reasons (cultural or disease features) for the differences observed.

A final limitation of this study is its cross-sectional design, which precluded assessment of test–retest reliability or responsiveness to change in health or minimally clinically important difference. A small sample of subjects were administered the LupusQoL-US at two time points 1 week apart to obtain test–retest reliability. However, larger longitudinal studies are needed to confirm test–retest reliability estimates, ascertain its responsiveness to change and minimally clinically important differences.

In summary, there is evidence of acceptable reliability and validity of the modified LupusQoL version (LupusQoL-US) (online supplementary data) among US patients with SLE, but the tool functions slightly differently among US subjects than in UK subjects with SLE. Our study provides evidence of the cross-cultural validity of this tool and can be used to assess HRQoL among US patients with SLE. Larger studies among US patients may provide more insight into the psychometric properties of the US version of the measure, as well as its responsiveness to change in disease activity.