Development and preliminary validation of a systemic lupus erythematosus-specific quality-of-life instrument (SLEQOL)

Abstract

Objectives. Systemic lupus erythematosus (SLE), a chronic illness with an unpredictable and variable course, profoundly affects the quality of life (QOL). General health questionnaires are used to assess QOL in SLE, but a disease-specific instrument could offer enhanced responsiveness and content validity. We detail the steps we took to develop and validate a new SLE-specific QOL instrument, SLEQOL.

Methods. Rheumatology professionals nominated items that they felt were important determinants of QOL of SLE patients. One hundred SLE patients were asked to assess the importance and frequency of occurrence of these items and to suggest those that had not been listed. Item reduction was performed using Rasch model and factor analyses to create a new questionnaire in English. This final questionnaire was administered to a cohort of 275 patients to study its psychometric properties.

Results. Fifty-one items covering a wide range of QOL concerns were identified. The patients’ responses led to the elimination of 11. The new questionnaire of 40 items was found to have Cronbach's alpha of 0.95 and to consist of eight domains covering physical, mental and social QOL issues. It has good test–retest reliability, poor to fair cross-sectional correlation with the SF-36, with poor correlation with lupus activity or damage indices. The SLEQOL was more responsive to change than the SF-36.

Conclusions. We have developed a new 40-item SLEQOL in English and showed that it is valid for use in SLE patients in Singapore. It offers better content validity and responsiveness to change than the SF-36.

Systemic lupus erythematosus (SLE), a chronic autoimmune illness that usually begins in early adulthood, affects multiple organ systems and may be associated with considerable morbidity and mortality. In addition, SLE is incurable and runs a variable course over the patient's remaining years. The response and complication of drug treatment are generally unpredictable. These characteristics of the disease impact the quality of life (QOL) in unique ways, thus measuring the objective outcomes of morbidity and mortality only does not fully reflect the burden of disease borne by the sufferers [1]. The assessment of QOL has become an important facet of management of chronic diseases such as this and is considered to be very relevant for clinical practice, interventional trials and outcome monitoring [2, 3].

General health instruments have been shown to be valid for measuring QOL in SLE patients. They possess construct validity and responsiveness and they allow the comparison of QOL between different rheumatic diseases [4]. Criticisms levelled at their use include the argument that they are not sufficiently comprehensive for SLE, especially those that cater to the physical and neglect the mental dimension [5].

The QOL domains of importance to lupus patients include health, work and income, identity and independence, and social and family life, which can be measured with a variety of general health instruments [4–7]. Qualitative research has yielded as many as 12 concepts: uncertainty/unpredictability of lupus, fatigue, pain/symptoms, social support, misunderstood by others, fear, dependence/feelings of inadequacy/loss of self, limitations/restricted activities, personal self-management, medical treatment, emotional stress and financial issues [8]. The unpredictability of SLE in terms of the course of disease, extent of organ involvement and response to treatment is a significant component of the patient's QOL that is not well represented by general health instruments even though it may be indirectly assessed with a helplessness index [9, 10].

There are recognized advantages to using a disease-specific QOL instrument rather than a general health one [11]. Disease-specific instruments respond to smaller changes in the QOL (improved responsiveness). One benefit of this is that smaller number of patients will be needed in clinical trials. Disease-specific instruments also cover a wider range of QOL (superior face and content validity) issues arising from that illness.

As a part of a wide-ranging SLE study in our institution, we recently developed a SLE-specific QOL scale (SLEQOL) using an established methodology [12, 13]. In this paper, we describe the steps we took to develop and validate the SLEQOL in our patient population. We also compared our work with an independent QOL instrument for SLE that has been published recently [14].

Patients and method

Patients

All the patients fulfilled the 1982 American Rheumatism Association (ARA) classification criteria for SLE or the 1997 American College of Rheumatology (ACR) revision [15, 16]. They were recruited from our out-patient Rheumatology Clinic and from the hospital wards in a single institution. The study was approved by the Hospital's Ethics Committee. The patients had to be fluent in written and spoken English. Informed consent was obtained according to the Declaration of Helsinki.

Development of the QOL instrument

Because there was no SLE-specific QOL instrument for comparison, we built one from first principles. A team of rheumatologists and nurse clinicians, cognisant of the QOL issues in the literature and experienced in managing SLE patients, generated the initial list of QOL items in English. The items were assembled into a questionnaire and administered to 100 SLE patients. They were invited to suggest items not in the list that may be important to them. They were also asked to rate the items with regard to their frequency of occurrence and importance in the past month. Based on these responses, factor analysis, Rasch model analysis and expert review were used for item reduction. The reduced number of items was finally composed into a questionnaire and its psychometric properties were studied. The time frame for item recollection was now changed to 1 week in order to optimize responsiveness.

Data collection

Data were captured systematically during every study visit. In addition to clinical data and SLEQOL, the SLE Disease Activity Index (SLEDAI) [17], Systemic Lupus Activity Measure (SLAM) [18], Rheumatology Attitudes Index (RAI) [9, 10, 19], MOS 36-Item Short-Form Health Survey (SF-36) [20–22] and the Systemic Lupus International Collaborating Clinic/American College of Rheumatology damage index (SLICC/ACR-DI) [23, 24] were recorded. Patients with disease duration shorter than 3 yr were interviewed every 3 months.

The SLEDAI is a 24-item instrument for assessing SLE activity in nine organ systems. Clinical and laboratory data are required to complete the questionnaire. The score ranges from 0 to 105 points, with higher values signifying disease activity. SLAM assesses SLE activity from data on nine organ systems and seven laboratory values. The RAI, known as the Arthritis Helplessness Index before 1988, is a 15-item scale that is internally consistent and correlates significantly with the independent indices of control, self-esteem, anxiety and depression [19]. The items resolve into two factors, one depicting the subjects’ belief in their own ability to control their arthritis and the other, consisting of five items, the concept of helplessness. Initially designed for patients with rheumatoid arthritis, the RAI has been validated for use in lupus patients [9, 10]. The SF-36 is a 36-item QOL questionnaire. The items are summarized into eight domain scores, each derived from a summation and transformation of items scores. Each domain score ranges from 0 to 100, higher values representing better self-perceived QOL. The physical health domains are physical functioning, role physical, bodily pain and general health. The mental health domains are vitality, social functioning, role emotional and mental health. In this paper, we did not further collapse the scores into the physical and mental component summaries. SLICC/ACR-DI tracks irreversible organ dysfunction across 12 systems. The dysfunction must be present for 6 consecutive months in order to register as damage. The organs systems of interest are ocular (0–2 points), neuropsychiatric (0–6 points), renal (0–3 points), pulmonary (0–5 points), cardiovascular (0–6 points), peripheral vascular (0–5 points), gastrointestinal (0–6 points), musculoskeletal (0–6 points), dermatological (0–3 points), gonadal (0–1 point), diabetes (0–1 point) and malignancy (0–2 points). The score ranges from 0 to 46, higher scores signifying more damage.

Statistical analysis

Factor structure was studied by factor extraction with principal components analysis followed by varimax rotation. We determined the minimal clinically important difference (MCID) of the SLEQOL according to the method of Juniper et al. [25]. Test–retest reliability was evaluated with the intraclass correlation coefficient. We constructed a correlation matrix to compare the items in the SLEQOL with the SF-36, RAI and its helplessness subscale, the lupus activity scores and the damage score SLICC/ACR-DI using the Pearson correlation coefficient. As there is no single best responsiveness statistic for assessing HRQL (health-related quality of life) instruments [26], we used four of them to assess the evaluative property of SLEQOL and SF-36 with regard to SLE: Liang's relative efficacy (RE) [27], Liang's standardized response mean (SRM) [28], Kazis’ effect size [29] and Guyatt's coefficient [30]. We analysed the data with the SPSS Professional Statistics 8.0 (SPSS, Chicago, IL, USA). We performed Rasch model analysis [31] with BIGSTEPS, a program available at www.winsteps.com. All statistical tests were conducted at the 5% level of significance.

Results

Item generation

In May 2002, we asked a group of rheumatologists and nurse clinicians to nominate items that they believe were relevant to SLE patients. By 13 June 2002, 51 items had been identified (Table 1). As a gauge of their comprehensiveness, we noted that these items cover Wiginton's 12 concepts very well [8]. A draft questionnaire consisting of these items was administered to 100 patients in October 2002.

This group of 100 SLE patients comprises 11 males and 89 females. There were 84 Chinese, four Indians, 11 Malays and one whose ethnicity was not specified. The mean age was 39.4 ± 13.7 yr. Patient input was sought by encouraging these 100 patients to suggest activities in their lives that had been affected by their illness that had been omitted in the questionnaire. We also asked patients to score the importance and the frequency of occurrence of every item using a seven-point Likert scale. A seven-point scale was chosen over a five-point one because this allowed a finer assessment of his or her QOL and provided a better approximation to continuous data required for factor analysis.

Item reduction and creation of SLEQOL

We analysed the input from the 100 patients to produce the final questionnaire. The patients did not add any items to the list. The data on the importance of the items were subjected to Rasch model analysis and factor analysis, followed by expert review, for item elimination.

In Rasch model analysis, the probability of the respondent successfully performing a certain task is a function of the ability of the person and the difficulty of the task. It places data (both person ability and item difficulty) on some reasonable hierarchy on a single continuum of interest [31]. In this case, Rasch model analysis was used to locate the level of difficulty of each of the 51 items on a linear logit scale. The item of mean difficulty is assigned the value of zero. Negative measures denote that the items are more difficult. The INFIT and OUTFIT values (reported as mean squares) indicate the closeness of fit of the data to the stochastic Rasch model. Values less than 1.0 indicate observations are too predictable while those above 1.0 indicate unpredictability (model underfit). Zstd values are t-tests of the observed data against model-predicted values. Zstd values below zero indicate that the fit is too predictable and values above zero indicate lack of predictability. If the INFIT or OUTFIT values are between 0.6 and 1.4, Zstd can usually be ignored [32].

The most difficult item for our cohort of lupus patients is concern about being a financial burden to her or her family, followed by worry and concern that medicines do not work. The items with which our patients experience the least difficulty are social activities, getting in and out of bed and walking outdoors (Table 1).

The items resolved into 11 components on factor analysis. The first two components account for almost half of the variance and all the items for about three-quarters (Table 2). Using the results from these two analyses, we shortened the questionnaire according to a few principles, in the following order of importance. First, we ensured that there was a good distribution of questions across the different factors and difficulty levels. Second, items that have similar difficulty level and factor resolution were candidates for elimination. Third, items with INFIT or OUTFIT values below 0.6 or above 1.4 were also considered for removal [32].

Accordingly, the 11 items which were eliminated were: ‘lack of a good night's sleep’, ‘worry’, ‘slow in thinking’, fear of painful procedures’, ‘inability to go for long holidays’, ‘difficulty with dressing’, ‘difficulty with getting in and out of bed’, ‘difficulty with lifting a glass of water’, ‘difficulty with picking things from the floor’ and ‘difficulty with getting in and out of a car’. Wiginton's 12 concepts are still well represented by the remaining items.

A new questionnaire with 40 questions was devised in November 2002 (Appendix 1). The Flesch–Kincaid Grade Reading Level was 6.0 and Flesch Reading Ease was 70.2. The questions were grouped into six convenient subsections that were guided by the outcome of factor analysis. Later, we shall show that most of the subsections are internally correlated and correspond to QOL domains of relevance in SLE.

Preliminary validation of the SLEQOL

We proceeded to study the performance of the SLEQOL on SLE patients. We made use of our ongoing prospective study on SLE outcomes. Data from 275 patients were available at the time of analysis. In this group of patients, there were 249 females and 26 males, with a mean age of 40.1 ± 13.4 yr. The mean age at diagnosis was 31.6 ± 15.2 yr. There were 213 Chinese, 14 Indians, 41 Malays and seven of other races. Prospective data at 3-month intervals were available for 95 patients and repeat data collection in 2 weeks for 51 patients.

The SLEQOL has 40 items scored from 1 to 7, with higher values corresponding to worse QOL. Therefore the minimum score was 40 and the maximum 280, with a range that spanned 240 units. We took the summary score as the raw sum of these responses. It is also possible to derive the summary score from the sum of the mean of each of the six subsections, but this may place undue emphasis on those subsections with a greater number of items. This study was not designed to determine if and how of the items should be weighted. These issues have to be addressed in future analyses.

Face and content validity

These are subjective properties of QOL instruments. Face validity is satisfied when a group of informed individuals (health-care providers and patients) judges that an instrument measures part or all the experience of an illness. Content validity refers to the comprehensiveness of coverage of the instrument based on some existing theory or criteria. For example, the 40 items in SLEQOL adequately cover the 12 concepts in Wiginton's study [8]. By inviting 16 rheumatologists and rheumatology nurses and 100 patients to contribute to the items, we have ensured reasonable face and content validities. In a later section, where we present the factor analysis, we shall show that the QOL domains covered by SLEQOL include social and occupational activities, mood and self-image, physical functioning, physical symptoms, self-esteem and the unpredictability of the illness and its response to treatment.

Using cross-sectional data from 275 patients, we constructed a correlation matrix of the SLEQOL against the SF-36, RAI and its helplessness subscale, the lupus activity scores and the damage index SLICC/ACR-DI (Table 3). SLICC/ACR-DI was not computed for 29 patients because of incomplete data. The SLAM ranged from 0 to 52 (mean 3.52 ± 5.66), SLEDAI ranged from 0 to 39 (mean 2.74 ± 4.82) and SLICC/ACR-DI ranged from 0 to 6 (mean 0.67 ± 1.12).

Conventionally, Pearson's r>0.7 is regarded as showing very strong positive correlation; r ranging from 0.4 to 0.69 suggests strong positive correlation, r ranging from 0.30 to 0.39 suggests moderate positive correlation, r ranging from 0.20 to 0.29 suggests weak positive correlation, r ranging from 0.19 to −0.19 suggests no or negligible correlation, r ranging from −0.20 to −0.29 suggests weak negative correlation, r ranging from −0.30 to −0.39 suggests moderate negative correlation, r ranging from −0.40 to −0.69 suggests strong negative correlation and −0.70 or lower suggests very strong negative correlation. There was little correlation between SLEQOL and SLEDAI, and between SLAM and SLICC/ACR-DI. Generally, the summary score of the SLEQOL correlated better with the SF-36 than do its individual subsections. As support of the construct validity of SLEQOL, subsection 1 (questions on physical ability) correlated somewhat with the Physical Functional domain of the SF-36 and subsection 6 (questions regarding self-esteem, embarrassment and unpredictability) showed some correlation with the helplessness subscale of the RAI.

Rasch analysis of SLEQOL

Rasch model analysis of SLEQOL is shown in Table 4. The outcomes are generally the same though not identical to those in the previous investigation. Rasch model analysis is designed to produce person-free and instrument-free information on item difficulty. We believe that the slight differences may have resulted from the change of time frame of item recall from 1 month to 1 week. Ten of the 40 items were non-fitting (taken as INFIT or OUTFIT above 1.4 or below 0.6), namely items 2, 4, 5, 10, 13, 18, 19, 23, 31 and 39. We retained them to maintain the construct validity of the questionnaire.

Floor and ceiling effect

The ceiling or floor effect occurs when patients perceive that their condition has improved or deteriorated, respectively, beyond what a QOL questionnaire can measure. This is an inadequacy that is often difficult to eradicate. Table 5 shows the proportion of patients whose responses reached the maximum and minimum scores in the main sections of the SLEQOL and SF-36. The SF-36 and SLEQOL are scored from opposite directions but we have made the scoring consistent in this table. In this case, the floor values consistently represent good perceived QOL and ceiling values poor QOL.

The floor effect was more substantial than the ceiling effect in SLEQOL, consistent with the Rasch model analysis that many of the items were too easy for the patients. On the other hand, the ceiling effect was more obvious in the SF-36. While values above 20% are considered high [22], this does not invalidate the SLEQOL as the items contribute to the content validity. The finding that SLEQOL has a more significant floor effect and the SF-36 a ceiling effect suggests that they could compensate for each other's deficiency and they should be used together.

Test–retest reliability

Fifty-one patients with stable disease were asked to repeat the questionnaire after 2 weeks to determine the test–retest reliability. The intraclass correlation coefficient was 0.83 for the summary score. For subsections 1–6, the intraclass correlations were 0.59, 0.57, 0.72, 0.52, 0.60 and 0.80, respectively.

Internal consistency

For the entire SLEQOL, Cronbach's alpha was 0.95. For subsection 1 (questions 1–6), Cronbach's alpha was 0.85, for subsection 2 (questions 7–15) it was 0.90, for subsection 3 (questions 16–23) it was 0.89, for subsection 4 (questions 24–27) it was 0.76, for subsection 5 (questions 28–31) it was 0.93 and for subsection 6 (questions 32–40) it was 0.86.

The 40 items of the SLEQOL resolved into eight factors (Table 6). The first factor consisted of questions 7–13 (almost the entire subsection 2); they are the social and occupational activities. The second factor consisted of questions 16, 17, 19 and 28–31 (parts of subsections 3 and 5); they appear to be the mood-related and self-image questions. The third factor was made up of questions 1–6 (exclusively subsection 1), all regarding physical functioning. The fourth factor contained questions 16–23 (exclusively subsection 3), concerning physical symptoms. The fifth factor consisted of items 36–39 (part of subsection 6), concerned unpredictability of the response to treatment. Factor 6 was made up of items 32–35 (another part of subsection 6), regarding self-esteem. Factor 7 consisted of items 24–27 (exclusively subsection 4), all unpleasant aspects of the treatment of lupus. Factor 8, of very low eigenvalue, consisted of items 6, 14 and 15, regarding the ability to walk 3 km, ability to go out under the sun and making money.

Responsiveness of SLEQOL and SF-36 to changes in QOL

From the second interview onwards, patients were asked to rate their global change in QOL using a scale of integers from −7 to +7. A score of −7 indicates ‘a very great deal worse’, −6 ‘a great deal worse’, −5 ‘a good deal worse’, −4 ‘moderately worse’, −3 ‘somewhat worse’, −2 ‘a little worse’, −1 ‘almost the same, hardly any worse at all’, 0 ‘no change’, +1 ‘almost the same, hardly any better at all’, +2 ‘a little better’, +3 ‘somewhat better’, +4 ‘moderately better’, +5 ‘a good deal better’, +6 ‘a great deal better’, and +7 ‘a very great deal better’. Patients who rated the change as −1, 0 or +1 were regarded as having no difference in QOL, while those who rated it as +2, +3, −2 or −3 were felt to have a small but important change. Scores of +4, +5, −4 and −5 suggested moderate change and scores of +6, +7, −6 and −7 suggested large change [25].

We used a variety of responsiveness statistics to analyse the data. Liang's relative efficacy (RE) is formed by squaring the ratio of the appropriate t statistics of each instrument to be compared [27]. SRM is taken as mean change in score divided by the standard deviation of the change in scores [28]. Effect size is calculated as mean change divided by the standard deviation of the baseline scores [29]. Guyatt's coefficient is equal to the mean of change in scores divided by the between-subject variability or the within-person change in stable subjects [30].

In all, 119 data pairs from 95 patients were available for analysis. Other investigators have also used more than one pair of QOL data from one patient when available [33]. There were 12 data sets in which a decreased global QOL was reported, 50 in which it was unchanged and 57 that improved. Because there were only 12 in which the QOL worsened, we used the data from the unchanged and improved patients to generate the responsiveness statistics (Table 7). All the four responsiveness statistics show that the SLEQOL is more sensitive than any of the eight domains of the SF-36. However, in the patients whose QOL was unchanged, the SLEQOL scores varied to a greater degree than those of the SF-36. Therefore, it is reasonable to state that the SLEQOL was more sensitive but less specific to change than the SF-36.

However, it is also clear from Table 7 that subsections 3, 4 and 5 of the SLEQOL were the best indicators of change of QOL, superior to the summary score and all the domains of the SF-36. They were both very sensitive and specific to the changes in the QOL.

Minimal clinically important difference

The MCID is a valued property of a QOL instrument because it defines the quantum of change that warrants an intervention by the physician. However, there are many ways of defining the MCID. The methods to determine the MCID can be classified as the distributional, opinion-based and predictive approaches [34].

We derived the MCID for the summary score of SLEQOL by anchoring it to the patient's opinion of what constitutes minimal change of the global QOL, which constitutes the distributional approach.

The change of SLEQOL was −98 in one patient who assessed her global QOL change as large (deterioration), −30.4 in five patients who assessed it as moderate (deterioration), −25.33 in six patients who assessed it as small (deterioration), +5.48 in 50 patients who assessed as unchanged, +10.13 in 31 who assessed it as small (improvement), +32.75 who assessed it as moderate (improvement), and +13.8 in 10 patients who assessed it as large (improvement). That the summary score of the SLEQOL generally varies in the same direction and proportion as the self-perceived change in global QOL supports the construct validity of SLEQOL.

By taking the mean of the absolute difference of SLEQOL in the group of 37 patients who rated their global QOL change as +2, +3, −2 or −3, the MCID was calculated as 24.76, which may be rounded off to 25 for use.

Discussion

We have described how we developed a new instrument for measuring the QOL of SLE patients. We assembled the questionnaire from information from patients and health-care workers using a rigorous methodology. Subsequently, we showed that SLEQOL is valid for SLE because it possesses construct validity, face and content validity, internal consistency, test–retest reliability and responsiveness.

Though health-care professionals chose the items initially, the involvement of lupus patients followed immediately. Patients were asked to gauge the importance and frequency of the items. Crucially, they were asked to suggest additional items if they felt important areas were omitted. Qualitative interviews were not performed for two reasons. First, as physicians and rheumatology nurse educators, we are not experienced with qualitative research. Second, a good report on the qualitative experience in lupus is available [8]. Though not directly applicable across cultures, we felt that most of the domains described in that paper reflect our patients' concerns. An analysis of the 51 items shows that they cover Wiginton's 12 concepts adequately.

Our instrument may be compared with the 38-item SLE Symptom Checklist (SSC) [14]. The SSC was developed in Dutch and translated to English whereas the SLEQOL was developed entirely in English. The SLEQOL used a seven-point scale whereas the SSC employed a five-point one. The SSC and SLEQOL were elaborated upon different philosophies. Only six items are common to both: arthritis, fatigue, itch, sensitivity to sunlight, loss of concentration, and memory. The SSC consists entirely of queries on the physical symptoms related to lupus, while the SLEQOL seeks to encompass all the areas of QOL that may be affected.

That the SLEQOL only correlated weakly with the SLE activity indices is consistent with previous reports. For example, there was poor correlation of the SF-20 with SLEDAI [35, 36] and the SF-36 with BILAG [7]. On the other hand, other investigators have found that the SF-36 correlated with SLEDAI [37], the global disease activity measured by BILAG [38] and SLAM [39]. This suggests that lupus activity and damage do not consistently perturb the QOL. One possible explanation is that local factors and personality have more a direct effect on the subjective perception of QOL than life circumstances or disease [40].

We feel that the poor correlation between SF-36 and SLEQOL is due to the fact that they assay non-overlapping domains of the QOL. The SF-36 asks questions about QOL in general health while SLEQOL is specific about areas of concern in lupus.

The Rasch model analysis and distribution of the floor and ceiling responses suggest that the SLEQOL was rather easy for our patient cohort. Most of our patients were recruited from the out-patient clinic and were not in hospital wards, so they have either milder disease or controlled disease. More studies on other patient populations are needed. This may not be a large disadvantage because the instrument is likely to be employed to study the QOL of the worse-affected or deteriorating patients than the healthier patients. On the other hand, SF-36 suffered from ceiling effect and it will have limited ability to track patients as their QOL deteriorates.

At this moment, it is prudent to use SLEQOL together with a generic QOL instrument such as SF-36. First, though the SLEQOL is sufficiently comprehensive in its coverage of SLE-related QOL issues, but it ignores other aspects of health-related QOL. The employment of two instruments is relevant in SLE patients with concurrent conditions, for example, Sjögren's syndrome, osteoporosis or atherosclerosis. Second, the fortuitous discovery that SLEQOL has a significant floor effect while SF-36 has a ceiling effect means that each can offset the other's limitation. Third, data collected with a generic questionnaire enables the comparison of QOL in SLE with that in other diseases.

SLEQOL is unique because it comprehensively covers the physical and psychological aspects of QOL in SLE. The concept of unpredictability of the disease course and the outcome of treatment is well represented in SLEQOL and is absent in general health instruments.

We have documented the steps we have undertaken to construct a SLE-specific QOL instrument, and we showed that it possesses construct validity, face and content validity, internal consistency, test–retest reliability and responsiveness. Thus, the validity of the SLEQOL has been established. Full validation is awaited; work that remains to be done include resolving its role in clinical trials and routine practice and confirming its applicability in different populations.

This study was supported by a grant from the Biomedical Research Council of Singapore (01/1/28/18/016). We thank the Medical Outcomes Trust for permission to use the SF-36™ Health Survey. We also thank Mr Soo Yuen Chong for excellent support in data analysis.

The authors have declared no conflicts of interest.

References

Author notes

Department of Rheumatology, Allergy and Immunology, Tan Tock Seng Hospital, Singapore and 1Health Promotion Board, Singapore.