Systemic lupus erythematosus (SLE) is a multisystem inflammatory autoimmune disease, particularly affecting young females. A recent UK community study showed SLE prevalence ranging from 1 in 500 for individuals of Afro-Caribbean origin to 1 in 2800 for northern Europeans.1 SLE is heterogeneous with a variable course and prognosis ranging from mild disease to severe, even life-threatening, disorder. Some patients experience predominantly skin manifestations, others musculoskeletal symptoms, renal disease, neurological involvement, pericarditis, pleurisy or lung inflammation or a combination of these.2 It combines problems, such as pain and fatigue with psoriasis, depression and psychological disorders. Considerable emotional adjustment is often required.3

The current emphasis on evidence-based medicine has highlighted the absence of outcome measures suitable for use in clinical trials and economic analyses. OMERACT (Outcome Measures in Rheumatology Clinical Trials) has recommended that health-related quality of life (QoL) should be assessed in both randomised clinical trials and longitudinal observational studies of SLE.4 However, the complexity of SLE makes the selection of outcome measures difficult. This paper describes the development of the SLE quality of life questionnaire (L-QoL). The L-QoL is designed to be SLE-specific and capable of providing a unidimensional index of SLE impact. The development methodology combines the theoretical strengths of the needs-based QoL model5 with the statistical and diagnostic power of the Rasch model.6 The needs-based QoL model argues that life gains its quality from the ability and capacity of individuals to satisfy their needs.5 The model has been applied successfully in the development of disease-specific QoL instruments for a wide range of diseases.7^–14 The application of the Rasch model ensures that the fundamental scaling properties of an instrument (for example, unidimensionality and level of measurement) are assessed in addition to the traditional psychometric assessments of reliability and external construct validity.6

METHODS

Patient samples

Different samples were involved at each study stage and were recruited from two hospitals in the north of England and one in London. All patients fulfilled the Updated American College of Rheumatology Revised Criteria for Classification of SLE.15 Qualitative interview sample size was based on previous studies that showed that up to 50 patients are sufficient to exhaust potential themes.7 9 11 14 For Rasch analysis, samples of 45 cases per group are sufficient to test for differential item functioning (DIF) where, at 95% significance, a difference of 0.6 standard deviations within the residuals can be detected with 80% power.16

Interviews with patients with systemic lupus erythematosus

A fundamental requirement of QoL instruments is generation of content from relevant patients.17 L-QoL content was derived from qualitative interviews designed to explore the overall impact of SLE. Interviews were audio-recorded and verbatim transcripts produced.

Selection of items for the draft questionnaire

Thematic content analysis was used to identify potential items. Items were selected if they; were relevant to the needs-model, applicable to all potential respondents, reflected a single idea and were unambiguous, short and simple. A draft questionnaire was prepared.

Cognitive debriefing interviews

Interviews assessed face and content validity. Patients completed the draft L-QoL in the interviewer’s presence and then commented on the ease of completion and appropriateness of the instructions, items and response format. Problematic items were removed and a second draft measure prepared.

Initial postal survey (single administration)

This survey enabled item reduction and assessment of the L-QoL’s scaling properties. Patients completed demographic questions and the L-QoL. Rasch analysis was conducted to determine unidimensionality and absence of DIF. The Rasch model assumes that the likelihood of a person affirming an item is a logistic function of only two factors: the level of QoL of the person and the level of QoL represented by the item. The assumption is that the L-QoL is measuring a single underlying construct (forming a unidimensional scale). Statistics indicating fit to the model test how far the observed data match that expected by the model. A significant χ² indicates misfit to the model (a significance level of 0.01 was used to account for multiple testing). Additionally, individual person- and item-fit statistics are produced. Items failing to fit the Rasch model were removed. Data were also examined for DIF. If the L-QoL is unidimensional then with the exception of random variations, items should measure the same QoL across groups.18 Items displaying DIF by age or disease duration (both above or below the median) were removed. A third draft L-QoL was produced.

Postal survey 2 (two administrations)

This survey confirmed scaling properties and assessed traditional psychometrics. Patients completed demographic and perceived disease activity questions, the L-QoL and the Nottingham Health Profile (NHP) on two occasions, 2 weeks apart. The NHP is a measure of perceived distress across six domains (energy level, pain, physical mobility, sleep, social interactions and emotional reactions).19 Rasch analysis was used to determine unidimensionality, item hierarchical ordering, absence of DIF and scale level of measurement (ordinal or interval). Following Rasch analysis a further test of unidimensionality was conducted.20 Internal consistency was assessed using Cronbach’s α. A value of 0.70 or above was required.21 L-QoL reliability (an estimate of the instrument’s reproducibility over time) was assessed using Intraclass Correlations (ICCs) between Time 1 and Time 2 scores. A value of at least 0.85 was required.22 Construct validity was assessed by relating L-QoL scores to those on the NHP and to perceived general health, reported severity of illness and patient-perceived disease activity. Moderate associations were predicted between the L-QoL and NHP sections, indicating that they assess different but related constructs. It was also hypothesised that QoL would be worse for respondents reporting poorer general health or more severe SLE.

Ethics

The study was granted local research ethics committee approval from the United Leeds Teaching Hospitals NHS Trust. Participants gave their written informed consent according to the Declaration of Helsinki.23

RESULTS

Participant details for all study stages are shown in table 1.

Findings from qualitative interviews

Some issues raised by the interviewees were common to many chronic rheumatological conditions; including reduced energy, tiredness, lacking motivation and constant pain. Other issues characterised SLE more specifically, such as: concerns over the image portrayed to others, impact on personal relationships and concerns for the future. Many needs were frustrated by SLE, including those for: freedom from pain, adequate reserves of energy, confidence, independence, control over one’s life, attractiveness, giving and receiving love and belonging.

Selection of items for the draft questionnaire

Where possible, questionnaire items utilised patient wording. The item selection process yielded 169 potential questionnaire items. Removal of duplicate, idiosyncratic and poorly phrased items resulted in a pool of 55 items for the draft questionnaire. A dichotomous (“true/not true”) response format was selected, as previous studies had indicated that this was the most appropriate for such questionnaires.10

Cognitive debriefing interviews

The draft L-QoL was tested with 16 patients with SLE. The measure was well accepted and the items considered understandable and relevant. Eleven items were removed for reasons of clarity or relevance and the wording of two improved, resulting in a 44-item questionnaire.

Testing the psychometric and scaling properties of the systemic lupus erythematosus quality of life questionnaire

A high score on the L-QoL indicates worse QoL.

First postal survey

Packs distributed to 125 patients were returned by 95 (76%; table 1). Missing data were minimal with 16 items displaying 1.1%, nine items 2.2% and two items 3.2% missing data. Missing data were found for 14 (14.7%) respondents. The number of missing items per respondent was low (mean 2.86 (SD 2.66)), confirming item acceptability. Rasch analyses showed that while some items misfit, DIF was minimal. Ten items were removed producing a 34-item version L-QoL.

Second postal survey

Ninety three (63%) of the 147 questionnaire booklets distributed were returned. Of these, 76 (81.7%) were returned completed at Time 2. Again, missing data were minimal. At Time 1, 11 items displayed 1.1% and three items 2.2% missing data. At Time 2, 14 items displayed 1.3% and four items 2.6% missing data. Missing data were observed for eight (8.6%) respondents at each administration with two respondents displaying missing data on each occasion. The mean number of items missed per respondent was 2.1 (SD 2.23) at Time 1 and 2.9 (SD 2.87) at Time 2. Rasch analyses were conducted and nine further misfitting items removed. The fit of the final 25-item L-QoL to the Rasch model was good. (Overall item fit was −0.124 (SD 0.820) and overall person fit −0.701 (SD 0.660).) Item–trait interaction χ² significance was 0.564, showing invariance of the scale across the trait. Individual item fit χ² ranged from 0.0237 to 0.988 and item residuals from −1.537 to 1.748 (table 2).

No item displayed DIF for age or disease duration (p>0.002). Person-separation reliability was 0.926 indicating that the scale is able to differentiate four or more patient groups across the continuum and has a level of reliability consistent with use at the individual patient level. The proportion of independent t-tests of person estimates was 6.41% (95% CI 2 to 11), showing that the 25-item L-QoL conformed to the requirement of strict unidimensionality.

Respondents’ scores on the measures are shown in table 3. Relatively few scored at the L-QoL extremes. The ICC was 0.95, indicating excellent reliability. Internal consistency was 0.91 at Time 1 and 0.92 at Time 2. Moderate correlations were observed between the L-QoL and NHP scores. The highest correlation was with NHP–Energy (0.80); a predictable result given the importance of fatigue in SLE.

Scores on the L-QoL were unrelated to demographic variables, duration of illness, specific symptoms and medication use. In contrast, L-QoL scores were related to presence of fatigue (p⩽0.001), pain in joints (p⩽0.001) and swollen glands (p⩽0.02). As shown in fig 1, L-QoL scores were also related to perceived disease activity (at Time 2), perceived severity of SLE and ratings of general health and QoL. Employed respondents had lower L-QoL scores (p⩽0.001).

• Download figure
• Open in new tab
• Download powerpoint

Figure 1 Systemic lupus erythematosus quality-of-life (L-QoL) scores by specified groups.

DISCUSSION

The efficient and cost-effective management of chronic disease requires competing treatment regimens to be evaluated in terms of their ability both to control the disease and improve QoL. There is a long tradition of QoL being measured by more person-centred subjective evaluation but most current instruments are generic.24 Generic health status measures, such as the SF-20 and SF-36, have been employed in an attempt to capture the range of impairments and activity limitations resulting from SLE.25 However, the SF scales have been shown to miss several issues of importance to patients with SLE while including items that are of limited relevance.26 27

Existing SLE-specific instruments such as the Systemic Lupus Activity Measure (SLAM),28 SLE Disease Activity Index (SLEDAI),29 British Isles Lupus Assessment Group (BILAG;30) questionnaire and the European Consensus Lupus Activity Measure (ECLAM;31) focus predominantly on symptoms and/or activity limitation. While such information is important, it does not inform on the patient’s QoL. The SLEQOL and the LupusQoL have been produced in an attempt to overcome the absence of a QoL instrument specific to SLE. Unfortunately, neither is based on a stated model of QoL. The SLEQOL emerged from an interest in patient functioning rather than QoL and item generation was entirely from experts rather than patients. Its content addresses symptoms and functioning.32 Consequently, despite its name, the SLEQOL is more appropriately considered as a measure of health status. The development of the LupusQoL did involve patient interviews.33 The content is also primarily a mix of symptoms and activity limitations with a few items assessing body image, planning and feelings of being a burden. As neither of these instruments provide a valid single total score, their value to clinical studies is limited. The LupusQoL yields a profile of eight domains scores.

In contrast, the content of the L-QoL was derived from interviews with patients with SLE and the items are (as far as possible) in the patients’ words. Consequently, respondents find the instrument acceptable, comprehensive and relevant. It is quick and easy to complete (taking less than 5 min) making it suitable for use in clinical settings. Application of item response theory in the form of the one-parameter Rasch model showed that L-QoL was unidimensional with good item stability and minimal DIF. Thus, the instrument conforms to the strictest psychometric standards, which are increasingly required to ensure quality measurement in rheumatology.34 Test–retest reliability and internal consistency are excellent indicating that the L-QoL is suitable for use with individual patients. Two prerequisites of construct validity are that (1) the instrument is based on a model of the construct assessed, and (2) it has good reliability.35 As these requirements were met it is possible to infer that the L-QoL has construct validity. Further support for construct validity was provided by correlation with comparator measures.

SLE is a complex disease with a wide range of possible impacts on the patient. One important attribute of the L-QoL is that it provides a single unidimensional score representing the overall impact of SLE and its treatment on the patient. This has the advantage of providing an index of QoL that is easy to interpret. The cost of obtaining such a strictly unidimensional measure is the loss of items that otherwise would seem to have face validity for patients. In the current study this item reduction resulted in discarding approximately half of the original items taken into the first draft questionnaire used in the cognitive debriefing. Nevertheless, the original themes identified by patients are still well represented by the final item set and there were valid technical reasons (eg, item bias) for discarding the items. It remains to be shown that the L-QoL is responsive to changes in QoL.

A criticism of disease-specific measures is that they do not allow comparisons to be made across diseases. Modern psychometric techniques overcome this problem through item banking.36 Where instruments are designed to measure the same construct it is possible to “co-calibrate” them on to the same underlying metric (logit) scale. The L-QoL has been explicitly designed to be added to an item bank for “needs-based” QoL across the rheumatic disease and, therefore, scores are comparable with a range of other rheumatic diseases.37 38

It is concluded that the psychometric and scaling properties of the L-QoL indicate that researchers and clinicians can have confidence in the scores obtained by respondents on the measure. The fit to the Rasch model provides an interval scale translation for use in parametric analysis, including the calculation of change scores. It will serve as a valuable tool for assessing the impact of SLE and its treatment on QoL in clinical settings, trials and research studies. Such an instrument will allow accurate assessment of the effectiveness of interventions from the patient’s perspective.