Using the Short Form 6D, as an overall measure of health, to predict damage accrual and mortality in patients with systemic lupus erythematosus: XLVII, results from a multiethnic US cohort
To determine if overall health status as assessed by the Short Form 6D (SF-6D) index, a preference-based generic measure of health, is associated with the occurrence of damage accrual and mortality in patients with systemic lupus erythematosus (SLE).
We studied SLE patients (American College of Rheumatology criteria) from the LUpus in MInorities, NAture versus nurture cohort (LUMINA), a longitudinal multiethnic cohort. The contribution of the SF-6D as assessed at enrollment to damage accrual at the last visit and mortality was examined. All variables previously shown to be determinants of damage accrual and mortality and corroborated by univariable analyses were adjusted for in multivariable models (Poisson and Cox proportional hazards regressions, respectively). Damage accrual and mortality were the dependent variables. Similar analyses were performed examining the associations of the Short Form 36 summary measures (physical component summary [PCS], mental component summary [MCS]) with these outcomes.
In 552 patients, the SF-6D was negatively associated with damage accrual and mortality in the univariable analyses; the association with damage was confirmed in the multivariable analyses (χ2 = 9.020, P = 0.002) but the association with mortality was not confirmed (hazard ratio 0.495, 95% confidence interval 0.041–6.038). When the PCS and MCS were evaluated, the PCS, but not the MCS, was found to be associated with damage but not with mortality.
The SF-6D (and the PCS) as measured early in the disease course were found to independently predict damage accrual at the last visit, but not mortality. Although the SF-6D was originally conceived as a utility measure, it may be used to accurately assess overall health status in patients with SLE.
Preference-based generic measures of health are increasingly being used to measure quality of life, although they were originally developed for economic evaluations as utility measures (1, 2). The Short Form 6D (SF-6D) (3, 4) is a 6-dimensional preference-based scoring system derived from the Short Form 36 (SF-36); the 6 dimensions are physical functioning, role limitations, social functioning, pain, mental health, and vitality. The SF-6D algorithm generates health state values ranging from 1.0 (no difficulties with any of the 6 dimensions or perfect health) to 0.296 (most impaired level on all 6 dimensions, or worse than being dead) (5–7).
Although in the majority of studies the SF-6D has been used for its originally intended purpose, which is as a utility measure, it is now also being used to assess overall self-reported health status (or health-related quality of life [HRQOL]). Diseases or clinical situations in which the SF-6D has been used for this purpose include obesity, diabetes, chronic liver disease, coronary heart disease, human immunodeficiency virus, asthma, rheumatoid arthritis, total hip arthroplasties, and hip fractures (4, 8–15); until now, however, the SF-6D has not been used as a utility measure or as a measure of overall health status in patients with systemic lupus erythematosus (SLE). In contrast, the SF-36 has been extensively used in SLE (16–21); however, we thought it would be advantageous to obtain a single numeric value (the SF-6D index) that could be reliably used to predict subsequent health outcomes in these patients.
We have previously found that patients with SLE participating in a longitudinal study of outcome function at lower levels than individuals from the general population (22). However, at the time those analyses were conducted, the SF-36 had not been found to be a predictor of either damage accrual or mortality, 2 well-defined outcomes in patients with SLE. For the present study, we explored whether the SF-6D could be a predictor of either outcome and hypothesized that the worse the overall health status early in the course of the disease, the more likely that damage accrual and mortality would occur subsequently. Given that LUMINA (LUpus in MInorities, NAture versus nurture) has accrued a large number of patients and years of observation since the cohort was first established, we also reexamined the physical and mental summary measures (physical component summary [PCS] and mental component summary [MCS]) of the SF-36 as possible predictors of these 2 outcomes.
PATIENTS AND METHODS
As previously described (23), LUMINA is a longitudinal study of outcome in patients with SLE. All LUMINA patients met the American College of Rheumatology (ACR) criteria for the classification of SLE (24), had disease duration ≤5 years, were ≥16 years of age, were of defined ethnicity (Hispanic [Mexican or Puerto Rican ancestry], African American, and Caucasian), and lived in the geographic recruitment area of the participating centers (University of Alabama at Birmingham, The University of Texas-Houston Health Science Center, and The University of Puerto Rico Medical Sciences Campus). These patients were, for the most part, referred for medical care and enrolled into LUMINA from either the inpatient or outpatient services of these centers and their affiliated practices. Approximately 90% of all patients referred for possible enrollment into LUMINA agreed to participate and were enrolled; patients who refused to participate were, in general, comparable with those recruited into the cohort on socioeconomic/demographic and clinical features. The institutional review board of each participating center approved the LUMINA study; written informed consent was obtained from each participating patient according to the Declaration of Helsinki.
Prior to enrollment, all medical records were reviewed to confirm the patient's eligibility and to gather socioeconomic/demographic and relevant clinical data from the time of diagnosis to enrollment. Each patient had a baseline visit (T0), and followup visits were conducted every 6 months for the first year and yearly thereafter. A LUMINA study visit consisted of an interview, a physical examination, and laboratory tests. Data for missed study visits were obtained, whenever possible, by review of all available medical records. The time of diagnosis (TD) was defined as the time at which patients met 4 ACR criteria for SLE (24).
As previously reported (25, 26), the LUMINA database includes variables from the following domains: socioeconomic/demographic, clinical, immunologic, genetic, behavioral, and psychological. These variables were measured at T0 and at every subsequent visit. Only the variables included in these analyses will be described.
Variables from the socioeconomic/demographic domain included age, sex, ethnicity, and poverty level (as defined by the US federal government, adjusted for the number of individuals in the household) (27). Clinical variables included total disease duration (the time that elapsed between TD and the last visit [TL]), disease activity, damage accrual, medications, and mortality.
Disease activity was assessed using the revised Systemic Lupus Activity Measure (SLAM-R) (28); the average SLAM-R score, from T0 to TL, was calculated as a measure of disease activity over time. Damage was measured with the Systemic Lupus International Collaborating Clinics/ACR Damage Index (29) at T0 and TL. The cumulative exposure to glucocorticoids (prednisone equivalent) was also included and was calculated as the average dose of prednisone in mg per day.
The SF-6D index was derived from the SF-36 using the algorithm described by Brazier et al based on standard gamble methodology (30). In short, the original 8 dimensions of the SF-36 were reduced to 6 by excluding the general health dimension and combining the physical and mental role limitations dimensions into one. The final index includes items from the following dimensions: 3 items from physical functioning, 2 from role limitations, 1 from social functioning, 2 from pain, 1 from mental health, and 1 from vitality; this index provides a single numeric value ranging from 0.296 (worst possible health, worse than being dead) to 1.0 (perfect health).
The contributions of the SF-6D and the 2 measures of the SF-36 (PCS and MCS) as assessed at T0 to damage accrual at TL and to mortality were examined first by univariable analyses. Multivariable models with damage accrual and mortality as the dependent variables and all variables significant in the univariable analyses as independent variables were examined next. Given the distribution of the damage scores, damage accrual at TL was examined by Poisson regressions. For the mortality analyses, Cox proportional hazards regression models were examined. Two sets of models were examined for each of these outcomes, one in which the SF-6D was included as an independent variable and the other in which the PCS and MCS were included instead. Because the SF-6D index is a mathematical derivation of the SF-36, the PCS and MCS and the SF-6D index were not included in the same model. However, given the interaction between these constructs and poverty, and given that poverty is a predictor of both damage and mortality, additional analyses were conducted using these interaction terms. Statistical significance was defined as a P value ≤0.05. All statistical analyses were performed using SAS software, version 9.1 (SAS Institute, Cary, NC).
A total of 552 patients were studied. Of the 552 patients, 491 (89%) were women. There were 105 (19%) Texan Hispanics, 100 (18%) Puerto Rican Hispanics, 183 (33%) African Americans, and 164 (30%) Caucasians. Their mean ± SD age and total disease duration in years were 36.8 ± 12.6 and 5.4 ± 3.5, respectively.
First, the mean and median damage scores for the different variables were examined; continuous variables were divided into categories using either biologically plausible cutoff values (age, disease duration) or the median value for the variable being examined (SLAM-R score, glucocorticoid dose, and SF-6D and SF-36 PCS and MCS scores). Overall, higher damage scores were found in patients who were older, male, African American, and Texan-Hispanic, living below the poverty line, those who had longer disease duration, higher SLAM-R scores, and lower SF-36 PCS and MCS scores and SF-6D scores. These data are shown in Table 1. Both the T0 SF-6D and the SF-36 summary measures were found to be negatively associated with the later occurrence of damage. In addition, variables previously shown to be associated with damage accrual (31) were also found to be significant in these analyses (32). These data are depicted in Table 2.
Table 1. Mean and median Systemic Lupus International Collaborating Clinics/American College of Rheumatology Damage Index scores at last visit in LUMINA (LUpus in MInorities: NAture versus nurture) patients as per categories of the different variables examined*
|Age, years|| || |
|Sex|| || |
|Ethnicity|| || |
| Hispanic Texan||2.22||1|
| Hispanic Puerto Rican||0.56||0|
| African American||2.31||2|
|Poverty level†|| || |
|Disease duration, years‡|| || |
|Average SLAM-R score|| || |
|Average GC dose, prednisone mg/day|| || |
|SF-6D score at T0|| || |
|SF-36 PCS score at T0|| || |
|SF-36 MCS score at T0|| || |
Table 2. Variables associated with damage accrual at last visit by Poisson univariable regression analyses*
|Age, years||12.39||< 0.001|
|Ethnicity†|| || |
| Hispanic Texan||87.53||< 0.001|
| African American||101.26||< 0.001|
| Caucasian||38.22||< 0.001|
|Poverty level‡||44.47||< 0.001|
|Total disease duration, years§||109.14||< 0.001|
|Average SLAM-R score||180.32||< 0.001|
|SDI score¶||437.90||< 0.001|
|Average GC dose in mg||16.40||< 0.001|
|SF-6D score at T0||53.63||< 0.001|
|SF-36 PCS score at T0||63.33||< 0.001|
|SF-36 MCS score at T0||28.59||< 0.001|
Multivariable analyses: SF-6D.
Of the 552 patients, only 501 could be included in the multivariable analyses of damage due to missing values for either the dependent or the independent variables. The SF-6D was negatively associated with the occurrence of damage at TL after adjusting for age, ethnicity, sex, poverty, glucocorticoid use, disease duration, disease activity at TD, and damage accrual at T0 (or first recorded; χ2 = 9.020, P = 0.0027). These data are shown in Table 3.
Table 3. Variables independently associated with damage accrual at last visit by Poisson regression analyses*
|Age, years||14.99||< 0.0001|
|Ethnicity†|| || |
| Hispanic Texan||36.04||< 0.0001|
| African American||25.33||< 0.0001|
|Poverty level‡||19.74||< 0.0001|
|Total disease duration, years§||103.76||< 0.0001|
|SLAM-R score at diagnosis||16.31||< 0.0001|
|SDI score¶||171.79||< 0.0001|
|Average GC dose, mg prednisone/day||3.99||0.0459|
|SF-6D score at T0||9.02||0.0027|
Multivariable analyses: SF-36.
When the PCS and MCS measures of the SF-36 were entered into the Poisson regression, the variables retained were the same as in the previous model. The PCS was also found to be negatively associated with damage in this model (χ2 = 5.60, P = 0.0179) but the MCS was not (data not shown).
Alternative models: interaction terms.
When the interaction term between poverty and the SF-6D index was added to the model, neither poverty nor the interaction term was a significant predictor of damage accrual but the SF-6D was of borderline significance (P = 0.066). In contrast, when similar analyses were performed including the interaction term between poverty and the PCS, neither the PCS nor poverty remained significant but the interaction term was of borderline significance (P = 0.057).
Variables associated with mortality by Cox proportional hazards regression univariable analyses were entirely consistent with previously reported analyses of mortality conducted in our cohort (33). The SF-6D and the PCS measure of the SF-36 (but not the MCS) were negatively associated with mortality in these analyses. These data are depicted in Table 4.
Table 4. Variables associated with mortality at 10 years by univariable Cox proportional hazard regression analyses*
|Ethnicity†|| || || |
| Hispanic Texan||1.836||0.891–3.785||0.099|
| Hispanic Puerto Rican||0.213||0.028–1.639||0.137|
| African American||2.029||1.065–3.866||0.031|
|SLAM-R score||1.145||1.111–1.180||< 0.001|
|SDI score§||1.482||1.277–1.719||< 0.001|
|SF-6D score at T0||0.030||0.003–0.272||0.001|
|SF-36 MCS score at T0||0.983||0.954–1.012||0.247|
|SF-36 PCS score at T0||0.944||0.913–0.975||< 0.001|
Multivariable analyses: SF-6D.
At the time this study was performed, 64 deaths had been reported. The analyses included 52 of these 64 deaths; the others could not be included because of missing values for some of the independent variables included in the model. By Cox multivariable proportional hazards regression analyses, and after adjusting for age, ethnicity, sex, poverty, disease activity at T0, and damage accrual at T0 (or first recorded), the SF-6D was not associated with the occurrence of mortality (hazard ratio [HR] 0.495, 95% confidence interval [95% CI] 0.041–6.038). These data are depicted in Table 5.
Table 5. Multivariable Cox proportional hazard regression analysis of mortality at 10 years*
|Ethnicity†|| || || |
| Hispanic Texan||1.292||0.548–3.047||0.5584|
| Hispanic Puerto Rican||0.331||0.042–2.616||0.2945|
| African American||0.926||0.407–2.108||0.8550|
|SLAM-R score at T0||1.113||1.064–1.164||< 0.0001|
|SF-6D score at T0||0.495||0.041–6.038||0.5818|
Multivariable analyses: SF-36.
When the PCS measure of the SF-36 was included instead of the SF-6D, the resulting model was consistent with the SF-6D model; the PCS was not retained in the model (HR 1.014, 95% CI 0.959–1.073).
When poverty was excluded from the Cox regressions, the PCS and the SF-6D became strong predictors of mortality. If poverty, the SF-6D (or the PCS), and the corresponding interaction terms were included, the interaction term was significant in the case of the SF-6D (and the SF-6D and poverty were of borderline significance); this was not the case for the PCS regression in which neither the PCS nor the interaction term remained significant (data not shown).
Self-perceived HRQOL has become increasingly important in biomedical clinical research. In the LUMINA cohort, a longitudinal study of outcomes, we have previously explored intermediate and long-term outcomes including disease activity, damage accrual, self-reported HRQOL, and mortality (16, 22, 32, 34–36).
Although the SF-6D was originally designed for economic evaluations as a utility measure, it has been shown to be a reliable instrument to assess HRQOL (37, 38). We have now taken advantage of having the data necessary to derive the SF-6D (since the SF-36 had been administered to our patients) to assess its impact, when measured early in the disease course, on both damage accrual and mortality. We perceive the SF-6D to be a multidimensional measure of HRQOL represented by a single numeric value spanning the range of health status. This single value may be easier to grasp for clinicians and researchers alike than the different scales and summary measures of the SF-36; after all, perceived health, whether affected primarily in its mental or physical domains, is but one construct. We found that lower SF-6D values early in the disease course independently predict damage accrual at TL; this was not the case for mortality because the SF-6D was not retained in the model.
In parallel to examining the SF-6D index, we also reexamined the impact of the SF-36 summary measures (PCS and MCS) on both damage and mortality, given that not only has the LUMINA cohort accrued more years of observation, but also more patients now constitute this cohort. In previous analyses, significant and negative correlations between damage accrual at TL and the PCS and MCS measures of the SF-36 were found by univariable analyses; however, these variables were not retained in the multivariable analyses (32). In the present study, we found that both summary measures of the SF-36 were negatively associated with damage accrual at TL by univariable analyses; in the multivariable analyses, however, the PCS, but not the MCS, remained negatively associated with damage accrual, albeit the relationship was of lesser magnitude than for the SF-6D index. Although the SF-6D is a mathematical derivation of the SF-36, the manner in which it is derived and the items and dimensions included may better represent overall self-reported HRQOL and thus influence a patient's subsequent outcomes in a more decisive manner than the summary measures of the SF-36. In short, the SF-6D may be somewhat more sensitive than the SF-36 to detect those changes in HRQOL and more likely to detect subsequent impact on damage, particularly considering that the lower levels of the SF-6D index are much more sensitive than the upper levels (3). This of course will need to be corroborated in other patient populations before it is fully accepted. Conversely, damage accrual occurring early in the course of the disease has been found to be an independent predictor of overall physical function, but not mental function (19, 22). Other authors have not found a relationship between HRQOL and damage accrual in either direction (20, 39). It also should be noted that a certain degree of interaction between poverty and these measures of HRQOL was found. When interaction terms were added to the damage models, the SF-6D remained of borderline significance whereas the PCS lost its significance altogether.
Patients with SLE are now expected to live years, if not decades, after their diagnosis, although their life expectancy is still below that of the general population (40, 41); identifying the factors affecting this terminal outcome of the disease is therefore quite relevant. Mortality in lupus has been associated with socioeconomic factors such as age, ethnicity, and poverty, as well as with clinical factors including disease activity and damage accrual (33, 42–44), but the role of self-perceived HRQOL as a predictor of mortality in patients with lupus has not been supported by us or other researchers. Using either the SF-6D or the summary measures of the SF-36 as self-reported measures of HRQOL, we have not been able to demonstrate their role as predictive factors of mortality; however, it is entirely possible that their effect is not direct but rather is mediated by their effect on damage, which is known to affect survival (33). To this end, it is important to point out that patients with lupus may be experiencing a longer survival than in years past, yet their self-perceived HRQOL is less than optimal, in part because of the damage being accrued.
The alternative explanation, of course, is that poverty is such a strong predictor of mortality and that the effect of HRQOL ascertained either by the summary measures of the SF-36 or by the SF-6D cannot be fully appreciated unless poverty is removed from the models. In fact, our alternative models support this assertion. These data, taken together with the damage data discussed above, underscore the importance of poverty in the intermediate and long-term outcomes of a disease like lupus; we have recognized this fact from the very first analyses of mortality conducted in our cohort (33) and in recent analyses of damage (31).
The need to measure HRQOL in the assessment of patients with SLE enrolled in either randomized clinical trials or in longitudinal observational studies has been recognized by the Outcome Measures in Rheumatology Clinical Trials (OMERACT) group and by regulatory agencies (45, 46). In fact, functional assessment is now an integral part of ongoing clinical trials aimed at the study of new pharmacologic and biologic compounds (47, 48) for the treatment of patients with lupus. Whether we will continue to use the SF-36 or instead turn to the SF-6D to ascertain this construct cannot be determined from our work alone and the published literature to date. Our study demonstrates that low levels of self-perceived overall health early in the course of the disease may be premonitory of poor later outcomes and that physicians caring for these patients should be alert to this fact. Identifying and treating these patients appropriately and promptly is necessary if the long-term deleterious outcomes of the disease are to be prevented.
Dr. Alarco'n had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.
Study design. Fernández, Alarcón, McGwin, Sanchez.
Acquisition of data. Fernández, Vilá, Reveille.
Analysis and interpretation of data. Fernández, Alarcón, McGwin, Apte, Vilá.
Manuscript preparation. Fernández, Alarcón, McGwin, Sanchez, Reveille.
Statistical analysis. McGwin, Apte.