Abstract
Objective. Persistent systemic lupus erythematosus (SLE) disease activity is associated with increased morbidity and mortality. In a multicenter cohort of patients with prevalent SLE, we described persistence, patterns, and predictors of change in disease activity over time.
Methods. Based on SLE Disease Activity Index (SLEDAI)-2K scores at cohort entry, patients were classified into 4 groups: low (score < 4; LOW), moderate (4 to < 6; MOD), moderately high (6 to ≤ 10; MHIGH), and very high (> 10; VHIGH). Multivariable linear and longitudinal mixed linear regression models were used to identify predictors of change over time in SLEDAI-2K.
Results. There were 2019 participants, with declining followup data over 5 years (1326, 580, 274, 186, and 148 patients, respectively). At cohort entry, mean (± SD) age was 42 (± 17) years, disease duration 11 (± 10) years, and 90% were female. The 4 groups included 44% LOW (n = 891), 20% MOD (n = 400), 22% MHIGH (n = 442), and 14% VHIGH (n = 286); therefore, 36% had clinically important SLE activity. The proportion of patients in the LOW group at entry who moved to a higher activity level varied from 30% (167/557) at 1 year, to 49% (41/83) at 3 years, and 54% (30/56) at 5 years. Among 181 patients with MOD to VHIGH entry activity and 3 years of followup, 116 (64.1%) remained active. In all analyses, only higher SLEDAI-2K at cohort entry remained a significant predictor of higher SLEDAI-2K in subsequent years.
Conclusion. Higher SLEDAI-2K at study entry was the single major independent predictor of higher SLEDAI-2K over time, reflecting frequent persistence of active disease, even in patients with longstanding disease. This highlights gaps in the optimal treatment of SLE.
Systemic lupus erythematosus (SLE) is a complex auto-immune disease that manifests itself in many different organs in the body, primarily affecting women. SLE is a difficult disease to treat because of its heterogeneous features: no two patients are the same. Treatment aims to control disease activity and prevent damage. Damage reflects the accumulated and irreversible loss of organ function owing to either the disease itself or its treatment. Treating physicians must therefore offer patients the best treatment options that will balance the tradeoffs between risk of treatment side effects from intensive immunosuppressive therapy versus poor control of the disease, with the potential for irreversible critical organ damage.
While modern treatments for SLE have led to decreased mortality rates overall, mortality still remains unacceptably high compared to the general population1, and many patients have ongoing active disease in spite of treatment2. Persistent high disease activity over time has been clearly linked to both morbidity and mortality and is associated with accelerated damage accrual3,4. Similarly, treatments used commonly in SLE, in particular corticosteroids, also contribute to an accrual of both global and specific organ damage over time1,5,6. It is recognized that control of disease activity for many patients is suboptimal; only a small percentage achieve longterm remission that is sustained with no or minimal treatment7. While multiple treatments have been investigated in recent years, many have failed8–14 and many treatment options to reduce disease activity and limit corticosteroid use are not fully effective.
The aim of our study is to describe the distribution of levels of disease activity, both at cohort entry, in cross-sectional analyses, and longitudinally, in a cohort of prevalent patients with SLE, and to assess its associations with clinical characteristics. In addition, we examined changes in disease activity over time, and predictors of these changes. Suboptimal disease control, or persistently active SLE, has been identified as an ongoing burden for patients with SLE1,15. Information from this cohort will provide a clearer understanding of the burden of persistently active SLE.
MATERIALS AND METHODS
The 1000 Canadian Faces of Lupus study is a prospective multicenter study of SLE in Canada, enrolling patients from 2005 to 2008. Patients were enrolled at 14 sites across Canada, 10 adult and 4 pediatric rheumatology clinics. A detailed description of enrollment criteria and variables collected has been previously published16. Patients were eligible if they were identified by the site investigator(s) as having a clinical diagnosis of SLE. Both incident and prevalent cases were included. Because funding terminated shortly after the enrollment period ended, the number of annual followup visits available for analysis gradually declined.
Study variables
At the initial visit, all available medical records were reviewed by the site investigators, and clinical data were abstracted and entered onto a comprehensive standard form. Clinical manifestations of SLE were recorded, including those forming the American College of Rheumatology (ACR) criteria17 and those included in the revised Systemic Lupus Activity Measure (SLAM-R) and the revised SLE Disease Activity Index (SLEDAI-2K)18,19. In addition, autoantibody status was recorded for anti-dsDNA, anti-Sm, anti-RNP, anti-Ro/SSA, anti-La/SSB, and antiphospholipid antibodies (aPL; includes anticardiolipin antibodies and lupus anticoagulant). “Ever positive” results were collected from the medical record at the initial visit, while blood was tested at entry and annually to determine current status for antinuclear antibodies (ANA; anti-dsDNA, anti-Sm, anti-RNP, anti-Ro/SSA, anti-La/SSB, and aPL). Disease activity was measured at entry and annually using the SLEDAI-2K validated SLE activity scale18,19. Patients also filled out the Systemic Lupus Activity Questionnaire (SLAQ)20, a validated self-reported measure of disease activity, at entry and each followup visit, which includes a visual analog scale for global disease activity and fatigue. Current and past medication use were recorded and updated at each visit, and patients filled out a generic health status measure, the Medical Outcomes Study Short Form-3621. Components of the Systemic Lupus International Collaborating Clinics/ACR Damage Index22,23 were abstracted from the medical records and reviewed with the patients during the interview, and updated annually. Detailed sociodemographic data were collected including age, sex, highest education achieved, total household income, and self-reported ethnicity based on the format and categories used by Statistics Canada24. For the purposes of our analysis, patients were categorized according to the main self-chosen ethnic category.
Statistical analyses
Initially, data at cohort entry (Year 0) were analyzed, and 4 groups were created based on their SLEDAI-2K score: low (< 4), moderate (4 to < 6), moderately high (6 to < 10), and very high (> 10). These groupings were chosen by the investigators based on a SLEDAI score ≥ 6 as the standard definition of active SLE requiring treatment changes, with some investigators suggesting a score of 3 or 4 should define active disease25,26. We performed cross-sectional comparisons of the entry characteristics of the 4 groups, corresponding to different disease activity levels, using bivariate analyses: chi-square tests for categorical variables, as well as 1-way ANOVA and tests for trend for continuous variables. Patients who had sufficient followup duration were then classified into 4 similar disease activity groups at 1, 2, 3, 4, and 5 years after the initial visit. The results of the 6 groupings (years 0–5) were then cross-tabulated to estimate the probabilities of longitudinal transitions between activity levels at different times.
To determine cohort entry predictors of changes in disease activity during the first year after cohort entry, we relied on multivariable linear regression. SLEDAI-2K score (log-transformed because of the highly skewed distribution) at 1 year after cohort entry was used as the outcome variable. Potential predictors included cohort entry values of sociodemographic variables, ACR criteria, extractable nuclear antigens, aPL, current use of treatment, and the cohort entry (Year 0) SLEDAI-2K score. The final model was selected based on stepwise backward selection, with the p > 0.15 criterion for variables elimination, corresponding roughly to the Akaike information criterion27. Multiple Imputations by Chained Equation (MICE) was used to deal with missing data28 (section A1 of Supplementary Data 1, available with the online version of this article).
In separate, longitudinal analyses, we assessed predictors of SLE activity during the followup. In these analyses, SLEDAI-2K scores at years 1–5 were used as repeated measures of the outcome. Entry predictors included sociodemographic variables, ACR criteria, serology and current treatment, as well as the cohort entry SLEDAI-2K score. To account for both the between-subjects variation in the values of SLEDAI-2K scores and within-subject correlation of the repeated-over-time scores, we used linear mixed models in all longitudinal analyses, with a random intercept, and the first order autoregressive correlation structure29. Section A2 of Supplementary Data 1 (available with the online version of this article) describes handling of missing data in longitudinal analyses.
Two different repeated-measures linear mixed models were estimated, each addressing a different research question. The 2 models differed in the way that years of followup, corresponding to consecutive measurements of the SLEDAI-2K score, were analyzed. The first model included only the main effect of followup time as an adjustment covariate, and thus assessed the ability of cohort entry characteristics to predict the average-over-time level of disease activity during the followup. Similar to the linear regression analyses, the final model was selected through backward elimination, with p > 0.15 criterion for elimination. The second mixed linear model helped assess whether and how the associations between individual entry characteristics and post-entry repeated measures of SLEDAI-2K score varied with increasing followup time. To this end, the model included a series of 2-way interactions between (1) each of the entry variables, and (2) a time-varying covariate representing the followup time, in years since cohort entry. Then, we used backward elimination, with a p > 0.05 cutoff, to select statistically significant interactions with followup time, while forcing the main effects of all entry predictors and the followup time. A significant interaction with a given entry predictor would indicate whether its association with post-entry SLEDAI-2K score becomes either weaker or stronger, with increasing time since cohort entry, depending on the sign of the interaction coefficient30.
The 1000 Faces of Lupus study was approved by the regional Research Ethics Boards at each participating site, and this secondary analysis was approved by the University of Manitoba Health Research Ethics Board [HS14458(H2005:104)]. All patients provided informed written and verbal consent. Statistical analyses were conducted using RStudio 0.99.90331.
RESULTS
The cohort included 2019 patients. Mean age at cohort entry was 42 (SD 17) years; 90% were female, 63% were white, with a mean age at diagnosis of 31 (SD 15) years. Followup data were available for 1326, 580, 274, 186, and 148 patients at 1, 2, 3, 4, and 5 years, respectively. During the 5 years of observation, the median times between consecutive pre-scheduled yearly visits did not vary systematically either across the 4 disease activity groups or the followup time, and always ranged from about 11 to about 13 months (336–399 days). Demographic data for the 4 disease activity groups are presented in Table 1. Mean disease duration at cohort entry was 11.1 (SD 10.1) years, but was slightly less in those with very high disease activity, at 9.1 years, compared to 11.7 in those with low disease activity (p < 0.001). Patients with very high disease activity were also more likely to have lower incomes (Table 1), with 37.3% having incomes < $30,000 compared to 25.0% in the low disease activity group (p = 0.006). As expected, those in the higher disease activity groups also had higher physician’s global assessment (PGA) disease activity scores, SLAM-R, and SLAQ scores, had met a higher number of ACR classification criteria, and were more likely to have had renal involvement (Table 2).
Demographics of patients with SLE at cohort entry, by disease activity group.
Disease activity, clinical manifestations, and damage.
There were no differences in antimalarial use across disease activity groups, with close to 70% of patients taking these medications at their baseline visit. As expected, prednisone use at cohort entry was highest in the very highly active group at 64%, but was markedly high at about 40% even in those with low disease activity (p < 0.001; Table 3). Azathioprine and mycophenolate use did not differ between groups, while cyclophosphamide use, though infrequent, was most common in those with very highly active disease (p < 0.001). As expected, the overall proportion of patients taking immunosuppressants was highest in those with very highly active disease.
Treatment (at cohort entry).
Prednisone dose was available for only a subset of patients (Table 3). Mean daily prednisone dose was 24 mg in the very highly active group, and about half that dose (11–13 mg) in each of the other groups (p = 0.001). Daily doses > 7.5 mg were prescribed to one-third of very highly active patients and about 15% of those with low or moderate disease activity (p < 0.001).
Table 4 shows probabilities of a patient transitioning from 1 disease activity group to another, based on the SLEDAI-2K scores over 2 consecutive yearly visits. For example, patients with low disease activity at cohort entry had a 71% probability of having low activity 1 year later, and only a 5% probability of becoming very highly active. In contrast, a patient with very high initial disease activity had a probability of < 20% of attaining low disease activity 1 year later, and a 36% probability of continued very high disease activity. Across the years, patients with moderately high disease activity had about 50% probability of continued moderately high or very high disease activity 1 year later, whereas patients with low disease activity had about 14–25% probability of progressing to at least moderately high disease activity (Table 4). Figure 1 shows patients categorized by level of disease activity at cohort entry and describes their transitioning to specific disease activity groups across the followup time. For example, close to half of the initial low activity patients remained in this category throughout the followup (Figure 1a), whereas patients whose SLE was initially very highly active had about equal (15–35%) probabilities of moving into each of the 4 categories during the subsequent years of followup (Figure 1d).
Proportions (y-axis) of patients transitioning to specific disease activity groups across followup time (x-axis), depending on disease activity at cohort entry. SLEDAI: Systemic Lupus Erythematosus Disease Activity Index.
Transition matrices: probabilities of transitioning between different categories of disease activity over 2 consecutive visits, based on SLEDAI-2K score*.
In multivariable linear regression analyses, we used quantitative SLEDAI-2K scores rather than the 4 groups of disease activity to increase precision. Among SLE patients with the same entry SLEDAI-2K score, those with longer disease duration at cohort entry (p = 0.020) and highest income category (p = 0.015), patients without arthritis (p = 0.049) and women (p = 0.057) had lower SLE activity at 1 year after study entry (Table 5). However, as expected, the initial SLEDAI-2K score had by far the strongest association with the 1-year score (t statistic = 16.3, p < 0.0001, vs t statistics < 2.9 for all other entry predictors). For each additional 1-point increase of the initial SLEDAI-2K score, the SLEDAI-2K score at 1 year increased by 0.37 points (95% CI 0.32–0.42).
Multivariable linear regression analysis of changes in disease activity in the first year after cohort entry*.
In the first multivariable repeated measures linear mixed model, among all entry characteristics, only higher initial SLEDAI-2K score (t statistic = 15.15, p < 0.0001) and lower initial income (t statistic = 1.67, p = 0.095) were at least marginally statistically significant predictors of average overtime SLEDAI-2K during the followup. A 1-unit increase of the entry SLEDAI-2K score was associated, on average, with a 0.32-unit increase of the SLEDAI-2K score during the followup (95% CI 0.28–0.36; data not shown).
In the second multivariable mixed linear model, the only significant interaction with followup time involved the initial SLEDAI-2K score (p < 0.001). The estimated interaction coefficient indicated that the strength of the association between the entry and post-entry values of SLEDAI-2K score gradually decreased with increasing followup time. For example, after 4 years of followup, the effect of the entry SLEDAI-2K decreases to about one-half of its effect on the SLEDAI-2K score at 1 year. However, even after 4 years of followup, higher entry score predicts a significantly higher updated score (data not shown).
DISCUSSION
Disease activity in SLE is generally thought to be highest early in the course of the disease, declining somewhat over time32. This premise is consistent with the overall mean SLEDAI-2K score of 4.7 that we found in this cohort of prevalent patients with longstanding disease. Other authors have reported similar scores: in a Czech study, mean SLEDAI was 3.7 after nearly 8 years33, while in the Hopkins Lupus cohort, mean SLEDAI was 3.5 after 5 years of disease5. However, in spite of low mean disease activity, 36% of the participants had active SLE at cohort entry, with > 20% of those with followup data remaining active over the followup period, and at least one-third of patients had active disease at any given followup visit, as indicated by SLEDAI-2K ≥ 6. Moreover, > 40% of our patients with inactive disease (SLEDAI-2K < 6) were taking prednisone, and more than one-third required immunosuppression, reflecting rates generally similar to patients with active or very active disease. Among 600 patients from sites where prednisone dose was recorded, ∼15% of the patients in the low to moderately high disease groups were taking > 7.5 mg/day, and one-third of highly active patients were taking > 7.5 mg of prednisone per day. Prednisone doses > 7.5 mg/day are known to predict increased damage accrual over time34. A large Spanish cohort study has reported similar findings: 15% of 3568 patients had active disease (as defined by the SLEDAI ≥ 6), after a mean of about 8 years of disease, and more than half of all patients were taking prednisone2. This high frequency of prednisone and immunosuppressive treatment in these patients suggests that significant ongoing treatment is required to maintain even a relatively low level of disease activity.
We also found that the probability of a patient with quiescent disease becoming active, or flaring, at some time during the followup, was substantial. Those in the lowest disease activity group, SLEDAI-2K < 4, had a 30–50% likelihood of transitioning to more active disease in the next year (Table 4). Only a minority of patients achieved lasting low disease activity states, even with treatment (Figures 1a and 1b). Conversely, those with active disease (SLEDAI-2K > 6, and SLEDAI-2K > 10) had about 30% probability of remaining active over the followup period (Figures 1c and 1d). Other authors have reported similar findings. In 1999, Barr, et al also reported that the long-quiescent pattern (clinical SLEDAI = 0 for at least 1 year) was uncommon, seen in only 16% of patients, while about half were persistently active for at least 1 year35. Zen, et al found that only about 21% of patients achieved prolonged (≥ 5 yrs) remission while not taking corticosteroids36. These results emphasize the difficulty SLE clinicians and their patients experience in maintaining persistent control of disease.
We found that a higher SLEDAI score was the best predictor of high disease activity in subsequent years, in both linear and more complex mixed model multivariate analyses. No clinical or treatment variables (e.g., nephritis or immunosuppressive treatment) predicted longitudinal disease activity. Among demographic variables, only higher income was independently associated with lower disease activity at subsequent visits. The better longterm evolution of SLE disease activity for patients with higher income might be partly explained by their better access to care. Indeed, the proportion of patients reporting additional private insurance coverage for prescription drugs increased from 38.1% in the lowest income group to 88.6% in the highest income group (p < 0.0001). Similarly, at cohort entry, 84.7% of patients in the highest income group reported having no problem with the cost of their SLE medications compared to only 49.6% in the lowest income group (p < 0.0001). This finding has not been previously reported, although a multinational study found lower disease activity in countries with higher national social wealth37. In addition, Alarcon, et al reported that poverty was predictive of higher disease activity, especially among African Americans38. Lower incomes or socioeconomic status have been previously associated with increased damage accrual39,40,41,42, including in this cohort16. We had previously also found that lower educational attainment was associated with higher disease activity in this same cohort in cross-sectional analysis43. The relationship between socioeconomic status and SLE is not well understood, although it has become clear that the effect goes beyond access to care44. Because damage is thought to accrue from ongoing active disease as well as treatment, our findings most likely reflect a higher burden of disease experienced by those with lower incomes, but may also reflect better access to care. We were not, however, able to measure whether income had any effect on treatment delays or adherence in this cohort. No other demographic variables were predictive.
Persistent disease activity despite standard of care therapy is an important theme in SLE research and clinical care. Previous studies have shown that persistent disease activity is associated with increased damage accrual4,45,46. Gilboe, et al reported that higher disease activity early in the disease course predicts later active disease and damage45, and in another inception cohort, 25% of patients failed to achieve low disease activity within the first year; almost half of these continued to have persistent active disease throughout the 5-year followup period4. This points to our lack of ability to achieve sustained low disease activity or clinical remission, resulting in a combination of ongoing disease activity and damage accrual over time. Zen, et al recently found that a sustained 2-year remission was the minimal duration of remission associated with reduced damage46, emphasizing the need for more prolonged disease control to improve outcomes.
There are a number of limitations to this study. While the data were collected prospectively, patients entered the cohort at a mean of 11 years of disease, meaning that we had no information on disease activity earlier in their disease course. In addition, data were collected annually; this makes it possible that disease activity was underestimated (because flares may have occurred between annual visits) and direct correlations between disease activity and treatment decisions are difficult. Conversely, it is also possible that patients with low disease activity were more likely to be lost to followup, resulting in an overestimation of disease activity. Nonetheless, this cohort is large, multiethnic, and extends across Canada, with rigorously collected data, thus providing reliable, generalizable results.
Investigators had begun previously to define low disease activity in SLE, with development of the Lupus Low Disease Activity State (LLDAS), analogous to the low disease activity target long in use for rheumatoid arthritis47. LLDAS was defined as a SLEDAI ≤ 4, no new SLE activity compared to previous visit, prednisone dosage ≤ 7.5 mg/day, Safety of Estrogens in Lupus Erythematosus: National Assessment–SLEDAI PGA ≤ 1, and standard, well-tolerated doses of immunosuppressant and biologics. Although our dataset does not have all the elements needed to determine LLDAS (i.e., doses and tolerability of immunosuppressives), we do collect several of these elements. In our cohort, only half of the patients had a SLEDAI < 4 at cohort entry, and even in those patients it was generally not sustained, with about 30–40% probability of increased disease activity at subsequent visits (Table 4). Further, in a subset analysis, more than 15% of patients with SLEDAI < 4 were taking ≥ 7.5 mg/day of prednisone. Thus, only a small proportion of our cohort would meet the criteria for sustained LLDAS.
Glucocorticoids, most commonly prednisone, continue to be the “go-to” medication for both disease flares and active disease unresponsive to other therapies. This is not because clinicians fail to recognize the overwhelming evidence for the risk of damage from cumulative prednisone dosing5,48,49, but because of lack of effective therapeutic alternatives1,8,50. Our data, similar to others’, suggest that optimal control of disease activity is often not achieved with available steroid-sparing treatment options, with more than half of patients taking prednisone, often at substantial doses. This highlights gaps in the optimal treatment of SLE and the need for additional therapies.
ONLINE SUPPLEMENT
Supplementary material accompanies the online version of this article.
Acknowledgment
We thank Mellissa Moyen, national coordinator for CaNIOS (Canadian Network for Improved Outcomes in Systemic Lupus Erythematosus), the many site coordinators, and all the patients who participated in this study.
Footnotes
The 1000 Canadian Faces of Lupus study was funded by a grant from The Arthritis Society and supported by grants from Lupus Canada and the Lupus Society of Manitoba. This updated analysis was funded by GlaxoSmithKline (201364). AS and JR are employees of, and hold stock in, GlaxoSmithKline. SI was an employee of GlaxoSmithKline at the time of development of this manuscript.
- Accepted for publication June 28, 2018.
REFERENCES
In this issue
