Predictors of Survival in Systemic Lupus Erythematosus

Abstract

We conducted the current study to determine the impact of demographic factors, household income, clinical manifestations, disease activity, serologic tests, and calendar year on survival among patients with systemic lupus erythematosus (SLE). In a large prospective cohort of patients with SLE, we used the Kaplan-Meier method to estimate survival probabilities of SLE patients over time since diagnosis. We analyzed the predictors of survival in SLE using Cox proportional hazards models.

The study included 1378 patients with SLE, with a median follow-up in the cohort of 6.1 years. One hundred eighteen patients died (8.6%). The overall cumulative probability of survival after disease diagnosis at 5, 10, 15, and 20 years was 95%, 91%, 85%, and 78%, respectively. Based on a multivariable model, age at SLE diagnosis ≥50 years old (hazard ratio [HR] = 5.9; p < .001) and male gender (HR = 2.4; p = .004) were associated with poorer survival. Patients with annual family income <$25,000 had poorer survival (HR = 1.7; p = .040). The presence of hemolytic anemia in the first year after disease diagnosis (p = .016) or during the follow-up period (p = .031) increased the risk of death. A low complement level during the first year after diagnosis was the only serologic marker of poorer survival (p = .013 for low C3 level and p = .053 for low C4 level).

Abbreviations: HR = hazard ratio, SLE = systemic lupus crythematosus, SLEDAI = Systemic Lupus Erythematous Disease Activity Index.

INTRODUCTION

Systemic lupus erythematosus (SLE) is a chronic autoimmune disease with greater mortality (1.5-5 fold increased) in comparison to the general population^1,2,23,38. In past studies, multiple demographic predictors of poor survival have been identified, including age at onset^{1,5,7,20,21,23,31,40,43}, gender^2,4,22,23,43, ethnicity^21,31,34, socioeconomic status^{3,13,30,34,43,45}, disease related factors (disease activity)^{1,16,24,28,30,40}, renal disease^{1,8,9,12,13,16,23,33,44}, central nervous system lupus^9,16,23, serositis⁴⁰, hemolytic anemia¹⁸, and thrombocytopenia^1,26 (Table 1). However, there is disagreement among previous studies, especially regarding the role of ethnicity, socioeconomic status, clinical manifestations and disease activity. In addition, few previous studies^1,3,33 have been prospective or have included multivariate analyses.

TABLE 1: Predictors of Mortality in SLE, Previous Reports*

We evaluated survival in SLE over time in the Hopkins Lupus Cohort, a longitudinal prospective study of SLE patients. Our study was designed to collect all the variables of interest using a prospective protocol.

PATIENTS AND METHODS

The study cohort included all patients with SLE who consented for the Hopkins Lupus Cohort from 1987 through 2004. Patient inclusion in the cohort was based on the clinical diagnosis of SLE by a member of the Rheumatology Division; 96% of the patients satisfied at least 4 of the 1982 American College of Rheumatology revised criteria for the classification of SLE^15,37. Basic demographic characteristics (date of birth, age at SLE onset, ethnicity, gender, socioeconomic status, years of education, combined annual household income), presenting and cumulative clinical manifestations, Systemic Lupus Erythematosus Disease Activity Index (SLEDAI) scores⁶, the Systemic Lupus International Collaborating Clinics/American College of Rheumatology Damage Index for systemic lupus erythematosus¹⁴, and immunologic markers (serum C3 and C4 level, antinuclear antibody, anti-dsDNA, anti-Sm, anti-RNP, anti-Ro, anti-La, lupus anticoagulant by dilute Russell viper venom time and anticardiolipin antibody) have been recorded since cohort entry. Patients were seen at regular intervals of 3 months, or more frequently if medically indicated. At each patient visit, a complete history, physical examination, and routine laboratory testing were performed in a systematic fashion according to protocol.

Survival was determined from the time of SLE diagnosis to the last contact or date of death. Patients not seen in the Hopkins Lupus Cohort for 6 months or longer were traced using a combination of telephone and letter interviews. For patients who were not found, the United States Social Security Death Index was used to determine death.

Clinical manifestations within the first year after diagnosis of SLE were assessed directly for those who entered the cohort within a year of diagnosis, and retrospectively for those who entered later. These were grouped under neuropsychiatric disease, renal disease, pulmonary disease, hematologic disease without hemolytic anemia, and hemolytic anemia. Neuropsychiatric disease was defined as any 1 of the following: seizure, psychosis, encephalopathy, aseptic meningitis, cerebrovascular accident due to SLE, depression, lupus headache, and cognitive impairment. Renal disease was defined as any 1 of the following: proteinuria of ≥0.5 g/d, hematuria defined as ≥5 red blood cells/HPF, renal insufficiency defined as creatinine ≥1.5 mg/dL or <75% glomerular function rate, and renal failure defined as requiring dialysis or renal transplant. Pulmonary disease was defined as interstitial pulmonary fibrosis or pulmonary hypertension. Hematologic disease without hemolytic anemia was defined as any 1 of the following: leukopenia defined as white blood cell <4000 cells/mm³ when the patient was not on drugs known to cause bone marrow suppression, lymphopenia defined as absolute lymphocyte count <1500 cells/mm³, and thrombocytopenia defined as platelet count <100,000 cells/mm³.

From 1987 until 2004, 1530 patients were enrolled in the Hopkins Lupus Cohort. Of these, 152 patients were excluded from the survival analysis: 53 patients with incomplete data, 8 patients with a questionable date of diagnosis, 90 patients with only 1 cohort visit, and 1 patient with a questionable date of death. Therefore 1378 patients were included in the current study.

Statistical Analysis

The probability of survival since diagnosis with SLE was estimated using the Kaplan-Meier method¹⁹. The fact that many patients did not enter the cohort until years after their diagnosis with SLE was handled by allowing for delayed entry into the risk sets used in calculating the survival curves, as implemented by STATA 7.0 (Stata Corporation, College Station, TX). Estimates of the association between various predictors and SLE survival were obtained using Cox proportional hazards models, also allowing for delayed entry, and implemented in SAS 8.02 (SAS Inc, Cary, NC). Analyses depending on predictors measured in the first year after diagnosis that could not be ascertained from patient histories or chart reviews were based only on those patients who entered the cohort within a year of diagnosis. The assumption of proportional hazards was assessed by including an interaction between the key predictor and time in each model and assessing the importance of this term.

RESULTS

We included 1378 patients in this study: 1275 (92.5%) female patients and 103 (7.5%) male patients (female:male ratio = 13:1). Of these, 767 (56%) were white, 543 (39%) were African-American, and 68 (5%) were of other ethnicity. The mean age at SLE diagnosis was 32.3 plusmn 12.8 years (range, 5-75 yr). The median time from disease onset to the diagnosis of SLE was 0.75 years (range, 0-48.6 yr). Additional details regarding patient characteristics are shown in Table 2.

TABLE 2: Characteristics of Patients With SLE (n = 1378)

At the time of analysis, over the median length of 6.1 years of follow-up in the cohort, 118 (8.6%) patients died. The overall cumulative probability of survival at 5, 10, 15, and 20 years after disease diagnosis was 95%, 91%, 85%, and 78%, respectively (Figure 1). For patients who died, the median time from diagnosis of disease to death was 11 years.

FIGURE 1: Kaplan-Meier estimated survival function, starting at date of SLE diagnosis.

Impact of Demographic Parameters on Survival

Patients with SLE diagnosis after the age of 50 years (p < .001) (Figure 2), male gender (p = .005), African-American ethnicity (p = .004), and lower income (p < .001) had poorer survival (Table 3). After adjustment for the other variables, age at SLE diagnosis >50 years, male gender, and lower income remained significant predictors of mortality. African-American ethnicity just failed to meet statistical significance (p = .067).

FIGURE 2: Estimated survival by age at SLE diagnosis. The numbers on the lines refer to age at diagnosis in years.

TABLE 3: Association Between Various Time-Invariant Patient Characteristics and Survival

Impact of Calendar Time

Regarding calendar time, we divided the follow-up time into 3 periods, 1987-1994, 1995-1999, and 2000-2005. It appears that mortality rates are declining (Table 4) between the 1987-1994 period and the post-2000 period (adjusted hazard ratio [HR] = 0.6; p = 0.059).

TABLE 4: Association Between Calendar Time and Hazard Rates

Impact of Clinical Manifestations on Survival

The relationships between clinical manifestations within the first year after diagnosis of SLE and survival are shown in Table 5. The presence of renal disease (p = .007), and of hemolytic anemia (p < .0001) were significantly associated with poorer survival. Because the prevalence of these clinical manifestations may differ with patient age, gender, ethnicity, and household income, and because these demographic factors also influence survival, we examined the association of each clinical manifestation with survival in analyses that adjusted for the effects of patient demographic characteristics and calendar year. In these adjusted analyses, hemolytic anemia (p = .0003) and renal disease (p = .045) remained significantly associated with poorer survival. Finally, in a fully adjusted analysis, adjusting for the other clinical conditions, hemolytic anemia was still significantly associated with mortality (p = .016), whereas renal disease was no longer a significant predictor of survival in SLE (p = .27).

TABLE 5: Association of Baseline Disease Manifestation and Survival

We also estimated the impact of developing a disease manifestation on the survival hazard using a Cox proportional hazards model with time-varying covariates. This model assumed that the hazard changes by a factor when patients developed a disease manifestation during the follow-up period. This increased hazard was assumed to continue, even if the condition resolved. From this model, hemolytic anemia increased the risk of death (HR = 2.2; p = .031), even if not present initially. No other clinical manifestations were associated with mortality risk in this model.

The actual causes of death in the patients with hemolytic anemia were infection in 10 (41.7%), cardiovascular causes in 4 (16.7%), SLE-related causes in 4 (16.7%), renal failure in 3 (12.5%), and other causes in 3 (12.5%). The time between diagnosis of hemolytic anemia and death was 89.4 ± 72.3 months.

Impact of Immunologic Markers on Survival

At 1 point during cohort participation, antinuclear antibody, anti-Ro, anti-La, anti-Sm, anti-RNP, anticardiolipin antibodies, and lupus anticoagulant (dilute Russell viper venom time) were assessed (Table 6). There were no strong associations between the results of these tests and survival.

TABLE 6: Association of Serologic Findings and Survival

Inception Cohort

There were 542 patients who entered the cohort within 1 year of SLE diagnosis. It should be noted that there were only 35 events (deaths) in the inception cohort compared to 118 in the entire cohort. In this group of patients, low C3 or C4 level in the first year after diagnosis was associated with significantly poorer survival (p = .013 and .053, respectively) (Table 7).

TABLE 7: Association Between Serologic Findings and SLEDAI Score in the First Year After Diagnosis and Survival (Based on the Inception Cohort)

Impact of Initial SLEDAI Scores of SLE on Survival in Inception Cohort

Patients who had a SLEDAI score ≥10 in the first year of SLE disease had poorer survival than patients with a SLEDAI score <10. The estimated probability of 5-year survival rate was 82% with a SLEDAI score ≥10 and 98% with a SLEDAI score <10, respectively (p = .042) (see Table 7). This association remained high (HR = 2.0), but was no longer statistically significant after adjustment for the effect of patient demographic characteristic.

DISCUSSION

The Hopkins Lupus Cohort is the largest cohort (1378 patients) survival study to date, to our knowledge. One strength is the prospective ascertainment of risk factors in all cohort patients. The cohort's 20-year survival rate was 78%. One previous study¹ from Canada reported a 20-year survival rate of 68% in SLE. Our study contained 39% African-American patients (versus 6.6% in the Canadian study), yet the survival rate at 20 years was comparable.

Our data suggest that mortality rates in this cohort are declining over time. Controlling for demographic changes in the cohort and duration of disease, the HR decreased by 40% (95% confidence interval 60%-0%; p = 0.059), comparing the mortality rate of patients observed during 2000-2005 with that of patients observed during 1987-1994. Our study also confirms the progressive improvement in survival in SLE in North America (Table 8), from 50%-77% 5-year survival^12,22,25 in 1950-1975, to 64%-87% 10-year survival^{13,31,33,41,43} in 1975-1990 and 78% 20-year survival in the current study in 1990-2004.

TABLE 8: Survival Studies in SLE Starting from Date of Diagnosis, Previous and Present Report

Hemolytic anemia was the only major clinical manifestation that was significantly associated with mortality risk in adjusted analyses. Several previous studies have reported associations between clinical manifestations at the time of diagnosis and survival. In our study, we analyzed the impact of clinical manifestations on survival, not just in terms of the first year of disease, but in any given moment during the clinical course. We found that hemolytic anemia was significantly associated with poor survival, regardless of whether it was present at diagnosis or at a later time. However, the ultimate causes of death in patients with hemolytic anemia were diverse. The presence of hemolytic anemia increased the risk of death 2-fold, independent of other disease manifestations, patient age, gender, ethnicity, or income. This finding is in accordance with a previous study by Jacobsen et al¹⁸. Other studies reported anemia¹³ or hematologic disease²⁷ as predictors of poor survival. However, these more general terms did not identify hemolytic anemia, a manifestation of active lupus, specifically. Anemia in lupus can reflect chronic disease, renal insufficiency, or even iron-deficiency.

Renal disease was less strongly associated with survival. This result was consistent with previous studies^1,26,31 that used the same statistical analysis, a Cox proportional hazards model with time-varying covariates. In contrast, some previous studies reported renal disease as a predictor of mortality in SLE^{13,16,23,33,44}.

Our study confirms that, in SLE, socioeconomic status was a predictor of poor survival in SLE. In the past, there has been no consensus as to whether a poor outcome in African-American patients with SLE is primarily related to genetic features, to socioeconomic features, or to a combination of both. We found that household income was significantly associated with survival rates (86% survival in patients with an annual family income >$25,000, versus 70% in those with an annual family income <$25,000; p < .001), independent of other characteristics. Alarcón et al³ also found poverty to be associated with poor survival. We found that African-American ethnicity was associated with poorer survival with SLE in univariate analysis, although this association was somewhat attenuated after adjusting for demographic confounders, including income (p = 0.067). Thus, the difference in survival of African-American SLE patients may be due to both ethnicity and socioeconomic status.

Age over 50 years at first diagnosis was associated with poorer survival in our multivariate analysis model (HR = 5.6; p < .001). This is consistent with previous studies that found poorer survival with elder-onset lupus^{1,5,7,20,21,23,31,40,43}. For example, a Centers for Disease Control and Prevention study³² during 1979-1998 found that crude death rates in SLE increased with age, and Kaslow et al²¹ reported, during 1968-1972, that death rates in SLE increased at age 55 and above in both male and female patients.

Male gender in SLE is an independent predictor of mortality. The current study demonstrated that males have shorter survival than females (20-year survival 68% versus 79%, respectively (p = .004)) and the observed HR is 2.4. It is higher than the male/female age-adjusted mortality ratio (1.4) in the general United States population in the year 2003¹⁷. Although SLE is more common in females than in males, all past studies^2,4,22-24,43 and the current study were consistent that male gender in SLE is associated with poorer survival.

Low complement level at diagnosis was predictive of mortality in our SLE patients (HR = 2.8, p = .011 for low C3 level; and HR = 2.2, p = .050 for low C4 level). To our knowledge, this is the first study to find low C3 and C4 levels as predictors of mortality^24,26,33.

Consistent with most previous reports^3,10,16, no specific autoantibody was associated with survival. Massardo et al²⁴ found that anti-dsDNA was an associate of survival in univariate, but not multivariate analysis. Two studies^10,16 found that SLE patients diagnosed with the secondary antiphospholipid syndrome had higher mortality, but the antiphospholipid antibody, by itself, did not increase mortality. In the inception cohort, we failed to demonstrate association of anti-ds DNA with survival.

Consistent with other studies^20,24, our data confirm an important increased risk among those with a high initial SLEDAI score. This could be addressed only in the inception cohort, because disease activity was not recorded quarterly, by protocol, in patients who were diagnosed with SLE before joining the cohort.

One limitation of the current study is that for those not in the inception cohort, some of the baseline conditions were measured retrospectively, and might have been misclassified. Because they were assessed before the event, we believe that any misclassification would be nondifferential, and therefore would tend to bias our findings toward the null.

In conclusion, our study identified demographic (male gender, age at diagnosis over 50 years, and low income), clinical (hemolytic anemia), and serologic (low C3 and C4 level) predictors of mortality in SLE. Studies of the mechanisms and interplay of these predictors are needed, if survival in SLE is to improve.