To examine childhood-onset disease as a predictor of mortality in a cohort of adult patients with systemic lupus erythematosus (SLE).
To examine childhood-onset disease as a predictor of mortality in a cohort of adult patients with systemic lupus erythematosus (SLE).
Data were derived from the University of California Lupus Outcomes Study, a longitudinal cohort of 957 adult subjects with SLE that includes 98 subjects with childhood-onset SLE. Baseline and followup data were obtained via telephone interviews conducted in 2002–2007. The number of deaths during 5 years of followup was determined and standardized mortality ratios (SMRs) for the cohort, and across age groups, were calculated. Kaplan-Meier life table analysis was used to compare mortality rates between childhood- (defined as SLE diagnosis at <18 years of age) and adult-onset SLE. Multivariate Cox proportional hazard models were used to determine predictors of mortality.
During the median followup period of 48 months, 72 deaths (7.5% of subjects) occurred, including 9 deaths (12.5%) in subjects with childhood-onset SLE. The overall SMR was 2.5 (95% confidence interval [95% CI] 2.0–3.2). In Kaplan-Meier survival analysis, after adjusting for age, childhood-onset subjects were at increased risk for mortality throughout the followup period (P< 0.0001). In a multivariate model adjusting for age, disease duration, and other covariates, childhood-onset SLE was independently associated with an increased mortality risk (hazard ratio [HR] 3.1, 95% CI 1.3–7.3), as was low socioeconomic status measured by education (HR 1.9, 95% CI 1.1–3.2), and end stage renal disease (HR 2.1, 95% CI 1.1–4.0).
Childhood-onset SLE was a strong predictor of mortality in this cohort. Interventions are needed to prevent early mortality in this population.
The long-term survival for adults and children with systemic lupus erythematosus (SLE) has improved dramatically over the last several decades. While once associated with significant early mortality, the 5-year survival rates for SLE have improved from 64–87% in the 1980s to >95% today (1–4). Improvement in survival has been attributed to several important advances in SLE care including more timely diagnosis and treatment, and earlier recognition and more aggressive management of disease-related comorbidities. Despite this improvement in short-term survival, a significant percentage of SLE patients still die prematurely. Across studies, the 15-year survival rate ranges from 76–85% (1–3). In addition, the standardized mortality rate for patients with adult-onset SLE is 2 to 5 times higher than that for the general population (5–7) and is almost 20 times higher among young adults with SLE (8).
Approximately 15–20% of patients with SLE are diagnosed during childhood (9). Although the trends in improved survival for childhood-onset SLE appear to be similar to that of adult-onset SLE (10–13), little is known about the long-term outcomes of adults with childhood-onset SLE. Prior studies suggest that childhood-onset SLE has a more aggressive course than adult-onset SLE (14–16), likely leading to increased exposure to potentially toxic immunosuppressive medications, over a longer disease duration (17). Similarly, some studies suggest that patients with childhood-onset SLE accumulate disease damage more quickly, as compared with adult-onset SLE, and as a result may be at higher risk for early mortality (16).
The primary objective of this study was to determine predictors of mortality in a large, community-based cohort of patients with SLE, with a focus on describing the long-term mortality associated with childhood-onset SLE. Given the known increased risk of morbidity associated with childhood-onset SLE, we hypothesized that these subjects would have a higher risk of mortality in the followup period, as compared with their adult counterparts.
Between 2002 and 2003, 957 subjects were enrolled in the University of California, San Francisco (UCSF) Lupus Outcomes Study (LOS), an ongoing longitudinal study of a large cohort of individuals with SLE from the US. Details regarding eligibility and enrollment of participants have been described elsewhere (18). Briefly, subjects previously enrolled in the UCSF Lupus Genetics Project (19) were invited to enroll in the LOS. Participants were recruited from both clinical and community-based sources: 22% from UCSF-associated clinics, 11% from non-UCSF rheumatology offices, and 67% from various community-based sources (e.g., lupus support groups, conferences, newsletters, Web sites). All participants had a confirmed diagnosis of SLE based on review of the medical records. The study protocol was approved by the UCSF Committee on Human Research.
LOS data are derived from structured, 1-hour telephone interviews conducted by trained interviewers. Validated items from the annual survey pertaining to demographic and socioeconomic characteristics, cumulative disease manifestations, and recent SLE activity and general health status were used to determine predictors of mortality in the LOS. The data in this study include results from the first (baseline) interviews that occurred in 2002–2003, and the subsequent 4 years of annual interviews (2004–2007).
At baseline, information regarding various demographic characteristics was collected including age, sex, self-reported ethnicity, education, and poverty status. Ethnicity was dichotomized into white versus nonwhite. The highest level of education achieved either before or during the study followup period was categorized as having graduated high school or less, some college/trade school, or a college degree or higher. The percentage of participants living below the poverty level, defined as a household income <125% of the Federal poverty guidelines, was also determined.
Subjects were classified as having childhood-onset SLE if their age at diagnosis was <18 years.
Disease duration was calculated as the number of years since the diagnosis of SLE. Disease activity was assessed using the Systemic Lupus Activity Questionnaire (SLAQ) (20), a validated, patient-reported assessment of disease activity in SLE. The SLAQ correlates strongly with the Systemic Lupus Activity Measure-Revised and with the other self-report measures of SLE activity used in this study (21). SLAQ scores were not available from the first year (baseline) interviews, leading to missing data for 70 subjects who only completed the baseline interview. In addition to the SLAQ, disease activity was determined by patient assessment of disease activity (reported on a 0–10 scale, where 0 = no activity and 10 = extremely active); this measure was available in all years and correlated strongly (r = 0.73) with the SLAQ. General health status was assessed by the Short Form 12 physical component summary (SF-12 PCS) (22). The data reported for the SLAQ and SF-12 PCS reflect the interview results from the year preceding subject death or the most recent interview available prior to the end of the study followup period.
A history of renal involvement was determined based on the American College of Rheumatology (ACR) SLE criteria (23) for renal disease (based on medical records), as well as subject self-report of end-stage renal disease (ESRD) either at the baseline interview or at any point during the study followup period. Subjects were considered to have a positive history of cardiovascular disease if they reported a history of “heart disease” or a “heart attack” at the baseline interview or during the followup period.
In this analysis, the primary outcome of interest was mortality over 5 years of study followup (2003–2008). In the majority of cases, the study coordinators were notified of a subjects' death by family members at the time that the subjects were contacted to participate in the annual survey. To confirm the deaths and to obtain cause of death data, probabilistic linkage to the National Death Index (NDI) was performed for patients deceased or lost to followup prior to December 2006, the last month for which NDI data are currently available. The NDI was provided key subject data (name, date of birth, last known address) and previously-validated algorithms were used for selecting matches on the basis of the probability of a correct match.
From a combination of study and NDI records, we determined that there were 72 deaths among LOS subjects from 2002–2008. Forty-seven deaths occurred prior to December 2006 (the last month for which NDI data are currently available), and the majority of these deaths (35 of 47) were also listed in the NDI. Cause of death data, using International Statistical Classification of Diseases and Related Health Problems, Tenth Revision codes, were provided for 33 of 35 subjects (cause of death data were not listed for 2 subjects). In addition, 3 subjects who had been lost to followup were determined by NDI data to be deceased, for a total of 38 NDI listed deaths. Therefore, cause of death data were obtained for 36 of 72 deceased subjects. In the analysis, subjects lost to followup (approximately 3–5% per year) were presumed to be survivors.
The primary outcome variable for this study was mortality. Subjects were classified as deceased if they died during the followup period. Survivors included subjects who were alive as of the last contact with the study, or those who were lost to followup but not confirmed to be deceased. We compared the demographic and disease characteristics (disease activity, organ manifestations, general health status) of the 2 groups. Demographic and disease characteristics were expressed using means, medians, standard deviations, and proportions, as appropriate, and statistical tests of comparison (rank sum, t-test, or chi-square test) were employed.
Standardized mortality ratios (SMRs; ratio of the observed number of deaths to the expected number of deaths) were calculated for the overall cohort and for 5 age groups (19–34, 35–49, 50–64, 65–79, and ≥80 years). The number of expected deaths was derived from the death records from the National Vital Statistics Reports for 2006 (24).
Kaplan-Meier life table analyses, including calculation of unadjusted and age-adjusted log rank tests, were used to compare mortality rates between subjects with childhood and adult-onset SLE. Individual Cox proportional hazard models, adjusting for age and disease duration categorized as a dichotomous variable (disease duration of <10 years versus ≥10 years) were used to assess predictors of mortality. Covariates for these models were selected a priori and included childhood-onset SLE, male sex, ethnicity (white versus nonwhite), baseline measures of socioeconomic status including education (any college versus none), and insurance status (categorized as employer-based as the referent group, Medicare, and Medicaid), history of ESRD, history of cardiovascular disease, and the SLAQ score from the most recent interview. A multivariate model was constructed to determine the relative contribution of each predictor to mortality, and included age and disease duration, along with the covariates that were statistically significant (P < 0.05) in the individual Cox models. Education was chosen as the measure of socioeconomic status in the final multivariate model, however alternative models substituting other measures of socioeconomic status (poverty level and insurance source) yielded similar results. Statistical analyses were performed using SAS software, version 9.2 (SAS Institute) and STATA software, version 9.0 (StataCorp).
Baseline data were collected on 957 subjects. The mean age of the cohort was 46.8 years, 91% were female, 66% were white, 10% (n = 98) had childhood-onset SLE, and the median disease duration was 11 years. During the followup period, 72 deaths (7.5% of the subjects) occurred, including 9 deaths (12.5% of all deaths) among subjects with childhood-onset SLE. The median followup time was 48.1 months.
Differences in demographic, SLE-related, and general health characteristics between the surviving and deceased subjects are shown in Table 1. Deceased subjects were older and more likely to be male. There was no difference in mortality with regard to ethnicity. Deceased subjects had less education, were more likely to have incomes below the poverty level, and were more likely to have Medicare or Medicaid versus employer-based insurance.
|Survivors (n = 885)||Deceased (n = 72)||P|
|Age, mean (range) years†||49 (19–86)||56 (21–100)||0.0006|
|Some/graduated high school||16||32|
|Some college/trade school||45||42||0.0011|
|College degree or higher||39||26|
|Below poverty level‡||11.3||25.7||0.0008|
|Age at diagnosis, median (range) years||33 (2–75)||31 (13–87)||0.32|
|Disease duration, mean (range) years||16 (1–51)||20 (7–42)||0.0009|
|ACR renal criteria||26||40||0.008|
|End-stage renal disease||8||26||< 0.0001|
|Cardiovascular disease§||26||53||< 0.0001|
|Disease activity SLAQ score, mean ± SD¶||12.6 ± 8.0||14.6 ± 8.9||0.09|
|SLE activity score, mean ± SD||4.2 ± 2.8||4.5 ± 3.3||0.54|
|Hospitalization in the past year, %||23||51||< 0.0001|
|Childhood-onset SLE, no. (%)||89 (10)||9 (12.5)||0.51|
|General health status†|
|SF-12 PCS score, mean ± SD||38 ± 10.8||33 ± 8.4||< 0.0001|
Regarding SLE characteristics, deceased subjects had a longer disease duration than the survivors (20 versus 16 years; P = 0.0009), were more likely to meet the ACR criteria for renal disease (40% versus 26%; P = 0.008) and were more likely to have ESRD (26% versus 8%; P < 0.0001). Deceased subjects were also more likely to have reported a history of cardiovascular disease (53% versus 26%; P < 0.0001). There was no difference in the frequency of SLE flares in the 3 months prior to interview between the survivors and the deceased, and hospitalizations in the past year were much more common in the deceased group (51% versus 23%; P < 0.0001). The mean SF-12 PCS score was significantly lower in the deceased group than in the survivor group (33 versus 38; P < 0.0001).
The SMRs for all cause mortality, by age groups, are shown in Table 2. The overall SMR for the cohort was 2.5 (95% confidence interval [95% CI] 2.0–3.2). The excess mortality risk was strikingly high (20.4, [95% CI 9.3–38.7]) in the youngest age groups, but remained elevated through age 79 years.
|Age group, years||Adult-onset deaths (n = 63)||Childhood-onset deaths (n = 9)||Total deaths (n = 72)||SMR (95% CI)†|
Cause of death data have been obtained on 36 (50%) of 72 deceased subjects. SLE was coded as one of the causes of death in 19 (53%) of 36 cases, and 3 deaths were attributed directly to SLE. The most common causes of death not attributable to SLE alone included death due to circulatory disease (n = 12; including heart failure, arterial disease, and cerebrovascular disease [stroke]); infection (n = 7; including pneumonia and septicemia); malignant neoplasm (n = 4; including lung and lingual cancer and B cell lymphoma); pulmonary disease (n = 5; including pulmonary fibrosis, chronic obstructive respiratory disease, and respiratory failure); renal failure (n = 2); and other causes not related to SLE (n = 3; including “muscular dystrophy,” “exposure to uncontrolled fire,” and “poisoning by exposure to drugs and medications”).
Differences between the subjects with childhood-onset and adult-onset SLE are shown in Table 3. As has been demonstrated in a prior study, childhood-onset subjects were younger and were more likely to be male and nonwhite (17). Subjects with childhood-onset SLE were also more likely to have obtained a college degree or other higher education, and they were more likely to have employer-based health insurance. With regard to SLE characteristics, childhood-onset SLE subjects had a longer disease duration (19.5 versus 16.5 years; P < 0.0001), were more likely to meet the ACR criteria for renal disease (56.1% versus 23.6%; P < 0.0001), and were more likely to have ESRD (21.4% versus 8.4%; P < 0.0001) than adult-onset subjects. Rates of cardiovascular disease were similar between the 2 groups. Subjects with adult-onset SLE were more likely to report an SLE flare in the 3 months prior to the interview and had higher disease activity scores, but childhood-onset SLE subjects were more likely to be receiving prednisone and other immunosuppressive therapies (17) (data not shown). The mean SF-12 PCS score was significantly higher in the childhood-onset cohort (44 versus 37; P < 0.0001).
|Adult-onset (n = 859)||Childhood-onset (n = 98)||P|
|Age, mean (range) years†||52 (23–99)||33 (19–63)|
|Some/graduated high school||17||15|
|Some college/trade school||46||36|
|College degree or higher||37||49|
|Below poverty level‡||12.2||13.4||0.7|
|Age at diagnosis, median (range) years||35 (18–87)||15 (2–17)||< 0.0001|
|Disease duration, mean (range) years||16.5 (1–50)||19.5 (4–51)||< 0.0001|
|ACR renal criteria||23.6||56.1||< 0.0001|
|End stage renal disease||8.4||21.4||< 0.0001|
|SLAQ score, mean ± SD¶||13 ± 8.0||9 ± 7.9||< 0.0001|
|SLE activity score, mean ± SD||4.4 ± 2.8||3.0 ± 2.8||< 0.0001|
|Hospitalization in the past year||25||24||NS|
|General health status†|
|SF-12 PCS score, mean ± SD||37 ± 10.7||44 ± 9.0||< 0.0001|
|No. (%) of deaths||63 (7)||9 (9)||NS|
|Age at death, mean (range) years||52 (23–100)||33 (19–49)||< 0.0001|
Nine (9%) of 98 subjects with childhood-onset SLE died in the followup period, and are described in Table 4. The median age of death was 32 years (range 21–51 years) and the median disease duration was 15 years (range 8–36 years).
|Patient||Sex||Age at diagnosis, years||Age at death, years||Disease duration, years||History of ESRD||History of a “heart attack”||Cause of death|
The Kaplan-Meier survival curves for childhood- versus adult-onset disease are shown in Figure 1. Results from an unadjusted log rank test demonstrated no significant difference in the mortality rates between childhood- and adult-onset SLE (P = 0.34), but after controlling for age, the differences in mortality rates between childhood- and adult-onset SLE became statistically significant (P < 0.0001).
As shown in the first column of Table 5, in Cox proportional hazards models adjusting for age and disease duration at baseline, childhood-onset SLE, male sex, education, Medicare or Medicaid insurance, and a history of ESRD and cardiovascular disease were all associated with a higher risk of death in the followup period. Ethnicity and SLE disease activity were not associated with increased mortality risk. In the full multivariate model, childhood-onset SLE, education, and ESRD remained significant predictors of mortality.
|Variable||Individual models HR (95% CI)†||Full multivariate model HR (95% CI)‡|
|Childhood-onset SLE||3.5 (1.5–8.0)§||3.1 (1.3–7.3)§|
|Male||2.2 (1.2–4.0)§||1.6 (0.8–3.0)|
|Education||2.1 (1.2–3.4)§||1.9 (1.1–3.2)§|
|Below poverty level||2.4 (1.4–4.3)§|
|End stage renal disease||2.0 (1.2–3.4)§||2.1 (1.1–4.0)§|
|Cardiovascular disease||1.9 (1.2–3.1)§||1.6 (0.8–3.1)|
|SLE activity||1.06 (0.9–1.1)|
In this study, we examined the causes and predictors of mortality in a large community-based cohort of patients with SLE, which included a substantial number of subjects with childhood-onset SLE. We found that childhood-onset SLE was a significant predictor of mortality in our population. This finding highlights the importance of identifying modifiable risk factors to prevent early mortality among patients with childhood-onset SLE.
During the first 5 years of study followup, 7.5% of the LOS subjects died. This study confirms some of the previously identified predictors of increased mortality in SLE, including demographic characteristics such as male sex, lower educational attainment and income, and disease-related factors, including the presence of cardiovascular disease and ESRD, and poorer general health (5, 25–28).
The SMR for the cohort was 2.5, which is consistent with the SMRs calculated from other large SLE cohorts (5, 6). A 2006 study by Bernatsky et al utilizing the Systemic Lupus International Collaborating Clinics (SLICC) international multi-site cohort calculated an unadjusted SMR of 2.4 across participating sites, and an SMR of 2.2 for sites in the US (8). The SLICC cohort study also noted a particularly high SMR of 19.2 for adolescents and young adults (ages 16–24 years) with SLE. In our cohort, the SMR for subjects in the youngest age category (19–34 years) was also strikingly high at 20.4, and this calculation included 6 deaths among subjects with childhood-onset SLE. Among the subjects for whom cause of death could be determined, approximately half of the deaths were due to cardiovascular disease or infection, which is consistent with other studies describing causes of death in SLE (3, 5, 8, 29–32). SLE was listed as a cause of death in just over half of the subjects for whom cause of death records were available. At least 1 prior study has demonstrated an under-reporting of SLE as the cause of death on death certificates (33).
Nine subjects with childhood-onset SLE died during the followup period. It is both striking and disturbing that all of the deaths in the childhood-onset cohort occurred in young patients (range 21–51 years), with a median disease duration of 16 years. Three of these subjects self-reported a history of a “heart attack,” including one subject who died at age 26.
Given the difficulties inherent in conducting longitudinal, long-term followup studies of patients with childhood-onset rheumatic disease, little is known about the outcomes of patients with childhood-onset SLE once they reach adulthood. Tucker et al compared differences in morbidity and mortality among adolescent- and adult-onset SLE patients in the LUpus in Minorites: NAture versus nurture cohort, who were followed for an average of 6.8 and 5.6 years after diagnosis, respectively (16). The adolescent-onset cohort, who aged into early adulthood during the period of study followup, had increased disease-related damage as measured by a SLICC/ACR damage index score of 2.3 (versus 1.6 for the adult-onset group). Unadjusted mortality rates were almost twice as high among patients with adolescent-onset disease, although these differences did not reach statistical significance. The findings from the present study, which compares childhood- and adult-onset subjects in the context of a much longer mean disease duration, suggest that childhood-onset SLE patients continue to be at risk for early mortality over the duration of their disease. In addition, after adjusting for several important covariates, childhood-onset SLE was an independent predictor of mortality in our cohort.
Why would childhood-onset SLE subjects be at risk for early mortality, and what factors (e.g., differences in demographic or disease-related factors) explain this risk? As compared with adult-onset SLE, childhood-onset SLE subjects are more likely to be male and nonwhite, both previously identified risk factors for mortality in SLE. With regard to disease factors, one could presume that increased disease damage (e.g., increased rates of ESRD), earlier development of disease-related morbidities (e.g., cardiovascular disease) and longer exposure to SLE-related therapies and side effects could all predispose these patients to an increased risk of death at an earlier age, as compared with patients with adult-onset SLE with the same disease duration. However, after controlling for many of these factors in the multivariate model, childhood-onset SLE remained an independent predictor of mortality. This suggests that there must be another explanation for the increased mortality risk among patients with childhood-onset SLE. One hypothesis is that genetic differences, such as increased genetic susceptibility, leads to earlier disease onset and increased disease severity among childhood-onset SLE patients. Another possibility is that pediatric patients are more biologically vulnerable to the effects of SLE and its treatment.
Although this study is one of the first to examine the long-term mortality associated with childhood-onset SLE, there are several important limitations. First, because this study relies on subject self-reported outcomes, inaccuracies in subject reporting may occur. This limitation was addressed in part by validating a subset of the self-reported outcomes through chart review. Second, the LOS is not an inception cohort, so the conclusions should be interpreted in the context of a potential survival bias, because mortality data are only available over the period of study followup, and not from the date of diagnosis. Given the increased risk of early mortality among patients with childhood- versus onset-SLE, our sample may underestimate the true differences in long-term mortality of childhood- and adult-onset SLE, as the sicker childhood-onset SLE patients were less likely to have survived to enrollment in the LOS. Third, cumulative disease damage was not assessed in this study. The increase in mortality among childhood-onset SLE subjects could be explained by increased disease damage, as prior studies have shown an increase in cumulative disease damage in childhood versus adult-onset SLE (14–16). However, ESRD is a significant marker of disease damage and mortality in SLE, and this was included as an independent predictor of mortality in the multivariate model (34). Finally, given the small number of deaths in the childhood-onset cohort, our study was not powered to identify particular risk factors associated with mortality among the childhood-onset group.
In conclusion, our findings indicate that SLE patients with childhood-onset disease were at high risk for mortality at an early age. These results highlight the need for long-term, longitudinal studies of patients with childhood-onset chronic disease that better delineate the modifiable disease-related, behavioral, and health-care system factors that contribute to less favorable outcomes for patients with childhood-onset SLE. Studies are needed to develop prognostic models to identify patients at high risk for significant morbidity or early mortality for both childhood- and adult-onset SLE, and to design interventions to improve the quality of care for this vulnerable group of patients.
All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be submitted for publication. Dr. Hersh 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 conception and design. Hersh, Trupin, Yazdany, Panopalis, Julian, Criswell, Yelin.
Acquisition of data. Trupin, Julian, Katz, Criswell, Yelin.
Analysis and interpretation of data. Hersh, Trupin, Yazdany, Panopalis, Julian, Yelin.
The authors gratefully acknowledge the contributions of Stuart Gansky, DrPH, and Janet Stein, Stephen King, Jessica Spry, and Rosemary Prem.