To determine the features associated with mortality in a multiethnic US cohort of patients with systemic lupus erythematosus (SLE) within 5 years of study onset.
To determine the features associated with mortality in a multiethnic US cohort of patients with systemic lupus erythematosus (SLE) within 5 years of study onset.
Socioeconomic and demographic features (age, gender, ethnicity, marital status, education, occupation, poverty, and health-related behaviors [drinking, smoking, exercising]), clinical and immunologic features (disease duration, disease onset type, disease activity according to the Systemic Lupus Activity Measure [SLAM], disease damage according to the Systemic Lupus International Collaborating Clinics [SLICC] Damage Index [SDI], number of American College of Rheumatology criteria at diagnosis, organ system manifestations, fatigue and pain ratings, and medication usage and autoantibodies), immunogenetic features (HLA class II genotypes), and behavioral and psychosocial features (social support, illness-related behaviors, and helplessness), as obtained at enrollment into the study, were compared between survivors and deceased patients. Logistic regression analysis was used to determine significant independent risk factors for mortality.
Within 5 years of study onset, 34 of 288 patients have died. Fourteen deaths could be directly attributed to SLE and 11 to infections. In 1 patient the cause of death could not be determined. In the remaining 8 patients the cause of death was neither infectious nor disease-related. There were 10 deaths among Hispanics, 18 among African Americans, and 6 among Caucasians (P< 0.05). Variables associated with mortality in the univariable analyses included poverty, less than full-time employment, difficulty in accessing health care, shorter disease duration, cardiovascular and renal involvement, higher serum creatinine levels and lower hematocrit values, higher SLAM and SDI scores, lower use of antimalarial drugs, and higher use of (some) immunosuppressants. Specific autoantibodies and class II HLA genotypes were not associated with mortality. Poverty and higher baseline SLAM and SDI scores were independently associated with mortality in the multivariable analyses.
Disease activity, disease damage, and poverty appear to be the most important determinants of mortality in this multiethnic US cohort of SLE patients. These results have applicability to the management of patients with SLE, a disease that more severely affects disadvantaged minority population groups.
The overall probability of survival in patients with systemic lupus erythematosus (SLE) has improved dramatically over the last few decades(1–10). However, there are still patients with SLE who succumb relatively early in the course of their disease, either to overwhelming illness-related manifestations or to treatment-related complications(7, 11–18). Factors reported to be associated with early death in SLE include older age(7), non-Caucasian ethnicity(3, 7, 12, 17, 19–25), male sex(14, 26), poor socioeconomic status(14), renal or cardiovascular involvement(17, 27), vasculitis(28), anemia, increased serum creatinine(2, 7, 29, 30), renal biopsy findings (e.g., high chronicity index, high activity index)(31–34), cytopenias (2,7,15,29,35), and some immunologic parameters (e.g., low complement levels)(16, 23, 36). There is no consensus as to whether a poor outcome in SLE is primarily related to genetic features, to socioeconomic features, or to a combination of both. Most studies to date have included data obtained in populations with little or no diversity in terms of ethnicity or socioeconomic features (4,6,7,9,11,13,14,19,20,25–27,29,37–47) and thus may have limited generalizability.
We have constituted a multiethnic (Hispanic, African American, and Caucasian) cohort of patients with SLE and are following its members over time to determine the factors associated with a given disease course and outcome(48–53). We now present the features associated with mortality within 5 years of study onset in this multiethnic US cohort of SLE patients.
Lupus in minority populations, nature versus nurture (LUMINA) is a longitudinal study of outcome in SLE patients. The constitution of this cohort, the variables obtained, and the frequency of followup visits have been described in detail previously(48–53). In short, patients of either Hispanic (Mexican or Central American ancestry), African American, or Caucasian ethnicity, from either Texas (The University of Texas–Houston Health Science Center and The University of Texas Medical Branch, Galveston) or Alabama (The University of Alabama at Birmingham) and their catchment areas, with SLE according to the American College of Rheumatology (ACR, formerly the American Rheumatism Association) criteria(54) and disease of 5 years duration or less, already receiving care at these centers or presenting to them for the first time, were eligible to participate.
As has been described before, virtually all patients at the 3 institutions who were eligible to participate in the study, and who were invited, were enrolled. In addition, community rheumatologists in Birmingham, Alabama, and in Houston, Harlingen, and Corpus Christi, Texas, were invited to refer subjects for the study. Recruitment into the cohort took place initially between April 1994 and January 1996 and was resumed in July 1997. Prior to the enrollment visit, all previously available medical records were reviewed. This was done in order to confirm patients' eligibility and to gather as much information as possible relative to the patients' socioeconomic, demographic, and clinical features before disease onset and the enrollment visit.
Data were obtained from the following domains: socioeconomic and demographic, clinical and immunologic, immunogenetic, behavioral, and cultural. From the socioeconomic and demographic domain the following were obtained: age, sex, ethnicity, marital status, education (number of years of formal education), occupation, housing (ownership and density), income, health insurance (type and degree of coverage), health-related behaviors (smoking, drinking, exercising), and access to health care (distance from health care providers and perceived difficulty in accessing care). Income and number of persons in the household were used to define poverty according to US federal government (Department of Commerce) guidelines(55).
From the clinical and immunologic domain, the following were ascertained: disease onset (time at which a patient met 4 ACR criteria for the diagnosis of SLE), disease onset type (abrupt if manifestations evolved over 4 weeks or less; insidious otherwise), number of ACR criteria at diagnosis, presence of autoantibodies (antinuclear antibodies, anti-U1 RNP, anti-Smith, anti-Ro, anti-La, antiphospholipid antibodies [IgG and IgM isotypes], and the lupus anticoagulant), disease activity according to the Systemic Lupus Activity Measure (SLAM)(56), and disease damage, irrespective of its cause, according to the Systemic Lupus International Collaborative Clinics (SLICC) Damage Index (SDI)(57), measured only in patients with disease duration of 6 or more months at the enrollment visit, according to instrument requirements. Organ system manifestations and their attribution (lupus, treatment, or neither), fatigue (using the 9-item validated instrument described by Krupp&lsqbr;58&rsqbr;), pain (using a 10-cm visual analog scale), comorbidities, medications administered, and hospitalizations were also ascertained. In the immunogenetic domain, HLA–DR and HLA–DQ genotypes and HLA class III allotypes were determined(59, 60).
Finally, from the behavioral and cultural domain, the following variables were ascertained: social support (according to the Interpersonal Support Evaluation List, which includes 4 scales—appraisal, belonging, tangible, and self-esteem—and a summary measure of overall social support&lsqbr;61&rsqbr;), helplessness (according to the Rheumatology Attitude Index&lsqbr;62&rsqbr;), and illness-related behaviors (according to the Illness Behavior Questionnaire, which provides a summary measure with higher scores indicating more abnormal illness-related behaviors&lsqbr;63&rsqbr;).
All variables were ascertained at the enrollment visit. In addition, disease activity was ascertained at the time of disease diagnosis, utilizing all available medical records. For the analyses presented here, disease duration was also computed from the time of diagnosis (per ACR criteria) to the last visit (or to death, if applicable). For patients with less than 6 months of disease duration at the enrollment visit, the first available SDI score was used in the multivariable analyses (see below).
Followup visits were conducted every 6 months for the first 2 years of the study, and yearly thereafter. At each visit, a new release-of-information form was completed so that records could be obtained, if office visits or hospitalizations had taken place in the interval. Members of the LUMINA cohort were encouraged to maintain contact with study coordinators and to notify them of significant life- or illness-related events, including hospitalizations. Study coordinators maintained the interest of participants in the study through birthday and Christmas cards, as well as a semiannual newsletter (LUMINEWS). Study coordinators were also responsible for retrieving all available medical records prior to each followup visit, or prior to completing a death form, as the case may be.
Investigators at each site determined the cause of death and completed the corresponding study form. To this end they utilized all available medical records, as well as information from the death certificates. Subsequently, these forms were reviewed by the LUMINA principal investigator (GSA). The final adjudication of the main cause of death was based on the information provided on these forms plus all available clinical information from the study visits preceding the patient's demise. Three categories were considered: death due primarily to disease activity, death due primarily to an infectious process, and death due primarily to neither disease activity nor an infectious process; within this third category all vascular (cardiovascular or cerebrovascular) events were included.
Comparisons between deceased and surviving patients were done using the t-test and chi-square test for continuous and categorical variables, respectively. These analyses were done for all patients and by ethnic group. Fisher's exact test was used for categorical variables when appropriate. Variables with a P less than or equal to 0.10 in the univariable analyses, and those that were felt to be clinically relevant, were entered into multivariable logistic regression models. Also included in these models (regardless of significance) were age and ethnicity, our a priori exploratory variables of the outcome of interest, and disease duration. Several multivariable models were constructed. This was because some variables were felt to be related to the course and outcome of the disease instead of being true predictors of the final outcome (hospitalizations and medications, for example), and thus some models were run with and without these variables. Finally, we were interested in examining the possible predictive value of disease activity over the duration of followup, so additional models were constructed using “SLAM average” (disease activity at diagnosis, at enrollment, and at subsequent scheduled visits). Analyses were also performed using a time-dependent approach (Cox regression). Results were comparable to those obtained by multivariable logistic regression and therefore are not presented.
As shown in Table 1, 34 of 288 LUMINA patients (11.8%) have died within 5 years of study onset. Of those, 10 of 82 were Hispanics (12.2%), 18 of 120 were African Americans (15.0%), and 6 of 86 were Caucasians (7.0%) (P > 0.05). The corresponding survival curves for all patients within each ethnic group are shown in Figure 1 and Figure 2. The differences between these survival curves were not significant (P > 0.05). Time from disease onset to death was longer for the Hispanics (38.3 months) than for the African Americans (35.8 months) and Caucasians (32.8 months); these differences were, however, not statistically significant.
|Hispanic (n = 82)||African American (n = 120)||Caucasian (n = 86)||All (n = 288)|
|Time to death, months, mean ± SD||38.3 ± 28.1||35.8 ± 35.0||32.8 ± 28.3||36.1 ± 31.0|
|Causes of death, n (%)|
|Disease manifestations||8 (80.0)||5 (27.8)||1 (16.7)||14 (41.2)|
|Infectious processes||1 (10.0)||8 (44.4)||2 (33.3)||11 (32.4)|
|Other*||1 (10.0)||4 (22.2)||3 (50.0)||8 (23.5)|
|Undetermined||0 (0)||1 (5.6)||0 (0)||1 (2.9)|
|Total (%)||10 (12.2)||18 (15.0)||6 (7.0)||34 (11.8)|
As shown in Table 1, 14 of the deaths (41.2%) were considered to be related mainly to the disease, whereas 11 (32.4%) were considered to be primarily related to an infectious process. In 1 of the remaining 9 deaths, a cause could not be determined. Of the other 8 deaths, 6 (17.6%) were of cardiovascular or cerebrovascular etiology, 1 was due to a malignant process (melanoma), and 1 was due to cirrhosis. Hispanics were more likely to die of disease manifestations and African Americans were more likely to die of an infectious process; these differences were, however, not statistically significant (P > 0.05).
Associations of mortality with selected baseline socioeconomic and demographic features at enrollment into the study are shown in Table 2. There were no differences in sex, marital status, years of education, or home ownership between those patients who were alive and those who had died. Deceased patients were, overall, older (40.2 years) than the surviving patients (35.7 years), but these differences were not statistically significant. There was a significantly higher proportion of patients below the poverty line (64.0%) among the deceased than among the surviving patients (31.6%) (P = 0.002). There was also a significantly lower proportion of patients working full time among those who did not survive (17.2%) than among the surviving patients (37.5%) (P = 0.039).
|Surviving (n = 254)||Deceased (n = 34)||P value*|
|Sex, % women||88.2||91.2|
|Age, years, mean ± SD||35.7 ± 12.4||40.2 ± 16.9||0.06|
|Married/living together, %||44.6||41.4|
|Education, years, mean ± SD||12.5 ± 3.1||10.8 ± 3.3|
|Housing, ownership, %||63.4||67.9|
|Occupation, working full time, %||37.5||17.2||0.039|
|Below poverty line, %||31.6||64.0||0.002|
|Health insurance, %||72.9||58.1||0.095|
|Access to health care, % >20 miles away||48.1||42.9|
|Perception of difficulty in obtaining health care, %||3.8||21.4||0.002|
|Does not drive self to see doctor, %||42.6||61.5||0.094|
|Prior hospitalizations, %||26.0||67.9||<0.0001|
|Unhealthy behaviors,† %||67.7||64.7|
Selected health care variables are also shown in Table 2. There were no differences in health insurance coverage at enrollment into the study between surviving and deceased patients. The deceased patients were significantly more likely to have had prior hospitalizations for SLE than the surviving patients (67.9% versus 26.0%, respectively) (P < 0.0001). There were no differences in the proportion of patients living more than 20 miles away from health care providers as a function of survival status. However, there was a higher proportion of patients who perceived some degree of difficulty in obtaining health care among those who did not survive (21.4%) than among the surviving patients (3.8%) (P = 0.002). The deceased were more likely not to have driven themselves to see their physicians (61.5%) than the surviving patients (42.6%) (P = 0.094).
Selected clinical features are shown in Table 3. The number of ACR criteria and disease onset type were comparable among the surviving and deceased patients. Disease duration, measured either from diagnosis to enrollment into the study or from diagnosis to the time of death (or the last visit), was shorter among the deceased (12.9 and 36.1 months) than among the surviving patients (19.7 and 54.4 months) (P < 0.03 and 0.007, respectively). Organ system manifestations attributable to lupus were comparable among deceased and surviving patients, with the exception of cardiovascular (79.4% versus 59.0%) and renal (82.4% versus 53.7%) manifestations, which were more frequent among the patients who did not survive than among the surviving patients (P = 0.024 and 0.002, respectively). Baseline serum creatinine values had been higher among the deceased than they were among the surviving patients. The hematocrit values, in contrast, were lower among the deceased than among the surviving patients (P < 0.0001, in both cases). None of the autoantibodies studied (anti-dsDNA, anti-Sm, anti–U1 RNP, anti-Ro, anti-La, IgG, and IgM anti-phospholipid antibodies and the lupus anticoagulant) were found with higher frequency among the deceased than among the surviving patients (data not shown). HLA–DR and HLA–DQ alleles were examined for all patients and by ethnic group. None of the alleles examined were particularly overrepresented or underrepresented in the deceased relative to the surviving patients, and this was the case for all patients and by ethnic group (data not shown).
|Surviving (n = 254)||Deceased (n = 34)||P value*|
|Disease duration, months, mean ± SD|
|To enrollment||19.7 ± 16.7||12.9 ± 16.6||0.03|
|To last visit (or death)||54.4 ± 29.0||36.1 ± 31.0||0.007|
|Disease onset type, % abrupt||45.0||58.1|
|Number ACR criteria,† mean ± SD||5.5 ± 1.3||5.9 ± 1.5|
|Manifestations by organ system attributable to SLE, %|
|Serum creatinine (mg/dL), % of patients|
|Hematocrit (%), % of patients|
Assessments of disease activity and disease damage at different times are shown in Table 4. There were higher SLAM scores and SLAM global measures among the deceased than among the surviving patients. This was the case at diagnosis, at enrollment into the study, and across all visits (SLAM average). The SLAM scores and globals at enrollment into the study, and across time (average), were statistically significant (P ranging from 0.0065 to less than 0.0001). Likewise, the SDI scores (as well as the damage global) were higher among the deceased than among the surviving patients, and these differences were statistically significant (P = 0.0002 and P < 0.0001, respectively).
|Surviving (n = 254)||Deceased (n = 34)||P value*|
|SLAM score, mean ± SD|
|Enrollment||9.8 ± 5.2||17.6 ± 7.8||< 0.0001|
|Diagnosis||12.3 ± 6.8||13.5 ± 5.6|
|Average||9.5 ± 3.7||15.8 ± 5.3||< 0.0001|
|SLAM MD global, 0–10 cm, mean ± SD|
|Enrollment||2.6 ± 2.0||4.7 ± 3.4||0.001|
|Diagnosis||3.8 ± 2.8||5.0 ± 2.5|
|Average||2.8 ± 1.4||5.0 ± 2.3||< 0.0001|
|SLAM patient global, 0–10 cm, mean ± SD|
|Enrollment||3.5 ± 2.8||4.5 ± 3.5||0.0584|
|Average||4.4 ± 2.2||5.6 ± 2.9||0.0065|
|SDI, enrollment, mean ± SD||0.7 ± 1.2||1.8 ± 2.1||0.0002|
|Damage assessment, MD global, 0–10, mean ± SD||1.7 ± 3.1||4.7 ± 4.3||< 0.0001|
Past and current use of medications at enrollment into the study are shown in Table 5. Past or current use of antimalarial drugs was more common among the surviving patients (58.3% and 49.0%) than it had been among those who were deceased (22.6% and 16.1%) (P = 0.0002 and P = 0.0004 for past and current, respectively). Corticosteroid use, as well as the mean current dose, was comparable in the 2 groups. The mean highest daily dose of corticosteroids (prednisone equivalent) had been higher among the deceased (63.8 mg) than it was among the surviving patients (45.9 mg), and these differences were statistically significant (P = 0.04). Current, but not past, use of azathioprine had been more common among the deceased (13.3%) than it was among the surviving patients (6.6%) (P = 0.03). The proportion of patients receiving intravenous cyclophosphamide pulses at enrollment in the study was higher among the patients who did not survive than among the surviving patients (24.1% versus 14.7%), but these differences were not statistically significant. The mean number of cyclophosphamide pulses was comparable for the 2 groups when all patients were considered; however, if only those patients receiving pulses were considered, those who survived had received more pulses (5.9 ± 3.0) than those who died (3.2 ± 2.3) (P = 0.04).
|Surviving (n = 254)||Deceased (n = 34)||P value*|
|Hydroxychloroquine, % use|
|Corticosteroids, % use|
|Highest daily dose (ever), mg, mean ± SD||45.9 ± 38.0||63.8 ± 57.0||0.04|
|Mean daily dose, mg, mean ± SD||12.7 ± 17.1||19.6 ± 22.9|
|Azathioprine, % use|
|Cyclophosphamide, % use||14.7||24.1|
|All patients, number of pulses, mean ± SD||0.9 ± 2.4||0.8 ± 1.8|
|Only patients receiving pulses, mean ± SD||5.9 ± 3.0||3.2 ± 2.3||0.04|
Fatigue ratings, pain ratings, abnormal illness-related behaviors, social support, and helplessness at enrollment into the study were, overall, comparable among surviving and deceased patients (data not shown).
The results of the multivariable analyses are shown in Table 6. Variables consistently found to significantly affect early mortality in our LUMINA cohort include increased disease activity (higher SLAM scores), increased disease damage (higher SDI scores), and poverty. Of these variables, the one associated with the smallest P value was poverty, regardless of the model used (P ranging from 0.0167 to 0.0013).
|Variable||Odds ratios||Confidence intervals||P value|
|Below poverty line||4.06||1.50–11.01||0.0059|
|SLAM score at enrollment||1.09||1.01–1.17||0.0194|
|SDI index (first computed)||1.45||1.19–1.91||0.0094|
We have reported the mortality data for Hispanic, African American, and Caucasian SLE patients constituting the LUMINA cohort relatively early in their disease course (within 5 years of study onset). This is the first time in which these 3 ethnic groups have been studied simultaneously using the same protocol(48–52). Although the difference was not significant, the death rate was higher among African Americans and Hispanics than among Caucasians. The survival curves for the 3 ethnic groups paralleled these data. This is in concordance with data from other centers. Table 7 summarizes the survival experience of other lupus cohorts(3, 5, 7–12, 14, 16, 19, 22, 24–26, 29, 32, 35, 38, 41, 42, 57, 64–70). Although our 5-year survival rate is lower than the one described in primarily Caucasian cohorts (Europe, Canada), it is the same as or better than the survival rates for contemporary US cohorts that have a significant representation of African Americans, and it is much higher than the survival rates for cohorts from developing countries.
|Authors (ref)||Year||Place||n||Ethnicity||%||Comments†||5-year survival, %|
|Ginzler et al(3)||1982||US (multicentric)||1,103||Caucasian||58||Inception 1965–1976 cohort||77|
|African American||32||Risk factors: anemia, proteinuria, no.|
|Other||10||criteria, and low SES|
|Reveille et al(7)||1990||US (Alabama)||389||Caucasian||51||Established 1975–1984 cohort||89|
|African American||48||Risk factors: older age, thrombocytopenia,|
|Asian||1||African American ethnicity|
|Pistiner et al(64)||1991||US (California)||570||Caucasian||72||Established 1980–1989 cohort||97|
|African American||11||Risk factors: renal involvement,|
|Kumar et al(42)||1992||India (New Delhi)||163||Asian Indian||100||Established 1981–1990 cohort||68|
|Risk factors: disease activity|
|Anstey et al(23)||1993||Australia||22||Australian Aborigine||100||Established 1984–1991 cohort Risk factors: not determined||60|
|Massardo et al||1994||Chile (Santiago)||218||Hispanic‡||100||Established 1969–1991 cohort||87|
|(32)||Risk factors: disease activity, renal involvement|
|Abu-Shakra et al||1995||Canada (Toronto)||665||Caucasian||87||Established 1970–1994 cohort||93|
|(8)||African American||7||Risk factors: renal and lung involvement,|
|Asian||6||disease activity, older age|
|Ward et al(14)||1995||US (North Carolina)||408||Caucasian||51||Inception 1969–1983 cohort||82|
|African American||49||Risk factors: older age, male sex, low SES|
|Huicochea et al||1996||Mexico (Mexico||65||Hispanic‡||100||Inception 1970–1993 pediatric cohort||60|
|(12)||City)||Risk factors: not determined|
|Blanco et al(26)||1998||Spain (Madrid,||306||Caucasian||100||Established 1975–1993 cohort||90|
|Santander, and Alicante)||Risk factors: male sex, renal and central nervous system involvement|
|Uramoto et al(9)||1998||US (Minnesota)||48||Caucasian||94||Population-based 1980–1992 cohort||91|
|Other||6||Risk factors: not determined|
|Cervera et al(38)||1999||Europe||1,000||Caucasian||100||Established 1990 cohort||95|
|(multinational)||Risk factors: immunologic parameters (low complement, high anti-DNA antibodies)|
|Jacobsen et al||1999||Denmark||513||Caucasian||100||Established 1975–1995 cohort||91|
|(41)||(multicentric)||Risk factors: younger age|
|Mok et al(29)||2000||China (southern)||186||Asian||100||Inception 1992–1999 cohort||93|
|Risk factors: thrombocytopenia, corticosteroid use|
|Bellomio et al||2000||Argentina||366||Hispanic‡||100||Established 1990–1998 cohort||91|
|(25)||(multicentric)||Risk factors: cardiovascular and renal damage, hyperlipidemia|
|Alarcón et al||2001||US (Alabama and||288||Hispanic‡||28||Established/inception 1993–1999 cohort||86|
|(this paper)||Texas)||African American||41||Risk factors: older age, disease activity,|
As in other published series, the causes of death among our LUMINA patients related primarily to either unrelenting disease activity or infections (5,7,10,12,16,24,35,37,38,41,45,66,67,69,71–76). Although we realize the difficulties in adjudicating the cause of death in lupus patients, we have followed a consistent method of ascertainment and have included information from all available medical records as well as from death certificates. Infections may occur because of medications or because of genetically mediated mechanisms(77); such is the case for some mannose-binding lectin polymorphisms and complement deficiencies(78–80). Lupus patients are particularly prone to develop, for example, Salmonella sepsis or disseminated mycobacterial fungal or parasitic infections(71–76, 81). None of our patients experienced such infections but, rather, succumbed to infections with common microbial organisms.
Perhaps the main limitation of our study relates to the fact that the LUMINA cohort is not truly an inception cohort, because patients with disease up to 5 years in duration were enrolled. It can be argued, therefore, that patients who succumbed to lupus shortly after disease onset may not have been included in the cohort. We do not believe that this is the case, because we have maintained a very active surveillance for new lupus patients at our participating centers, and new patients, including those with very abrupt disease onset and catastrophic course, have been recruited into the cohort.
Another limitation of our study is that we did not measure the variables at disease onset or diagnosis (except for those patients who were enrolled into the study within weeks or months of disease onset) but rather at the somewhat arbitrary point in time at which patients were enrolled into the study. However, many of the variables, specifically the socioeconomic, demographic, and immunogenetic variables, are either relatively or definitively stable over time(39, 82, 83). Therefore, possible changes in some of these variables over the duration of the study are likely to have a negligible impact. Other variables are so tightly related to the course of the disease that it is hard to consider them true predictors of mortality; hospitalizations and medications fall within this category (patients receiving immunosuppresants or very high doses of corticosteroids could have died because of a superimposed infection or because of unrelenting disease activity despite the medications)(16, 38). Thus, the final multivariable model presented is one in which these variables have been excluded.
Our results were consistent regardless of the model examined. Disease activity as measured by the SLAM score, disease damage as measured by the SDI, and poverty as defined by the US government(55) emerged as significant predictors of mortality in our LUMINA patients. We were not surprised to find that disease activity, as measured by the SLAM, was predictive of mortality in our patients, but we were somewhat surprised that disease damage was retained consistently in the models explored, despite the fact that most of our patients died either as a consequence of active disease or because of an infectious process. Damage as measured by the SDI has not been previously reported to be associated with decreased survival. However, clinical manifestations (such as end-stage renal disease and atherosclerotic cardiovascular disease&lsqbr;8, 84–87&rsqbr;) that are scored in the SDI have been associated with mortality. These data suggest that patients who maintain overall high levels of disease activity at this point in time are also prone to accrue damage and to succumb to disease manifestations or to disease-related or treatment-related complications. Despite this apparent relationship between disease activity and damage, the interaction term between these 2 variables was not retained in the models in which it was entered. This probably relates to the strength of the associations; disease activity was a stronger predictor of mortality than disease damage. In contrast to most other published studies and earlier analyses of predictors of mortality in the LUMINA cohort(7, 14, 17, 34, 35, 47, 88–91), age was only of borderline significance in the univariable analyses and was not significant in the multivariable analyses performed.
Although abnormal hematocrit and serum creatinine values were significantly associated with mortality in the univariable analyses, they were not significant in the multivariable analyses, suggesting that the SLAM and the SDI are more comprehensive measures of the overall impact of the disease and do not represent just the involvement of one particular organ system or a given disease manifestation. A similar explanation may account for our failure to identify other specific organ system manifestations or laboratory abnormalities (e.g., antiphospholipid antibodies) as predictors of the outcome of interest. It should be noted, however, that the values entered into the models were those obtained at entry into the study and not those obtained immediately prior to the patient's demise. Medications were not consistent predictors, although hospitalizations were; as noted, both can be regarded as proxies for more severe disease. None of the psychosocial parameters or overall measures of distress (pain, fatigue, and helplessness) were retained in the multivariable models; this contrasts with some earlier analyses of predictors of mortality in our LUMINA cohort(90, 91). We do not have a good explanation for these discrepant findings.
The lack of association between mortality early in the course of the disease, specific autoantibodies, and HLA–DR and DQ alleles is noteworthy, suggesting that these parameters may not be particularly important predictors of this outcome at this point in time (unless they predispose to intervening events such as infections). It should also be noted that the genetic markers associated with disease activity at disease onset were not the same across the 3 ethnic groups(52); therefore, ethnic-specific analyses need to be performed if the role of these genetic markers is to be determined. Relatively small numbers within each specific ethnic group make such analyses less likely to provide valid information. Of course, similar reasoning may be applied to all other variables—that is, they were examined in the context of the entire LUMINA cohort in the multivariable analyses, rather than within each ethnic group.
The most important finding of our study is that, despite the apparent association of non-Caucasian ethnicity with mortality in our LUMINA cohort, poverty and not ethnicity was the variable consistently found to be significant in the multivariable analyses of our data. Of course it can be argued that disease activity and disease damage reflect to a certain extent the ethnic background of our patients. However, analyses of our data suggest that the important variables, which mediate disease activity over the course of the disease, are socioeconomic and demographic rather than ethnic/racial or genetic. The definition of poverty that we used in our study is the one established by the US Department of Commerce and utilized in population studies. Individuals are classified as being above or below the poverty line based on family income and size. We, of course, recognize that there are other approaches to defining poverty or socioeconomic status. Indeed, when income was determined as the median of the household income for the corresponding census tract, it was also found to be a significant predictor of mortality in a lupus cohort, although not as powerful a predictor as it was in our study(14). More recently, it has been proposed that a more sensitive measure of financial resources and socioeconomic status is the measurement of wealth rather than income (adjusted to family size)(92–94). In fact, even among individuals below the poverty line, ethnic minorities tend to be less wealthy than individuals from the ethnic majority, as these individuals are more likely to have fewer assets and more financial liabilities than individuals from nonminority groups(92–94). Unfortunately, we did not collect these data on the members of the LUMINA cohort during the early years of our study (but we are doing so at present); thus we do not have data to support these assertions.
Our findings regarding poverty in this cohort are consistent with a body of literature that indicates that, regardless of the condition being studied, patients with a low socioeconomic status fare much worse than those with an overall better socioeconomic status(95, 96). As to lupus, in particular, lack of health insurance has been shown to be associated with increased disease activity and disease damage in a cross-sectional study designed to include patients from the 2 major US ethnic groups (African American and Caucasian)(39). Special attention to these socioeconomic and demographic features is warranted when managing lupus patients with characteristics similar to those that we have studied. That is to say, in addition to the pharmacologic therapy chosen, special emphasis should be placed on adherence to the regimen prescribed, identification of flares or comorbid events, family and social support, and the like. Our data thus have direct applicability to the management of patients with SLE, a disease known to more commonly affect disadvantaged ethnic minorities.
Current LUMINA investigators and staff:
At the University of Alabama at Birmingham: Graciela S. Alarcón, MD, MPH, Holly M. Bastian, MD, MSPH, Barri J. Fessler, MD, Gerald McGwin, Jr., MS, PhD, Jeffrey Roseman, MD, PhD, MPH, Alfred A. Bartolucci, PhD, Martha Sanchez, MD, MSc, and Ellen Sowell, Research Coordinator.
At the University of Texas–Houston Health Science Center: John D. Reveille, MD, Alan W. Friedman, MD, Chul Ahn, PhD, Jo McLain, BSN, RN, Robert Sandoval, BA, and Rudyard Lanete, BA.
At the University of Texas Medical Branch at Galveston: Bruce Baethge, MD, Penny Stanton, CNA, and Heather Melts, RN.
The authors express their gratitude to Drs. Gene V. Ball and William J. Koopman for their critical review of the manuscript and to Ella Henderson for her most expert assistance in its preparation.