Drs. Linnik, Heilbrunn, and Joh have stock ownership or options in La Jolla Pharmaceutical (LJP). Dr. Strand serves as a consultant to LJP. Dr. Hurley has received consulting fees and/or honoraria (more than $10,000) from LJP.
Relationship between anti–double-stranded DNA antibodies and exacerbation of renal disease in patients with systemic lupus erythematosus
Article first published online: 7 APR 2005
Copyright © 2005 by the American College of Rheumatology
Arthritis & Rheumatism
Volume 52, Issue 4, pages 1129–1137, April 2005
How to Cite
Linnik, M. D., Hu, J. Z., Heilbrunn, K. R., Strand, V., Hurley, F. L., Joh, T. and LJP 394 Investigator Consortium (2005), Relationship between anti–double-stranded DNA antibodies and exacerbation of renal disease in patients with systemic lupus erythematosus. Arthritis & Rheumatism, 52: 1129–1137. doi: 10.1002/art.20980
- Issue published online: 7 APR 2005
- Article first published online: 7 APR 2005
- Manuscript Accepted: 11 JAN 2005
- Manuscript Received: 17 SEP 2004
To examine the relationship between changes in anti–double-stranded DNA (anti-dsDNA) antibody levels and the risk of renal flare in patients with systemic lupus erythematosus (SLE), using data from 2 randomized, controlled trials.
Analyses were based on 487 patients with SLE and a history of lupus nephritis who had an anti-dsDNA antibody titer ≥15 IU/ml at baseline, as measured by Farr assay. Results are presented for the combined population of patients, the placebo arms, and the drug treatment arms in which a dsDNA-based bioconjugate (abetimus sodium; LJP 394) was used.
Changes in anti-dsDNA antibody levels were inversely correlated with changes in the C3 level (P < 0.0001 in both trials). Cox proportional hazards regression models showed that changes in anti-dsDNA antibody levels correlated with the risk of renal flare. The models predicted that a point estimate of a 50% reduction in anti-dsDNA antibody levels is associated with a 52% reduction (95% confidence interval [95% CI] 26–68%, nominal P = 0.0007) and a 53% reduction (95% CI 33–69%, nominal P < 0.0001) in the risk of renal flare in the 2 trials, respectively. In the 2 trials, the incidence of renal flare was lower in patients with sustained reductions in anti-dsDNA antibodies (3.0% and 4.1%, respectively) than in patients with stable or increasing antibody levels (21.3% and 20.3%, respectively).
Changes in anti-dsDNA antibody levels were directly correlated with the risk of renal flare and inversely correlated with changes in the C3 level. Reducing anti-dsDNA antibody levels may represent a therapeutic objective in SLE patients with lupus nephritis, because it is associated with a reduced risk of renal flare.
Anti–double-stranded DNA (anti-dsDNA) antibodies are diagnostic for systemic lupus erythematosus (SLE) (1) and have been implicated in the underlying pathogenesis of SLE renal disease and other disease manifestations (2–7). Immune complexes containing anti-dsDNA antibodies are trapped in the glomerulus or are formed in situ on glomerular surfaces, where they can fix complement and precipitate a local inflammatory response (8–13). Anti-dsDNA antibodies are present in higher concentrations in the kidney relative to the systemic circulation, and these antibodies appear to have a higher avidity for dsDNA than do antibodies in the circulation (14–16). Factors that may contribute to the pathogenicity of anti-dsDNA antibodies include IgG isotype and ability to fix complement, charge, and affinity for dsDNA (2, 11, 17, 18).
Increases in circulating anti-dsDNA antibody levels often precede exacerbations of SLE, and prophylactic treatment of patients following rises in anti-dsDNA antibody levels has reduced the occurrence of subsequent disease flares (4, 19–23). Decreasing levels of C3 also correlate with increased disease activity (24–26). Together, these studies demonstrate that increases in anti-dsDNA antibody levels and decreases in C3 levels represent meaningful predictors of a subsequent increase in disease activity. However, some studies have shown only a limited association between anti-dsDNA antibodies and SLE flares (27, 28).
Although increases in anti-dsDNA antibody levels are associated with SLE exacerbations, there are limited data directly addressing the potential clinical benefit associated with reductions in anti-dsDNA antibody levels in patients with SLE. This paucity of information extends to patients with lupus nephritis. Studies that specifically address the relationship between anti-dsDNA antibodies and risk of renal flare are often based on a subset analysis within a broad-spectrum end point of SLE flare.
To address this issue, we conducted a retrospective analysis of data from 2 randomized, well-controlled trials that evaluated time to renal flare in patients with SLE who had a history of renal disease and positive anti-dsDNA antibodies by Farr assay at the time of enrollment. The analysis populations comprise the combined population of patients, the placebo-treatment arms, and the drug-treatment arms in which abetimus sodium (abetimus; previously known as LJP 394) was used. Abetimus, consisting of >97% oligonucleotide dsDNA, was designed to specifically bind to anti-dsDNA antibodies. In clinical trials, both short-term and long-term administration of abetimus have been shown to reduce anti-dsDNA antibody levels in patients with SLE (29–31).
The analyses presented do not address the clinical benefit of abetimus relative to placebo; the 90-05 and 90-09 trials failed to reject the null hypothesis with respect to their prospective primary end point of time to renal flare. Rather, the analyses take advantage of a relatively large, well-characterized patient population to assess the relationship between reductions in anti-dsDNA antibodies and associated risk of renal flare.
PATIENTS AND METHODS
The LJP 394-90-05 and LJP 394-90-09 studies (90-05 and 90-09, respectively) were multicenter, randomized, placebo-controlled trials with the primary objective of determining whether treatment with abetimus prevented or delayed the time to renal flare when compared with placebo. Secondary objectives included an assessment of the effect of abetimus administration on anti-dsDNA antibodies, C3, major SLE flares, and results obtained with concomitant medications, relative to placebo. The mean duration of protocol participation was 371 days in 90-05 and 310 days in 90-09. This report includes data on patients with high-affinity antibodies for the abetimus DNA epitope at baseline, established retrospectively in 90-05 (n = 189 [92 in the abetimus group, 97 in the placebo group]) and prospectively in 90-09 (n = 298 [145 in the abetimus group, 153 in the placebo group]) (29, 32, 33). The study design, treatment regimen, and efficacy results from the 90-05 and 90-09 studies have been previously reported (29, 34). The 90-05 and 90-09 studies were approved by the appropriate institutional review board or ethics committee. All patients provided written informed consent.
Inclusion and exclusion criteria.
The inclusion and exclusion criteria were designed to identify patients with stable renal disease at baseline. Key criteria for study entry included a diagnosis of SLE according to criteria established by the American College of Rheumatology (1), a previous episode of SLE renal disease within 4 years before study entry, and anti-dsDNA ≥15 IU/ml by Farr assay at study entry. Patients were excluded if they 1) had evidence of a renal flare within 3 months of screening or within 4 months prior to dosing; 2) were receiving >20 mg/day of prednisone (or equivalent), >200 mg/day of azathioprine, >25 mg/week of methotrexate, and/or any dosage of cyclophosphamide within 3 months of screening; or 3) had a serum creatinine level >2.5 mg/dl (90-05 study) or >3.5 mg/dl (90-09 study). Patients were permitted standard-of-care treatment while they were enrolled in the trials.
Anti-dsDNA antibody and C3 levels.
Serum samples for anti-dsDNA antibody determinations were collected at least once per month and analyzed by Farr assay at 1 of 2 central laboratories (1 in the US and 1 in Europe). The laboratories used anti-dsDNA assay kits from the same supplier (Diagnostics Products, Los Angeles, CA); assays were validated before implementation, and reproducibility of each new batch of kits was confirmed against the supplier standards and an independent set of standards. All complement assays were also conducted at the 2 central laboratories. Correlations between anti-dsDNA antibody levels and C3 levels were based on data from visits when both were measured (every 4–6 weeks in the 90-05 study and every 4 weeks in the 90-09 study).
A responder analysis was developed to identify patients with sustained reductions in anti-dsDNA antibody titers over the duration of the trial. Responders were defined as having ≥10% reduction from baseline in anti-dsDNA antibody levels for at least two-thirds of all observed values. Anti-dsDNA antibody data that were obtained subsequent to the initiation of high-dose corticosteroid and/or cyclophosphamide therapy, which indicated treatment failure, were imputed to have a value representing less than a 10% reduction from baseline values.
Designation of renal flare.
The criteria for the protocol-defined renal flare end point were identical in the 90-05 and 90-09 studies. Classification as renal flare required that it be attributed to SLE by the treating physician, with concurrence of the medical monitor, and that ≥1 of the following 3 criteria were met: 1) a reproducible increase in 24-hour urine protein levels to (a) >1,000 mg/24 hours if the baseline value was <200 mg/24 hours, (b) >2,000 mg/24 hours if the baseline value was 200–1,000 mg/24 hours, or (c) >2-fold the baseline value if the baseline value was >1,000 mg/24 hours; 2) a reproducible increase in serum creatinine levels of >20% or at least 0.3 mg/dl, whichever was greater, accompanied by proteinuria (>1,000 mg/24 hours), hematuria (≥4 red blood cells [RBCs]/high-power field [hpf]) and/or RBC casts; or 3) new, reproducible hematuria (≥11–20 RBCs/hpf) or a reproducible increase in hematuria by 2 grades compared with baseline, associated with >25% dysmorphic RBCs, glomerular in origin, exclusive of menses, accompanied by either an 800-mg increase in 24-hour protein levels or new RBC casts. All laboratory values were determined at a central laboratory, and all renal flares in the 90-09 trial were adjudicated by an independent committee. Renal biopsies were not required during the trials.
Statistical methodology for continuous and categorical variables is presented in the Results section. Comparisons of time-to-event variables were performed using the log rank test. All treatment group comparisons were conducted at the 2-sided significance level of 0.05; P values for retrospective analyses are identified as nominal.
The relationship between change in anti-dsDNA antibody levels and risk of renal flare was analyzed using Cox proportional hazards regression models (35). The models included change from baseline in log10 anti-dsDNA antibody levels as a time-dependent covariate and baseline covariates, including age >40 years, race (white or black), sex, treatment group, baseline C3 level ≥75 mg/dl, and baseline anti-dsDNA titer ≥75 IU/ml. Log-transformed data were used to account for the asymmetric distribution of the anti-dsDNA data.
The demographics and baseline disease characteristics of the patient populations in studies 90-05 and 90-09 are summarized in Table 1. Treatment groups were well balanced at baseline, both within and between trials.
|Characteristic||LJP 394-90-05||LJP 394-90-09|
|Abetimus (n = 92)||Placebo (n = 97)||Abetimus (n = 145)||Placebo (n = 153)|
|Mean ± SD||37.2 ± 10.8||34.6 ± 10.0||37.1 ± 11.1||35.2 ± 9.9|
|Male||9 (9.8)||7 (7.2)||18 (12.4)||21 (13.7)|
|Female||83 (90.2)||90 (92.8)||127 (87.6)||132 (86.3)|
|White||33 (35.9)||44 (45.4)||69 (47.6)||68 (44.4)|
|Black||34 (37.0)||27 (27.8)||33 (22.8)||43 (28.1)|
|Other||25 (27.2)||26 (26.8)||43 (29.7)||42 (27.5)|
|Mean ± SD||70.5 ± 18.1||73.6 ± 19.6||74.1 ± 21.0||73.4 ± 18.5|
|Disease duration, years|
|Mean ± SD||8.97 ± 6.78||8.25 ± 6.12||10.0 ± 7.69||9.1 ± 6.91|
|Mean ± SD||94.6 ± 158.9||107.1 ± 183.7||86.5 ± 137.1||75.8 ± 69.4|
|Mean ± SD||78.4 ± 22.3||80.7 ± 21.6||87.0 ± 25.4||83.5 ± 28.24|
|Below the lower limit of normal||45 (48.9)||41 (42.3)||46 (31.7)||62 (40.5)|
|Serum creatinine, mg/dl|
|Mean ± SD||1.1 ± 0.36||1.1 ± 0.37||1.1 ± 0.40||1.1 ± 0.36|
|≥1.5||11 (12.0)||11 (11.3)||20 (13.8)||23 (15.0)|
|Urine protein, mg/day|
|Mean ± SD||851 ± 1,353||1,059 ± 1,652||963 ± 1,329||1,131 ± 1,939|
|≥500||32 (34.8)||37 (38.1)||60 (41.4)||65 (42.5)|
|Corticosteroids and immunosuppressive agents|
|Prednisone||74 (80)||77 (79)||121 (83)||127 (83)|
|Azathioprine||26 (28)||33 (34)||38 (26)||36 (24)|
|Mycophenolate mofetil||0 (0)||0 (0)||20 (14)||22 (14)|
Relationship between anti-dsDNA antibodies and C3.
An inverse correlation was observed between change in anti-dsDNA antibody levels and change in C3 levels. The results in Table 2 demonstrate that changes in anti-dsDNA antibody titers were inversely correlated with changes in C3 levels in the combined patient populations and the individual abetimus and placebo treatment groups in the 90-05 and 90-09 studies.
|Association||LJP 394-90-05||LJP 394-90-09|
|Nominal P||Association||Nominal P||Association|
|Anti-dsDNA and C3†|
|Baseline covariates and changes in anti-dsDNA antibodies with time to renal flare‡|
|Log10 change in anti-dsDNA from baseline||0.0007||11.1 (2.8–44.8)||<0.0001||12.5 (3.8–41.1)|
|Baseline anti-dsDNA (≥75 IU/ml or not)||0.0058||3.4 (1.4–7.9)||NS||–|
|Baseline C3 (≥75 mg/dl or not)||NS||–||0.0001||0.27 (0.14–0.53)|
|Black race or not||0.0240||4.1 (1.2–13.9)||NS||–|
The cumulative percentage of patients with renal flare according to treatment assignment is presented in Figure 1. Renal flares were observed in 28 of 189 patients (14.8%) and 41 of 298 patients (13.8%) in the 90-05 high-affinity and 90-09 intent-to-treat populations, respectively. In 90-05, renal flares were observed in 7 of 92 patients (7.6%) and 21 of 97 patients (21.6%) in the abetimus and placebo groups, respectively. In 90-09, renal flares were observed in 17 of 145 patients (11.7%) and 24 of 153 patients (15.7%) in the abetimus and placebo groups, respectively.
Cox proportional hazards regression analysis.
Cox proportional hazards regression models were developed to estimate the association between risk of renal flare, selected baseline covariates, and change from baseline in log10 anti-dsDNA antibody levels as a time-dependent covariate. Three baseline covariates, anti-dsDNA antibody titer ≥75 IU/ml, C3 level ≥75 mg/dl, and black race, exhibited a significant association with time to renal flare in 1 of 2 trials, as shown in Table 2.
The models showed a statistically significant association between change from baseline in anti-dsDNA antibody levels and risk of renal flare in the combined treatment groups from each trial, as shown in Table 2. The magnitude of this association was consistent in the 90-05 and 90-09 studies, with estimated hazards ratios of 11.1 (95% confidence interval [95% CI] 2.8–44.8) and 12.5 (95% CI 3.8–41.1) for a 1-log change in anti-dsDNA antibody levels, respectively. The consistency of the estimates and the associated standard errors from the 2 studies are reflected in the similar 95% CIs for the 2 estimates.
The relative risk of renal flare associated with a change in anti-dsDNA antibody levels from baseline was modeled using the calculated hazards ratio, as shown in Figure 2. These models predicted that increases in anti-dsDNA antibody levels are associated with an increased risk of renal flare, and that decreases in anti-dsDNA antibody levels are associated with a decreased risk of renal flare. For example, the Cox model predicted that a point estimate of a 50% reduction in anti-dsDNA antibody levels is associated with a 52% reduction (95% CI 26–68%, nominal P = 0.0007) and a 53% reduction (95% CI 33–69%, nominal P < 0.0001) in the risk of renal flare as compared with patients with no change in anti-dsDNA antibody levels in the 90-05 and 90-09 studies, respectively.
Analysis of individual treatment groups revealed an association between change in anti-dsDNA antibody levels and risk of renal flare in the placebo population, with estimated hazards ratios of 30.4 (nominal P = 0.0009) and 48.9 (nominal P < 0.0001) for a 1-log change in anti-dsDNA antibodies in the 90-05 and 90-09 trials, respectively. Analysis of the abetimus treatment groups did not reveal significant associations, with estimated hazards ratios of 6.4 and 2.5 for the 90-05 and 90-09 trials, respectively, consistent with the demonstration of a drug-treatment effect on the variables used in the model.
Responder analysis among patients with sustained reductions in anti-dsDNA antibody levels.
A responder analysis was developed to identify patients who had a sustained reduction in anti-dsDNA antibody titers below the baseline level. The overall frequency of responders was 35.5% (67 of 189) in the 90-05 trial and 40.6% (121 of 298) in the 90-09 trial, as shown in Figure 3A. The frequency of responders in the abetimus and placebo treatment groups was 58.7% and 13.4%, respectively, in the 90-05 study, and 55.2% and 26.8%, respectively, in the 90-09 study, as shown in Figure 3C. The median percent reduction in anti-dsDNA antibody levels from baseline in responders and nonresponders is shown in Figure 4. The median percent reduction in the responder populations was ∼40–50% below baseline; this reduction was substantially greater than the minimum requirement of a 10% reduction from baseline.
The responder population was subsequently evaluated to determine whether patients with sustained reductions in anti-dsDNA antibody levels exhibited clinical benefit in terms of a reduction in the incidence of renal flare. The frequency of renal flare in the combined population was lower in treatment responders (3.0% and 4.1%) than in nonresponders (21.3% and 20.3%) for the 90-05 and 90-09 studies, respectively (nominal P = 0.0004 and P < 0.0001 by Fisher's exact test for 90-05 and 90-09, respectively). Two of 28 renal flares (7%) were observed in the responder group in the 90-05 trial, and 5 of 41 renal flares (12%) were observed in this group in the 90-09 trial. The incidence of renal flare in the combined treatment groups is shown in Figure 3B, and the incidence in the individual treatment groups is shown in Figure 3D.
The current report provides an analysis of the relationship between changes in anti-dsDNA antibody levels and exacerbation of renal disease in patients with SLE who had a positive anti-dsDNA antibody titer at baseline by Farr assay and a history of renal disease. Three key findings emerge from this analysis. First, the observed inverse relationship between anti-dsDNA and C3 levels in both trials provides support for the pathogenic nature of anti-dsDNA antibodies. Second, increases in anti-dsDNA antibody levels were associated with an increased risk of renal flare in this population of patients with a previous history of renal disease. Third, reductions in anti-dsDNA antibody levels were associated with a reduced risk of renal flare.
The 90-05 and 90-09 studies provide an opportunity to investigate the relationship between anti-dsDNA antibodies and renal exacerbations, as well as the natural history of the disease, because the 2 studies represent a relatively large number of patients with identical definitions of renal flare as a primary end point and similar inclusion and exclusion criteria. In addition, the availability of data from the 2 trials provides an opportunity to evaluate the reproducibility of findings. Together, these studies provide a substantial data set for evaluating the potential therapeutic benefit of a reduction in anti-dsDNA antibodies.
The 90-05 and 90-09 trials also address concerns that have been raised about previous studies in this patient population, including the frequency of sampling of anti-dsDNA antibodies, the type of assay used to measure anti-dsDNA, and end points that are not restricted to renal exacerbations. For example, frequent sampling for anti-dsDNA antibodies appears critical for accurately determining a correlation between antibody levels and disease activity. In studies that have shown an association between rising antibody levels and disease activity, antibody levels were sampled frequently (weekly to monthly) (4, 6, 21). Studies in which antibodies were sampled infrequently, at intervals of every 3 months or longer, did not demonstrate this association (27, 28). In the 90-05 and 90-09 studies, anti-dsDNA was measured at least once per month by Farr assay.
The observed association between anti-dsDNA antibodies and renal flares demonstrated using the Cox proportional hazards regression models supports previous observations that increases in anti-dsDNA antibodies often precede exacerbations of disease (4, 6, 21). This is consistent with observations that therapeutic interventions based on increases in anti-dsDNA antibodies prior to clinical manifestations may prevent exacerbations or flares of SLE disease (19, 20). The use of a responder analysis to identify patients with sustained reductions in anti-dsDNA antibody levels provided an independent confirmation of the Cox regression analysis, because there was a clear and reproducible reduction in the incidence of renal flare in patients who exhibited sustained reductions in anti-dsDNA antibodies.
The Cox proportional hazards regression models identified 3 baseline variables that were associated with risk of renal flare, although none of the variables was identified in both trials. All of the significant baseline variables, elevated anti-dsDNA antibody level, low C3 level, and black race, have established relationships with SLE nephritis (6, 7, 14, 36–38). Although the trial was not powered to establish significance for these baseline variables, we interpret these observations in the context of the literature to indicate that SLE patients with hypocomplementemia and elevated anti-dsDNA antibody titers are predisposed to renal relapse.
The pathogenic nature of anti-dsDNA antibodies in patients with lupus nephritis is supported by the observation that changes in complement levels mirror changes in anti-dsDNA levels. The association between anti-dsDNA and C3 has clinical relevance, because circulating complement levels are considered an important indicator of disease activity in lupus nephritis (26, 39).
The minimum requirement for classification as a responder was at least a 10% reduction in the anti-dsDNA antibodies for at least two-thirds of all observed values. This requirement resulted in a responder population in which the median percent reduction in anti-dsDNA antibodies was ∼40–50% below baseline. This represents a substantial reduction in antibody levels for most of the responder patients; the mean baseline level of anti-dsDNA antibodies ranged from 76 IU/ml to 107 IU/ml. The absolute magnitude of reduction in anti-dsDNA antibodies in responder patients would be greater in patients with high initial Farr values and lower in patients whose antibody levels were close to the minimal inclusion requirement value of 15 IU/ml.
The characterization of patients with sustained reductions used all available anti-dsDNA antibody values for each patient; the clinical benefit in the responder population was then examined. Similar results were obtained from sensitivity analyses that varied the minimum percent reduction in anti-dsDNA antibody levels from 10% to 30% and varied the required percentage of qualifying values from 50% to 80%. An additional sensitivity analysis that censored anti-dsDNA antibody levels after a renal flare demonstrated little effect on the distribution of patients with sustained reductions in the 2 trials. In addition, the distribution of renal flares in the combined population and individual treatment groups continued to show clinical benefit in the responder population. In the placebo treatment group, the frequency of renal flare in responders and nonresponders was virtually identical to that observed in the primary analysis. In the abetimus treatment group, the frequency of renal flare was 5.4% and 9.4% in responders and 11.1% and 15% in nonresponders in the 90-05 and 90-09 trials, respectively.
The current treatments for lupus nephritis may decrease anti-dsDNA antibody levels in addition to suppressing other aspects of the immune system. These agents are used in spite of known toxicities, because it has been documented that the ability to induce remission of lupus nephritis has a direct relationship to renal survival (40–44).
Although the current results demonstrate that reductions in anti-dsDNA antibody levels are associated with a reduced risk of renal flare, a prospectively defined, statistically significant demonstration of a reduced risk of renal flare in patients treated with abetimus has yet to be established. Treatment with abetimus has been shown to be an effective approach for reducing anti-dsDNA antibody levels (29–31, 45). Abetimus is >97% dsDNA and is expected to interact predominantly with target molecules capable of recognizing dsDNA. With the exception of this effect on anti-dsDNA antibodies, there is no other known direct biologic effect of treatment with LJP 394 at the doses used in these trials. This compares with the placebo arm in which changes in anti-dsDNA antibody levels likely reflect a broader change in the autoimmune status of the individual, induced either pharmacologically or by the normal waxing and waning of the SLE disease process.
In conclusion, analyses of data from 2 large randomized, controlled trials demonstrate an association between levels of circulating anti-dsDNA antibodies and the risk of renal flare in patients with SLE and lupus nephritis. The analyses specifically establish that reductions in anti-dsDNA antibodies are associated with a reduced risk of renal flare and an increase in C3 levels in patients with SLE who have a history of renal disease. These data suggest that patients with SLE who have reductions in anti-dsDNA antibody levels are more likely to have a favorable clinical prognosis than are patients with stable or increasing anti-dsDNA antibody levels.
We would like to acknowledge the constructive comments provided by Drs. K. Frank Austen and Gary Koch.
- 18Clinical application of serologic abnormalities in systemic lupus erythematosus. In: WallaceDJ, HahnBH, editors. Dubois' lupus erythematosus. 5th ed. Baltimore: Williams & Wilkins; 1997. p. 925–42.
- 29LJP 394 Investigator Consortium. LJP 394 for the prevention of renal flare in patients with systemic lupus erythematosus: results from a randomized, double-blind, placebo-controlled study. Arthritis Rheum 2003; 48: 442–54., , , , , , et al, and the
- 34Clinical efficacy results from a RCT of LJP 394 in SLE patients with history of renal disease [abstract]. Arthritis Rheum 2003; 48 Suppl 9: S582., , , , , , et al.
- 35Analysis of survival data. New York: Chapman and Hall; 1984., .