Objective To determine trends in survival among adult and paediatric patients with systemic lupus erythematosus (SLE) from 1950 to the present.
Methods We performed a systematic literature review to identify all published cohort studies on survival in patients with SLE. We used Bayesian methods to derive pooled survival estimates separately for adult and paediatric patients, as well as for studies from high-income countries and low/middle-income countries. We pooled contemporaneous studies to obtain trends in survival over time. We also examined trends in major causes of death.
Results We identified 125 studies of adult patients and 51 studies of paediatric patients. Among adults, survival improved gradually from the 1950s to the mid-1990s in both high-income and low/middle-income countries, after which survival plateaued. In 2008–2016, the 5-year, 10-year and 15-year pooled survival estimates in adults from high-income countries were 0.95, 0.89 and 0.82, and in low/middle-income countries were 0.92, 0.85 and 0.79, respectively. Among children, in 2008–2016, the 5-year and 10-year pooled survival estimates from high-income countries were 0.99 and 0.97, while in low/middle-income countries were 0.85 and 0.79, respectively. The proportion of deaths due to SLE decreased over time in studies of adults and among children from high-income countries.
Conclusions After a period of major improvement, survival in SLE has plateaued since the mid-1990s. In high-income countries, 5-year survival exceeds 0.95 in both adults and children. In low/middle-income countries, 5-year and 10-year survival was lower among children than adults.
- systemic lupus erythematosus
Statistics from Altmetric.com
If you wish to reuse any or all of this article please use the link below which will take you to the Copyright Clearance Center’s RightsLink service. You will be able to get a quick price and instant permission to reuse the content in many different ways.
Survival of adults with systemic lupus erythematosus (SLE) is widely recognised to have improved between the 1950s and 1990s, with 5-year survival increasing from 50% to 60% to more than 95%.1–5 However, it is not clear if survival has continued to improve, as only one study that reported trends in survival included data after 2000.3 Some evidence suggests that the improvement in survival may have slowed between 1980 and 1990.6 Studies that have reported improvement in survival have largely been from high-income countries.2 3 7 8 It is unclear if similar improvements have occurred among patients in low/middle-income countries (LMIC), where socioeconomic barriers may limit access to care. Few data are available on 10-year and 15-year survival trends.
Similarly, 5-year survival in paediatric SLE has improved from 60% to 70% in the 1950s to more than 90% in the 1980s.9–12 These estimates are based on isolated studies, and information on recent trends are lacking. Paediatric patients experience a longer disease course and extended exposure to disease and medication complications.13 However, few studies have provided survival estimates beyond 5 years.14–16
We evaluated changes in survival from the 1950s to 2016 in adults and children with SLE in both high-income countries and LMIC based on a systematic literature review. We hypothesised that survival in SLE steadily improved over time, with greater improvement in studies from high-income countries. We also examined if the principal causes of death in patients with SLE have changed over time.
Data sources and search strategy
We conducted a systematic review of the published literature on survival in adult and paediatric patients with SLE (omitting neonatal lupus). The study protocol was developed based on Preferred Reporting Items for Systematic Review and Meta-Analysis guidelines17 and Meta-analysis of Observational Studies in Epidemiology recommendations.18 We searched the PubMed, Embase and Scopus databases from their inceptions to 7 June 2016, without language restrictions. The search strategy and terms were developed in collaboration with a medical informationist (online supplementary appendix A). We also reviewed the references of these studies and review articles for additional publications. The study was exempted from human subjects review by the National Institutes of Health Office of Human Subjects Research Protection.
Supplementary file 1
Two investigators independently reviewed the titles and abstracts, and when necessary, full texts, to determine eligibility for inclusion. We included prospective or retrospective cohort studies of overall survival in adult or paediatric patients with SLE. Studies were considered as paediatric if the inclusion criteria specified an age of 0–17 years. We excluded: (1) animal studies; (2) case reports, case series, reviews, meta-analyses and abstracts; (3) studies on unrelated topics; (4) studies of selected SLE subsets (patients with specific clinical manifestations, hospital inpatients, elderly onset patients, adults with childhood-onset SLE and male only cohorts); (5) studies based on administrative data; (6) studies with incomplete data on number of deaths or follow-up; (7) adult studies with fewer than 20 patients and paediatric studies with fewer than 10 patients; and (8) studies of the same cohort as included articles.
Two investigators read the full text of adult (MGT and MMW) and paediatric studies (LL and MMW) and independently performed data extraction and quality assessment. Discrepancies were resolved by consensus. We collected data on the year and country of publication, study design, inception or prevalence cohort, sample size, years of patient enrolment, patient demographic characteristics, SLE duration at study entry, proportion with nephritis or central nervous system (CNS) involvement during disease course, follow-up duration and number of deaths. We extracted data on 5-year, 10-year, and 15-year survival when provided and separately identified studies with Kaplan-Meier plots. We extracted data on causes of death and classified these as due to either SLE, infection, cardiovascular disease, malignancy or other causes.
We evaluated study quality using items adapted from the Newcastle-Ottawa scale (online supplementary appendix B), as used previously.19 20 We considered studies of inception cohorts as high quality.
Overall survival was the outcome of interest. Survival data were reported as either Kaplan-Meier plots, as summary survival estimates from time-to-event analyses (but without a Kaplan-Meier plot), or as per cent mortality over the observation period. We treated each format differently to arrive at pooled survival estimates.19 For studies that reported Kaplan-Meier plots, we reconstructed individual patient data from the plots.21 For studies that reported summary survival estimates, we used these to generate estimates of individual patient data. For studies that reported percent mortality, we used these percentages.
For each study, we modelled the time to death as a Weibull distribution. Each study contributed one survival estimate. We used Bayesian estimation with Markov chain Monte Carlo methods to obtain posterior distributions of the pooled Weibull estimates, which we used as the basis of 5-year, 10-year and 15-year survival estimates and corresponding 95% credible intervals. R V.3–12 package rjags was used for analysis.22 For prevalence cohorts, we accounted for left truncation by adding the median duration of SLE at entry to the estimation of time to death. A full description of the methods is provided in online supplementary appendix C.
To examine trends over time, we divided calendar years from 1950 to 2016 into overlapping 5-year periods. We pooled studies that contributed data in a given five-calendar-year interval, starting from the calendar year of the midpoint of enrolment and ending with the end of follow-up. We sequentially repeated the analysis using the subset of studies represented in each 5-year calendar window. Different studies therefore entered and dropped out of the analysis across calendar years, akin to a moving average. We weighted studies by their sample size so that larger studies had greater influence on the pooled estimates.
We accounted for two main sources of heterogeneity, age and development status, by stratification. Within the adult and paediatric studies, we performed separate analyses for studies from high-income countries and LMIC, using World Bank criteria for the midpoint year of patient enrolment in each study.23
We performed a sensitivity analysis in which the contribution of individual studies was limited to 10 years from the midpoint of enrolment. This analysis limits the late influence of studies with long-term follow-up, when fewer patients might be under observation. This analysis may therefore be more sensitive to changes in survival over calendar years. Additionally, we performed a separate analysis of inception cohorts (ie, high-quality studies).
We examined trends in the prevalence of nephritis and CNS involvement over time using Spearman correlations, weighted by sample size. We also examined trends in causes of death across decades, based on the year of start of enrolment. We tested trends across decades in the proportion of deaths due to SLE using weighted linear regression (beta coefficient b for the indicator variable for decade). The analysis was weighted by number of deaths, so that larger studies had greater influence. Because studies of prevalence cohorts may miss deaths early in SLE, which may have different causes than late deaths, we repeated this analysis using only inception cohorts. We considered p values ≤0.05 as statistically significant. We used SAS program (V.9.3) for these analyses.
We included 171 studies: 125 adult and 51 paediatric (five studies included stratified data on both adults and children) (figure 1 and online supplementary appendix D). Sixteen per cent were prospective cohort studies and 84% were retrospective cohort studies.
Although our goal was to study adult and paediatric patients separately, age of inclusion was not clearly specified in all studies. Thirty-four of the 125 adult studies (27%) included only adult patients (age 18 years or older) or reported data separately for adults, while 32% did not report if they solely examined adults or also included children. We assumed most patients were adults based on the departments from which the studies originated. Forty-one per cent of the 125 studies reported that they also included children (range 1.3%–26%, median 9%) but did not provide age-stratified results. For ease of description, we termed these age-unrestricted studies as ‘adult’ studies, but it is important to recognise that some included paediatric patients. The adult studies comprised 46 317 patients and paediatric studies comprised 6862 patients.
The adult studies included from 21 to 3679 patients (90.2% women, mean age (at diagnosis or study entry, as reported in the primary studies) 33.8 years, mean follow-up 6.8 years). Eighty per cent of studies used American College of Rheumatology (ACR) classification criteria for the enrolment of patients, 25% examined inception cohorts, 22% examined community-based cohorts and 32% provided Kaplan-Meier plots. Only 39% of studies reported the proportion of patients lost to follow-up, which was less than 20% in 88% of these studies. The proportion of patients with nephritis decreased substantially over time in adult studies from both high-income countries and LMIC, as did the proportion of patients with CNS involvement in high-income countries, indicating a change in the nature of patients enrolled over time (figure 2).
Paediatric studies included between 13 and 1393 patients (82.3% girls, mean age at diagnosis 12.4 years, mean follow-up 5.3 years (range 1.7–13.1 years). Ninety-two per cent of studies used ACR classification criteria for the enrolment, 16% examined inception cohorts, 12% examined community-based cohorts and 51% provided Kaplan-Meier plots. Only 31% of paediatric studies reported on losses to follow-up. The proportion of patients with nephritis decreased over time in studies from high-income countries, as did the proportion with CNS involvement in studies from LMIC (figure 2).
Survival in adult studies
Eighty-two studies were from high-income countries and 43 were from LMIC (online supplementary figure 1). Among studies from high-income countries, there was a progressive increase in survival from the mid-1950s to 1990, after which survival estimates were stable (figure 3). In 2008–2016, the 5-year, 10-year and 15-year survival estimates in high-income countries were 0.95 (95% credible interval 0.94 to 0.96), 0.89 (0.88 to 0.90) and 0.82 (0.81 to 0.83), respectively.
Supplementary file 2
Data from LMIC did not extend prior to 1970, but subsequent trends in survival were similar to those in high-income countries (figure 3 and online supplementary figure 2). In 2008–2016, the 5-year, 10-year and 15-year survival estimates in LMIC were 0.92 (0.91 to 0.93), 0.85 (0.84 to 0.87) and 0.79 (0.78 to 0.81), respectively.
Results were very similar in the sensitivity analysis that truncated follow-up at 10 years after the midpoint of enrolment, indicating that studies with very long follow-up did not overly influence the findings (online supplementary figure 3). Survival estimates were also very similar in the 25 inception cohort studies from high-income countries, but slightly lower in the seven inception cohort studies from LMIC (online supplementary figure 4). Only one inception cohort study, which had 5-year survival of 0.80, contributed to the LMIC estimate after 2008, accounting for the recent decrease in survival. Differences in survival estimates between high-income countries and LMIC were somewhat greater in inception cohorts, with 5-year survival of 0.94 and 0.89, 10-year survival of 0.88 and 0.81 and 15-year survival of 0.83 and 0.74, respectively, in 2008–2016.
Survival in paediatric studies
Thirty-three studies were from high-income countries and 18 were from LMIC (online supplementary figure 5). Only three studies reported 15-year survival, and therefore we did not estimate survival for this time point. Among studies from high-income countries, there was a sharp increase in survival from the 1960s to the 1970s, followed by slower improvement (figure 4). In 2008–2016, the 5-year and 10-year survival estimates from high-income countries were 0.99 (0.98 to 1.00) and 0.97 (0.96 to 0.98), respectively.
Data from LMIC did not extend prior to the 1970s. The increase in survival was pronounced between 1970 and 1990, followed by a plateau if not a slight decrease (figure 4). Between 1980 and 2000, survival persistently lagged that of high-income countries. By the end of the study period, 5-year and 10-year survival estimates from LMIC were 0.85 (0.83 to 0.88) and 0.79 (0.76 to 0.82), respectively, substantially lower than those from high-income countries.
Results were similar in the sensitivity analysis which limited the influence of studies with very long follow-up (online supplementary figure 6). There were too few paediatric inception cohort studies (n=8) for separate analysis.
Causes of death
Causes of death were reported in 87 adult studies, 22 of which examined inception cohorts (table 1). Among studies from high-income countries, the proportion of deaths attributed to SLE decreased over time in all studies and in inception cohorts (both p for trend=0.01). Similarly, the proportion of deaths due to SLE was lower in more recent years among inception cohorts from LMIC, but there was no trend among all studies from LMIC. Deaths from infections increased over time in adult inception cohorts from both high-income countries and LMIC.
Causes of death were reported in 39 paediatric studies, five of which reported on inception cohorts. There was no significant trend in cause of death in paediatric studies, possibly due to the smaller number of studies. In studies from LMIC, the frequency of deaths due to SLE demonstrated a trend to increase over time. In inception cohorts from high-income countries, SLE was the cause of more than 50% of deaths in recent studies.
Our results showed that survival in patients with SLE gradually increased from the 1950s to the mid-1990s, and then plateaued. Although it is widely recognised that survival in SLE has improved substantially over the past decades, the time course of this improvement has not been clear. Seven adult studies have reported survival trends over different calendar periods in the same cohort. In six studies, follow-up started between 1950 and 1970 and ended in the 1980s or 1990s.2 3 7 8 24–26 Only one study had follow-up into the 2000s.3 Trends in survival have also been examined in studies that compared the relative risk of death in SLE to the general population.3 27 28 For example, Urowitz et al found that the standardised mortality ratio decreased from 12.6 in 1970–1978 to 3.46 in 1997–2005.3 Our results indicate that survival in adults with SLE has not continued to improve through the 2000s.
The increase in survival to the mid-1990s is likely multifactorial. Some of the improvement is undoubtedly attributable to better treatment of SLE and advances in general medical care. However, some of the improvement may also be due to inclusion of milder cases after more widespread use of antinuclear antibody testing.29 30 Milder cases of SLE might have been underdiagnosed in earlier decades, leaving only more severely affected patients included in survival studies. Supporting this is our finding that lower proportions of adult patients had nephritis or CNS disease in more recent studies. Part of the apparent improvement in survival may therefore reflect less spectrum bias in more recent decades. Shorter delay in diagnosis may have also contributed to longer apparent survival in more recent studies (eg, lead-time bias).
Our pooled estimates are similar to the limited data from studies in the 2000s.31–37 The plateau in survival since the mid-1990s may reflect ongoing limitations in the appropriate implementation of treatment, or in the control of comorbidities or complications such as infections. It may also represent persistent poor outcomes among patients with treatment-resistant SLE. Whether newer medications such as mycophenolate mofetil and rituximab affect survival at the population level is not clear.
Among adult studies, survival was comparable in LMIC and high-income countries. However, disparity was somewhat more apparent in inception cohort studies, particularly for 15-year survival. Differences in survival between high-income countries and LMIC were more striking in paediatric SLE studies, with a gap in 10-year survival of 0.97 and 0.79, respectively. Even more concerning is that these rates seem to plateau at this level. Barriers to healthcare access and limited availability of experienced clinicians and treatments may influence the diagnosis and management of SLE in LMIC.38 39 Some severely affected patients may die before reaching specialists able to make a diagnosis, and therefore may not be included in survival estimates.
SLE was less frequently a cause of death in recent years among adult patients, supporting previous literature.4 8 40 The introduction of effective immunosuppressive treatments may have reduced deaths directly due to SLE, but likely also increased complications, notably infection-related deaths.41 The inclusion of more patients with milder SLE may have also contributed to fewer SLE-related deaths recently. In the 2000s, infections and cardiovascular disease were the main causes of death in large inception cohorts.42–44 Our pooled results indicated no increase in the proportion of deaths due to cardiovascular disease or cancer over time among adults.
The leading cause of death in paediatric patients in LMIC continues to be SLE. High rates of lupus nephritis combined with barriers to treatment may contribute to this finding. Infections were a major cause of death among paediatric patients in high-income countries. Serious infections are common among paediatric patients.45 Interventions to increase vaccination in children could further improve survival.46
Our study has several strengths. We examined both adult and paediatric studies, and separately analysed studies by income level, because pooling these may misrepresent the survival experience in each group. We also separately examined inception cohort studies.
The most important limitation is that few studies examined inception cohorts. Studies of prevalence cohorts are more likely to underestimate mortality, because they will not capture deaths early in the course of SLE. Few studies were community based, which would provide a more representative view of survival than studies from referral centres. Also, only 39% of studies reported a time-to-event curve, despite having survival as an outcome, and only 38% reported the proportion lost to follow-up. Few paediatric studies had more than 10 years of follow-up. Although we summarised data of many studies, we accounted for two main sources of variation—age and development status—by stratification. There were insufficient data to stratify further by gender or race. Some adult studies included small or unknown proportions of children, which might have affected the survival estimates.
Our results indicate that overall survival has not increased over the past 20 years in patients with SLE in high-income countries. Progress in improving survival will depend on a comprehensive understanding of the major preventable causes of death. This may be achieved most quickly by a focus on preventing infections and improving the outcomes of patients with serious infections.41
This work used the computational resources of the NIH HPC Biowulf cluster (http://hpc.nih.gov).
Handling editor Tore K Kvien
Contributors MGT and MMW conceived the study. All authors designed the study, and FH, AD and MMW did the analysis. MGT drafted the manuscript, and all authors provided critical review and approval of the final version.
Funding This work was supported by the Intramural Research Program, National Institute of Arthritis and Musculoskeletal and Skin Diseases, National Institutes of Health (ZIAAR041153).
Competing interests None declared.
Ethics approval National Institutes of Health Office of Human Subjects Research Protection.
Provenance and peer review Not commissioned; externally peer reviewed.
Data sharing statement Data on which the study is based are publicly available.
Correction notice This article has been corrected since it published Online First. The third author’s name has been corrected to Jinxiang Hu.