PATIENTS AND METHODS
The study was approved by the ethics review boards of all participating institutions.
Patients with definite SLE according to American College of Rheumatology (ACR) criteria (18, 19) or clinical criteria were eligible for inclusion, provided they were older than age 16 years. The study encompassed 23 collaborating lupus centers in 6 countries (Canada, the US, the UK, Iceland, Sweden, and Korea) (Table 2). Patients included in the study had been followed up in outpatient clinics and/or in the hospital as inpatients. Although most of the investigators were based at tertiary academic centers, enrollment of patients by community physicians was actively encouraged.
Table 2. Participating centers: international cohort study of malignancy in SLE
|North America|| || || |
| Calgary, Alberta, Canada||522||Patients enrolled from regional physician network†||ACR SLE criteria‡|
| Halifax, Nova Scotia, Canada||109||Consecutive patients enrolled at first clinic visit†||ACR SLE criteria|
| London, Ontario, Canada||90||Unselected patients followed up from first clinic visit§||ACR SLE criteria‡|
| Montreal, Quebec, Canada|| || || |
| Hopital Maisonneuve-Rosemont||120||Assembled using hospital discharge and clinic records§||ACR SLE criteria‡|
| Montreal General Hospital||309||Consecutive patients enrolled at first clinic visit†||ACR SLE criteria|
| Notre-Dame Hospital||120||Unselected patients followed up from first clinic visit¶||ACR SLE criteria|
| Saskatoon, Saskatchewan, Canada||306||Consecutive patients enrolled at first clinic visit†||ACR SLE criteria‡|
| Toronto, Ontario, Canada||873||Consecutive patients enrolled at first clinic visit†||ACR SLE criteria‡|
| Vancouver, British Columbia, Canada||81||Unselected patients followed up from first clinic visit§||ACR SLE criteria‡|
| Winnipeg, Manitoba, Canada||158||Unselected patients followed up from first clinic visit§||ACR SLE criteria‡|
| Baltimore, MD||453||Unselected patients followed up from first clinic visit¶||ACR SLE criteria‡|
| Birmingham, AL||166||Inception cohort (subset consenting to cancer linkage)†||ACR SLE criteria|
| Chapel Hill, NC||223||Unselected patients followed up from first clinic visit¶||ACR SLE criteria‡|
| Chicago, IL||469||Unselected patients followed up from first clinic visit¶||ACR SLE criteria|
| New York, NY|| || || |
| Albert Einstein University||240||Unselected patients followed up from first clinic visit§||ACR SLE criteria‡|
| Downstate University–Brooklyn||957||Consecutive patients enrolled at first clinic visit†||ACR SLE criteria‡|
| Pittsburgh, PA||1,050||University of Pittsburgh/regional rheumatologists¶||ACR SLE criteria‡|
|United Kingdom|| || || |
| Birmingham, England||439||Unselected patients followed up from first clinic visit†||ACR SLE criteria|
| London, England||273||Unselected patients followed up from first clinic visit†||ACR SLE criteria‡|
| Lanarkshire, Scotland||1,937||Assembled using hospital discharge registry§||SLE discharge diagnosis#|
|Other centers|| || || |
| Lund, Sweden||114||Inception cohort, enrollment at SLE diagnosis†||ACR SLE criteria|
| Reykjavik, Iceland||221||Unselected patients enrolled in national registry¶||ACR SLE criteria|
| Seoul, Korea||317||Unselected patients followed up from first clinic visit¶||ACR SLE criteria‡|
|Total||9,547|| || |
Data were collected on patient birth date, sex, dates of lupus diagnosis and cohort entry, and date of death, if applicable. Observed cancers were determined by linkage of the study patients' information to regional cancer registries (see Acknowledgments). Vital statistics linkages were performed for patients who were lost to followup or deceased, with the National Death Index in the US cohorts, and, in the non-US cohorts, with regional vital statistics registries. At 3 centers, ethics committee approval did not permit linkage of lost-to-followup patients to vital status registries. A total of only 515 study subjects were from these centers, very few of whom were lost to followup, but to be conservative, we assumed that any lost-to-followup patients from these centers remained at risk until the end of the observation interval.
For each cancer type, we determined the SIR. In secondary analyses, SIRs were estimated for subgroups according to sex, age group, and duration of SLE.
The expected numbers of cancers were calculated by multiplying person-years at risk in the cohort by the geographically appropriate age-, sex-, and calendar year–matched cancer rates. The person-years for each subject were determined by subtracting the later of 2 entry dates (the beginning of the cancer registry observation interval or the first visit to the respective lupus clinic) from the earlier of 2 exit dates (end date of cancer registry data or death). SIRs were obtained by dividing the observed number of cancers by the expected number, and 95% CIs were calculated using previously described methods for Poisson parameters (20).
Since SIRs for cancer may differ across centers, we also fit a hierarchical random-effects model allowing differences among centers in their cancer rates, rather than assuming a single fixed rate across all centers. SIR estimation using this hierarchical modeling represents a compromise between the pooling of data across sites (our primary analysis, which assumes no variation in cancer experience from one center to the next) versus independent estimates for each center (the other extreme, which would preclude estimation of the SIR across all centers). We used the Gibbs sampler as implemented with WinBUGS 1.4 software to estimate the model parameters, with 95% credible intervals (21).
At the first level of our hierarchical model, the number of observed cancers within each center (i) was assumed to follow a Poisson distribution, with mean θi = λiti, where λi is the rate for center i and ti is the total person-years at center i. The second level of the model specifies a gamma distribution, λi∼gamma (α, β), for the center mean. Diffuse prior distributions were used for the gamma priors, α∼exp (1.0) and β∼gamma (0.1,1.0) so that the data would dominate the posterior distribution (22). A robustness check of variations in the prior parameters was also performed, as detailed below.
The 9,547 patients were observed for a total of 76,948 patient-years (average followup 8 years). The calendar period of observation was 1958–2000. Most (71%) of the patients entered into the observation interval within the first 2 years of their SLE diagnosis. As may be expected, given that SLE is a disease primarily of women, 90% of the patients were female (Table 3). In Scotland, patients were identified by hospital discharge data without having specific confirmation of SLE by ACR classification criteria or known clinical confirmation by an appropriate specialist. However, examination of the data from this site revealed that the cancer incidence and patterns were similar to the findings at other centers. Sensitivity analyses with and without data from this site yielded very similar SIRs; therefore, the data from all 23 sites were included for all analyses presented herein.
Table 3. Patient-years of observation by age and systemic lupus erythematosus (SLE) duration*
|Age, years|| |
|SLE duration, years|| |
Within the observation interval, 431 cancers occurred. The data confirmed an increased risk of cancer among patients with SLE, particularly for specific cancer subtypes. For all cancers combined, the SIR estimate was 1.15 (95% CI 1.05–1.27). For all hematologic malignancies, it was 2.75 (95% CI 2.13–3.49), and for non-Hodgkin's lymphoma (NHL), it was 3.64 (95% CI 2.63–4.93). The data also suggested an increased risk of lung cancer (SIR 1.37; 95% CI 1.05–1.76), and hepatobiliary cancer (SIR 2.60; 95% CI 1.25–4.78). Further results concerning specific cancer types are shown in Table 4.
Table 4. Cancers observed and expected, with standardized incidence ratios (SIRs) and 95% confidence intervals (95% CIs)*
|Hematologic cancers|| || || || |
| Non-Hodgkin's lymphoma||42||11.5||3.64||2.63–4.93|
| Hodgkin's lymphoma||5||2.1||2.36||0.75–5.51|
|Reproductive cancers|| || || || |
|Other cancers|| || || || |
The results of analyses of SIRs by sex, age, and SLE duration are provided in Table 5. In men, the 95% CI for the SIR estimate for all cancer occurrence included the null value, as well as the possibility of an SIR below or above 1. For hematologic cancers, the 95% CI excluded the null value for both men and women. The results were consistent with similar relative risks for hematologic malignancies among SLE patients of both sexes. For all cancers combined, as well as for hematologic malignancies only, the SIR estimates were highest in patients with early SLE, particularly those who had been diagnosed in the last year. However, the majority of the cancers in SLE patients did occur >1 year after diagnosis.
Table 5. Cancers observed and expected, with standardized incidence ratios (SIRs) and 95% confidence intervals (95% CIs), according to sex, age, and duration of systemic lupus erythematosus (SLE)
|Total cancers|| || || || |
| Sex|| || || || |
| Age, years|| || || || |
| SLE duration, years|| || || || |
|Hematologic cancers|| || || || |
| Sex|| || || || |
| Age, years|| || || || |
| SLE duration, years|| || || || |
To address a potential surveillance or detection bias (23) or the possibility that some of the SLE cases represented paraneoplastic phenomena, we repeated the calculation of the SIRs, excluding all observed cancers in patients with an SLE duration of ≤1 year. This led to an SIR estimate of 1.1 (95% CI 1.0–1.2) for all cancers and 2.5 (95% CI 1.9–3.3) for hematologic cancers.
The Bayesian hierarchical (random-effects) model produced a point estimate for the total cancer SIR of 1.16 (95% CI 1.06–1.27), similar to the results with the primary analysis approach (in which data were pooled). This model used diffuse, noninformative prior distributions for the gamma priors; to investigate the robustness of the hierarchical modeling to changes in the prior distribution, we considered the literature regarding cancer rates in SLE (7–15), and substituted clinical prior values for the gamma prior distribution, i.e., α∼exp (0.7) and β∼gamma (38, 2) for overall cancer rates. The checks for robustness produced SIR and observed cancer rate estimates that were very stable (unchanged within 2 decimal points for the SIR).
In this study we have confirmed that certain cancers, particularly hematologic cancers, occur more frequently in patients with SLE than in the general population. The most important contribution of this work is the definition of the magnitude of risk of NHL in SLE. These results are of importance to both the patient and the physician; awareness of the association should guide appropriate followup care, directed by clinical judgment. Specifically, the presence of symptoms that could be associated with a hematologic malignancy (such as lymphadenopathy and weight loss) should not be ascribed to SLE before some reflection. Of course, such symptoms combined with a history and investigational findings that are consistent with an SLE flare should be treated as a flare, but always with the understanding that if the symptoms are unusually severe or persistent, a malignancy might be present.
Limitations of previous studies related to sample size, completeness of ascertainment, and use of an appropriate comparison population were avoided in the present study. Our sample size was vastly larger than any other clinical cohort previously studied. Cancer occurrence within the subjects at each center was determined systematically by linkage with the appropriate regional cancer registries, and the same registries provided data on comparison population incidence rates. Of course, although we believe our cohort is representative of the general lupus population, claims of representativeness must be made with caution since unobserved selection biases may occur.
Our estimates of the relative risk of cancer in SLE are conservative, since we assumed that any lost-to-followup patients who were not identified in the registry linkages (as having had a tumor or as having died) remained alive and at risk for cancer up to the end of the observation interval at that center. Some of these patients may have moved out of the area served by the registry (for example, to another country) and developed cancer or died without documentation. With our conservative assumption, these persons would still contribute person-time to the cohort, which would inflate the total number of person-years. This would slightly inflate the denominator (expected cancers) for our SIR estimates.
One might thus argue that our approach may have produced results that underestimate cancer risk in SLE to an extent. However, our estimate is within the spectrum of risk magnitude found in most studies of cancer in SLE (Table 1), several of which used the same conservative strategy that we did. In sensitivity analyses, we repeated our calculations using the last date of followup for lost-to-followup patients, for the centers that were able to provide those data. This resulted in the exclusion of several cancer events as well as shortening of the observation time, but the net effect was only a slight change in the estimates, with the SIR being 1.29 (95% CI 1.15–1.44) for all cancers. For NHL, the SIR from the sensitivity analysis using the less conservative strategy was 4.32 (95% CI 2.97–6.07).
We note that there were 2 lost-to-followup patients not found in the regional tumor registry linkage who were recorded (in the national vital statistics linkage) as having died of malignancy in a region served by another tumor registry. One of these patients had a hematologic malignancy (leukemia). These could represent cancers that developed in patients who had moved outside the region served by the cancer registry to which they had been linked. Including these 2 cases would have slightly increased the point estimate for our conservative SIR estimates (for total cancers and hematologic cancers) only to the second decimal point.
We did in fact implement some strategies to prevent excessive biases toward the null value. For example, a bias toward the null value can be created when the observation interval is allowed to extend back to the period prior to enrollment, since this creates an artificial inflation of followup time, termed “immortal patient years” (24). We ensured that this did not occur in our study, beginning the observation interval only at the time of cohort enrollment.
It must be acknowledged that a substantial proportion (29%) of the subjects were enrolled more than 2 years after SLE had been diagnosed. The cohort SIR estimates represent the average SIRs for SLE patients across all time windows, weighted by the amount of person-time in each window. In the setting in which the relative risk of cancer (SLE versus the general population) differs according to duration of SLE, missing person-time in the early time period (i.e., the first 2 years) would mean the overall cohort SIR estimate is a little lower than it should be. To address this, we performed time window analyses for SIR estimates (Table 5), in which the SIR for each time window is calculated for all person-time within that window only. The results in fact do suggest a time trend, such that the increase in cancer incidence in SLE relative to that in the general population is highest early in the course of SLE. The only published study of cancer incidence in an SLE inception cohort was the study by Nived et al (11). In their investigation, the SIR estimate (1.5; 95% CI 0.8–2.6) was, within error, comparable with our estimate. In addition, we have in progress a study to ascertain cancer occurrence in a large SLE inception cohort with a target enrollment of 1,500 patients (25).
Given that only ∼10% of the total SLE population is male, we recognize the limited precision of our estimates in men. However, our data represent the largest cohort of men with SLE ever assembled and provide useful information on malignancy in a clinical cohort of male SLE patients. Our estimates of relative risks of hematologic malignancies were similar in SLE patients of both sexes.
Of interest, the SIR point estimates suggest that the increased risk of cancer is highest early in the course of SLE, particularly in the first year after diagnosis. We acknowledge that some of the cancers detected in the first year of followup might be paraneoplastic presentations masquerading as SLE. Several of the components of the ACR criteria for SLE (18, 19), e.g., positive antinuclear antibodies and cytopenias, are nonspecific and could be seen both in SLE and in hematologic malignancies (26). Thus, we cannot rule out the possibility that the SLE diagnoses in some of the cancer cases that occurred within the first year since SLE diagnosis were in fact paraneoplastic phenomena. This issue might be resolved if the patient continued to meet ACR criteria for SLE even after the malignancy was in remission. However, because intermediate and high-grade NHL often leads to death within a short period of time (27), it may never be established whether the autoimmune disorder would have persisted beyond the active malignancy. We therefore also calculated the SIRs excluding cancers diagnosed in the first year of SLE. Since the estimates changed little, it seems unlikely that the association between cancer and SLE reflects a paraneoplastic process alone. Furthermore, the majority of the cancers occurred more than 1 year after SLE had been diagnosed.
The presence of elevated risk of cancer after the first year of SLE also suggests that our findings are not simply due to the discovery of subclinical malignancies during the diagnostic evaluation (including laboratory and radiographic investigations) for SLE, despite the fact that cancer may be more likely to be recognized in patients with SLE compared with individuals who do not have similar medical followup. Cancer occurring in the general population can remain undetected during life and be found only at autopsy, if at all, unless there is reason for diagnostic evaluation. Therefore, surveillance or detection bias (23) is a potential limitation when investigating cancer risk in patients with chronic disease. In the latter subjects, such as patients with SLE, regular contact with physicians may mean more regular screening procedures (e.g., mammograms and Pap tests), potentially resulting in the early detection of small or early neoplasms that may never have surfaced clinically. This could inflate the data on cancer incidence in the SLE cohort relative to the general population.
There are strong reasons to believe that this potential bias does not entirely explain our findings of increased risk of malignancy in SLE, however. Breast cancer, a neoplasm amenable to screening, was not increased in our study, in contrast to the striking increase in hematologic malignancies. Hematologic malignancies seem unlikely to be subject to this bias since there is no formal screening strategy for early detection. Bias could still operate in that hematologic malignancies may be revealed sooner in an SLE patient (during a periodic clinic visit) than in the general population. This might create a “lead time” in diagnosis but not an increased malignancy incidence, and thus would not bias the incidence rate or SIR if the followup time in the cohort study is adequate. Furthermore, in a study of 1,193 women with SLE, the stage distribution of diagnosed cancer cases did not differ from that in the general population (28), suggesting that increased scrutiny does explain our current results. In addition, some evidence suggests that patients with chronic rheumatic diseases may undergo cancer screening much less often than guidelines recommend (29). Finally, recent data indicate that cancer mortality (not just incidence) is increased in SLE (30), which provides further evidence of a true increased risk of cancer in SLE.
Although our findings have enabled us to precisely define the association between SLE and malignancy, they do not allow us to evaluate mechanisms responsible for the association. Numerous pathogenic mechanisms are possible. Intriguingly, evidence of genetic predispositions common to autoimmunity and hematologic malignancies (31, 32) has implicated abnormalities in apoptosis (cell death regulation) (33) and other pathways.
In an attempt to generate hypotheses regarding events that might link NHL to SLE, we have recently compiled information regarding NHL subtype, which was available for a subset (n = 21) of the cases of NHL that developed in our multicenter international SLE cohort (34). Of these 21 NHL cases, the most common NHL type was diffuse large B cell lymphoma, a relatively aggressive type. Interestingly, this lymphoma subtype appears to dominate the NHLs that develop in rheumatoid arthritis, an autoimmune disease which, like SLE, is associated with an increased risk of NHL (35). In contrast, only 1 NHL case in our SLE cohort (of the 21 in which the subtype was known) was a marginal-zone lymphoma related to mucosa-associated lymphoid tissue (MALT). MALT-type lymphomas, which are low-grade (indolent) lesions, have been associated with primary SS, another autoimmune disease associated with NHL (5).
Given this apparent difference in the subtype predominance of NHL cases that occur in SLE versus primary SS, it appears unlikely that identical pathologic processes are occurring in the cancers that develop in patients with these autoimmune diseases. We are unable, with the design of our current study, to draw conclusions regarding the risk of NHL in SLE patients with secondary SS. However, to evaluate the impact of SS on cancer risk in SLE, we currently have in progress a case–cohort study based on our multicenter cohort sample.
In general, more aggressive lymphoma types have been associated with immunosuppression (36), which might appear to favor the hypothesis that exogenous exposures (i.e., medications) may mediate cancer risk in SLE. The potential impact of immunosuppressive medications on cancer risk (37–41) has created widespread concern for SLE patients and their physicians. However, our finding of increased cancer risk even early in SLE (and thus not likely to be related to cumulative treatment) suggests that drug exposure is not the sole cause of the association. An increased risk of cancers (including hematologic, lung, and hepatobiliary tumors) has been shown in patients with other autoimmune disorders, including rheumatoid arthritis and scleroderma (42, 43), but again the extent to which the association is due to the autoimmune condition itself, or to related exposures (such as medications) is not known.
We observed, in our sample, a decreased incidence of endometrial and possibly breast and ovarian cancers. Endometrial cancer occurrence in clinical cohort studies of malignancy in SLE (7, 9, 12–15) has been reported very rarely, and precise estimates of the effect have, to date, been unavailable. The estimates for breast and ovarian cancer risk in cohort studies of malignancy in SLE have also been quite variable, and generally imprecise.
It is known that both endogenous estrogen levels (44) and unopposed hormone replacement therapy (HRT) increase the risk of endometrial cancer (45, 46). It is possible that this contributed to our observed findings, for two reasons. First, women with SLE are at risk of premature ovarian failure (47), in part due to medication exposure (48), and in those cases endogenous estrogen exposure is arrested. Because of clinical concern that exogenous estrogens may cause lupus flares (49), the overall population of female SLE patients may have been less likely to receive HRT than women in the general population. Of course, at different clinical centers, there is variability regarding prevalence of HRT use. The hypothesis that lower use of HRT in SLE patients mediates a decreased risk of endometrial cancer is perhaps further supported by our SIR point estimates suggesting decreased risk of ovarian and breast cancers in SLE, which are also estrogen-sensitive (50–53). However, some of our preliminary work indicates that factors (as yet unknown) other than HRT may influence breast cancer risk in SLE (54). The lack of homogeneity of the estimates of ovarian and breast cancer risk in previously reported single-center SLE cohorts (7–15) might suggest a complex interplay of risk factors that differ across centers or time.
It could be hypothesized that our finding of an increase in the frequency of certain cancers among SLE patients compared with the general population is partly attributable to differences in exposures to “traditional” cancer risk factors, such as smoking. However, the available data suggest that the prevalence of smoking and several other traditional risk factors in patients with SLE is similar to that in the general population (6). Moreover, factors such as smoking and postmenopausal hormone use (that have been fairly strongly associated with several nonhematologic cancers) are not very strong risk factors for NHL (55, 56).
In summary, our results confirm that certain cancers occur with increased frequency in SLE. In particular, the risk of NHL in patients with SLE was found to be increased ∼3–4-fold compared with the risk in the general population. Numerous pathogenic mechanisms are possible, but hypotheses regarding the reasons for the association remain largely speculative. There are inadequate data on how cancer risk in SLE may be related to medication exposure or to clinical characteristics such as the presence of secondary SS. To evaluate the impact of these exposures on cancer risk in SLE, cooperative studies by the SLICC and CaNIOS are currently in progress. Their results should provide much-needed insight into the causes of the association between cancer and SLE.
We are grateful to Angela Allen and Natalie Gonzalez, who functioned as research co-coordinators responsible for all US sites. We thank Drs. Simon Bowman, Linda Lee, Moon-Ho Leung, Ibraheem Nahr, and Martha Sanchez for their significant assistance in providing patient access and collection of data, and Stephanie Heaton, RN, for assistance with data collection. Staff at the following tumor registries performed the linkage studies and provided population data: the Quebec Tumour Registry, Nova Scotia Tumour Registry, Cancer Care Ontario, Cancer Care Manitoba, Saskatoon Cancer Centre, Alberta Tumour Registry, British Columbia Cancer Registry, Illinois State Cancer Registry, Maryland Cancer Registry, Univeristy of North Carolina Cancer Registry, New York State Tumor Registry, Pennsylvania Cancer Incidence Registry, Alabama State Cancer Registry, Thames Cancer Registry, British West Midlands Cancer Registry, the Swedish Cancer Registry, the Icelandic Cancer Registry, the Information and Statistics Division of the Common Services Agency for NHS Scotland, and the National Statistical Office in Korea. The Pennsylvania Department of Health requested a disclaimer for any responsibility on their part for any analyses, interpretations, or conclusions. The National Death Index and regional or national vital statistics registries provided vital status information on patients who were deceased or lost to followup.