Increased risk of severe infections and mortality in patients with newly diagnosed systemic lupus erythematosus: a population-based study

Abstract

Objective

To evaluate the risk of severe infection and infection-related mortality among patients with newly diagnosed SLE.

Methods

We conducted an age- and gender-matched cohort study of all patients with incident SLE between 1 January 1997 and 31 March 2015 using administrative health data from British Columbia, Canada. Primary outcome was the first severe infection after SLE onset necessitating hospitalization or occurring during hospitalization. Secondary outcomes were total number of severe infections and infection-related mortality.

Results

We identified 5169 SLE patients and matched them with 25 845 non-SLE individuals from the general population, yielding 955 and 1986 first severe infections during 48 367 and 260 712 person-years follow-up, respectively. The crude incidence rate ratios for first severe infection and infection-related mortality were 2.59 (95% CI: 2.39, 2.80) and 2.20 (95% CI: 1.76, 2.73), respectively. The corresponding adjusted hazard ratios were 1.82 (95% CI: 1.66, 1.99) and 1.61 (95% CI: 1.24, 2.08). SLE patients had an increased risk of a greater total number of severe infections with crude rate ratio of 3.24 (95% CI: 3.06, 3.43) and adjusted rate ratio of 2.07 (95% CI: 1.82, 2.36).

Conclusion

SLE is associated with increased risks of first severe infection (1.8-fold), a greater total number of severe infections (2.1-fold) and infection-related mortality (1.6-fold).

systemic lupus erythematosus, severe infection, mortality, risk, cohort

Rheumatology key messages

• One in five SLE patients developed severe infections; 21% of mortality related to these infections.

• SLE is associated with an 82% increased risk of severe infection when compared to non-SLE.

• SLE is associated with a 61% increased risk of infection-related mortality.

Introduction

Systemic Lupus Erythematosus (SLE) is a chronic disease with a broad spectrum of autoantibodies and clinical manifestations. It is a complex disease in which the body’s immune system mistakenly attacks healthy tissues in many parts of the body. There have been clinical improvements in controlling inflammatory manifestations of SLE, but a recent study suggested that survival rates of SLE patients have not improved in recent years [1] and are still at least 2- to 3-fold greater than the general population [2, 3].

Infections are a leading cause of morbidity and premature mortality in patients with SLE. Previous studies reported that 14–45% of SLE patients had severe infections requiring hospitalization and up to 50% of deaths were due to infections [4–7]. In a European multicentre lupus cohort of 1000 patients from seven countries, 36% of the patients had an infection during follow-up and 25% of all deaths were caused by infection [8], similar to reports from British [9] and Spanish cohorts [10]. Furthermore, the largest European SLE study on 3658 patients observed that 19% suffered from a severe infection [7]. Because these studies used prevalent and clinic-based lupus cohorts, they were subject to an inherent survivorship bias as only healthier survivors were included and previous infections and deaths could not be included. Other studies used selected samples (e.g. children and women) so their conclusions lack generalizability to all SLE patients [11, 12]. The limitations from existing studies including selected samples, small sizes and prevalent cohorts can negatively affect the accuracy of both the absolute and relative risk estimates of infections in SLE at the population level.

To address these knowledge gaps, we conducted a large population-based study of all patients with incident SLE between 1 January 1997 and 31 March 2015 in British Columbia (BC), Canada. The aim of the study is to determine whether SLE is an independent risk factor for severe infections and infection-related mortality compared with the general population.

Methods

Universal healthcare coverage is available for all residents of the province of British Columbia, Canada (4.7 million in 2015) [13]. Population Data BC includes data on all provincially funded healthcare service data from 1 January 1990 to 31 March 2015, including all registration information on healthcare professional visits [14], hospitalizations [15], cancer registry [16], vital statistics [17] and all dispensed medications in outpatient settings for all BC residents since 1 January 1996 [18]. Several population-based studies have been successfully conducted using Population Data BC [19–22].

Study design and cohort definitions

Using data from Population Data BC, we assembled a 1:5 matched cohort study with incident SLE patients (SLE cohort) compared with age-, gender- and index year-matched individuals who were randomly selected from the general population (non-SLE cohort).

SLE cohort

The case definition of incident SLE included the following: (i) age ≥ 18 years; (ii) two principal International Classification of Disease, Ninth Revision, Clinical Modification (ICD-9-CM 710.0) or ICD Tenth Revision, Clinical Modification (ICD-10-CM M32.1, M32.8, M32.9) codes for SLE at least 2 months apart within 2 years from any type of physician or hospital visit; and (iii) no SLE diagnosis in a 7-year run-in period prior to the first ICD code for SLE to ensure incident SLE cases. Of the SLE patients, 99.4% had at least one of the two ICD codes diagnosed by rheumatologists or from the hospitalization dataset [23]. This definition has 97.6% sensitivity and 97.5% positive predictive value in the Swedish registry data [24]. The date of the second ICD in the pair of ICD codes to confirm SLE was defined as the SLE diagnosis date. Once a patient was confirmed to have incident SLE, a look-back algorithm was applied to search for SLE-related resource use in the patient’s history. The date of the first ever ICD code for SLE was defined as the index date (i.e. SLE onset date), which was the start of the study follow-up for an SLE patient.

Non-SLE cohort

A total of 2 000 000 randomly selected BC residents registered with the provincial medical services plan during the study period were used to establish the comparison non-SLE cohort. The randomly selected individuals without any history of SLE were matched to SLE patients (5:1) on the same year of birth and gender, and were assigned the SLE index date (i.e. first-ever SLE visit) and the SLE diagnosis date (i.e. the date of the second ICD code in the pair of ICD codes to confirm SLE) of the SLE patient they were matched to. Because by design the SLE patient was alive between the patient’s index date and diagnosis date, to avoid immortal time bias, the corresponding selected non-SLE individuals had to remain alive between the assigned SLE index date and assigned diagnosis date.

Ascertainment of outcomes

The primary outcome was the first severe infection during follow-up. Severe infections were defined as infections necessitating admission to hospital or occurring during a hospitalization [25]. Fifty-eight different types of infections (Supplementary Table S1, available at Rheumatology online) were selected a priori by a panel of experts and identified using ICD-9 and ICD-10 codes [25]. Although the list of ICD codes used to define severe infection was originally from an RA population, it already exhaustively included major infection types and therefore we are confident they are applicable to SLE given the similarities between RA and SLE. We chose infections necessitating hospitalization as this case definition has a 95.4% positive predictive value to identify severe infections [26]. Secondary outcomes were the total number of severe infections and infection-related mortality during follow-up. We defined the latter secondary outcome as a death with at least one of the above 58 types of infections listed as the primary cause of death or as other contributing causes of death in the death axis coding as recorded in the individual’s vital statistics record [17].

Covariate assessment

Baseline covariates were assessed within 12 months prior to the index date (first ever ICD code for SLE). Covariates included prior hospitalized infections before the index date, gender, age, the modified Romano Charlson comorbidity index for administrative data [27], baseline medication use categorized as ever use or never use (glucocorticoids, fibrates, statins, anti-diabetic medications, anticoagulant therapy, other cardiovascular disease drugs, NSAIDs, hormone replacement therapy, cyclooxygenase-2 inhibitors and oral contraceptives), comorbidities (hypertension, cerebrovascular disease, alcoholism, ischaemic heart disease, myocardial infarction, congestive heart failure, depression, malignancy and chronic obstructive pulmonary disease-related diseases), and health resource utilization (number of outpatient visits and hospitalizations).

Statistical analyses

We calculated the incidence rates (IRs) of outcomes per 1000 person-years. For the primary outcome (first severe infection during follow-up), individuals were followed from the index date until they either experienced the first severe infection, died, left BC or the study period ended, whichever happened first. To compute IRs for the secondary outcomes (total number of severe infections and infection-related death), follow-up ended at death, migration out of BC or the end of study, whichever occurred first, but continued beyond first severe infection.

To further adjust for potential confounders, multivariable Cox proportional hazard models [28] were used to compute the adjusted hazard ratios (aHR) of infections and infection-related mortality for the SLE cohort compared with the non-SLE cohort, adjusting for baseline variables. Results were reported as aHR with 95% CI. Additionally, we used Poisson regression with over-dispersion [29] to determine the adjusted rate ratios (aRR) of total number of severe infections that occurred during follow-up for SLE compared with the non-SLE cohort.

Sensitivity analyses

To examine the robustness of our results, we performed three sensitivity analyses. First, to assess the effect of an unmeasured confounder (i.e. smoking), we calculated the aHR and aRR by adding the simulated unmeasured confounder in the multivariable Cox and Poisson models, respectively. To simulate the smoking distribution for individuals, we used a smoking prevalence ranging from 42% to 46% in the SLE cohort [30], a prevalence of 31% for the non-SLE cohort (corresponding to the estimated prevalence of smoking for the general population of Canada aged 15 and older) [31] and odds ratios (OR) for the association between smoking and infection ranging from 2.20 to 2.60 [30]. Second, we used the Fine-Gray method [32] to compute the crude cumulative incidence function (CIF) of first severe infections and infection-related mortality, while accounting for competing risks of death due to causes unrelated to infection. Gray’s test [33] was used for comparing the crude CIFs between the two cohorts. To further adjust for potential confounders, multivariable subdistribution proportional hazard models were used [34]. Last, because the medication data were fully captured for all BC residents only after 1 January 1996 [18], there existed uncertainty about the baseline medication data (12 months prior to the index date) for 611 SLE cases whose index date was before 1 January 1997. We therefore conducted sensitivity analyses that excluded these 611 SLE cases.

All statistical analyses used SAS version 9.4 (SAS Institute, Cary, North Carolina, USA).

Ethical approval

All inferences, opinions and conclusions drawn in this research are those of the authors, and do not reflect the opinions or policies of the Data Steward(s). No personal identifying information was made available as part of this study. Procedures used were in compliance with British Columbia’s Freedom of Information and Privacy Protection Act. Ethical approval was obtained from the University of British Columbia’s Behavioural Research Ethics Board (H15-00887).

Results

Baseline characteristics

Table 1 summarizes the baseline characteristics for SLE and non-SLE cohorts. During the study period, we identified 5169 newly diagnosed SLE patients (86% female) with mean age of 46.9 years at the index date (first ever ICD code for SLE). The mean and median time between the index date (first ever ICD SLE code) and the SLE diagnosis date (the second ICD code in the pair of ICD codes to confirm SLE) was 3.1 and 0.9 years, respectively. Among those who had prior severe infection within a year before the SLE index date, the average duration from the most recent prior severe infection to the SLE index date was 154.6 days for the SLE cohort and 178.0 days for the non-SLE cohort.

Table 1

Baseline characteristics of individuals with SLE and without SLE

Variablea	SLE cohort (n = 5169)	Non-SLE cohort (n = 25 845)
Demographics
Age, mean (median)	46.9 (47)	46.9 (47)
Female, n (%)	4384 (86.2)	22 270 (86.2)
Rural, n (%)	785 (15.2)	3334 (12.9)
Neighbourhood income quintile, n (%)
1 (Lowest)	1014 (19.6)	4380 (17.0)
2	950 (18.4)	4558 (17.6)
3	978 (18.9)	4577 (17.7)
4	922 (17.8)	4841 (18.7)
5 (Highest)	858 (16.6)	4762 (18.4)
Unknown	447 (8.7)	2727 (10.6)
Health resource utilization, mean (median)a
Number of outpatient visits	22.9 (19.0)	7.0 (10.6)
Number of hospitalizations	0.1 (0.0)	0.0 (0.0)
Comorbidities, n (%)a
Alcoholism	48 (0.9)	133 (0.5)
Hypertension	781 (15.1)	3253 (12.6)
Cerebrovascular accidents	64 (1.2)	90 (0.4)
Ischaemic heart disease	345 (6.7)	697 (2.7)
Myocardial infarction	31 (0.6)	77 (0.3)
Congestive heart failure	82 (1.6)	161 (0.6)
COPD-related diseases	131 (2.5)	329 (1.3)
Depression	722 (14.0)	2284 (8.8)
Malignancy	261 (5.1)	856 (3.3)
Charlson comorbidity index, mean (median)	0.6 (0.0)	0.2 (0.0)
Medications, n (%)a
NSAIDs	2030 (39.3)	3697 (14.3)
HRT	492 (9.5)	1434 (5.6)
Glucocorticoids	1281 (24.8)	737 (2.9)
Anticoagulant therapy	166 (3.2)	220 (0.9)
CVD drugs excluding anticoagulant therapy	1049 (20.3)	3446 (13.3)
Fibrates/statins	297 (5.5)	1372 (5.3)
Anti-diabetic medications	169 (3.3)	854 (3.3)
History of infection, n (%)a
Prior hospitalized infection	1105 (21.4)	3095 (12.0)

a

All baseline characteristics were measured over 1 year prior to the start of follow-up except that age was measured at the start date of the follow-up. COPD: chronic obstructive pulmonary disease; CVD: cardiovascular diseases.

Compared with the non-SLE cohort, the SLE cohort had significantly higher numbers of all outpatient visits and hospitalizations, greater Charlson comorbidity index scores and a higher prevalence of all selected comorbidities and prior hospitalized infection. In the SLE cohort, the most used prescriptions during 12 months prior to the index date were NSAIDs and cyclooxygenase-2 inhibitors (39%), glucocorticoids (25%), followed by cardiovascular disease drugs excluding anticoagulant therapy (20%).

Time to the first severe infection

During follow-up we observed 955 first severe infections (mean follow-up time of 9.4 years) in the SLE cohort compared with 1988 (mean follow-up time of 10.1 years) in the non-SLE cohort. The IR for severe infections in the SLE cohort was 19.7 events per 1000 person-years, while the IR in the non-SLE cohort was 7.6 events per 1000 person-years. Among the patients who had infections, the mean time to first infection was 7.4 and 7.9 years from the index date for the SLE and non-SLE cohorts, respectively.

Multivariable Cox proportional hazard models were used to estimate the association of SLE with the first post-SLE-onset infection. The age- and gender-adjusted HR for first severe infection for SLE was 2.67 (95% CI: 2.47, 2.88) compared with the non-SLE cohort. The fully aHR adjusting for all baseline covariates was 1.82 (95% CI: 1.66, 1.99; Table 2).

Table 2

Risk of severe infection in SLE relative to non-SLE during follow-up

Post-SLE diagnosis first severe infection
	SLE cohort (n = 5169)	Non-SLE cohort (n = 25 845)
No. of events	955	1988
IR per 1000 person-years	19.74	7.61
IRR (95% CI) Age and gender adjusted HR (95% CI)	2.59 (2.39, 2.80) 2.67 (2.47, 2.88)	1.00 1.00
Fully adjusted HR (95% CI)a	1.82 (1.66, 1.99)	1.00

Post-SLE total number of severe infections
Infection episodes	1898	3114
IR per 1000 person-years	38.4	11.87
IRR (95% CI) Age and gender adjusted rate ratio (95% CI)	3.24 (3.06, 3.43) 3.28 (2.90, 3.72)	1.00 1.00
Fully adjusted rate ratio (95% CI)a	2.07 (1.82, 2.36)	1.00

Infection-related mortality
No. of infection-related death events	114	269
IR per 1000 person-years	2.17	1.00
IRR (95% CI) Age and gender adjusted HR (95% CI)	2.20 (1.76, 2.73) 2.34 (1.88, 2.91)	1.00 1.00
Fully adjusted HR (95% CI)a	1.61 (1.24, 2.08)	1.00

a

Adjusted for covariates listed in Table 1. HR: hazard ratio; IR: incidence rate; IRR: incidence rate ratio.

Total number of severe infections

The SLE cohort had a total of 1898 severe infections and 363 SLE patients (7%) had recurring severe infections with a range of 2–20 episodes, while the non-SLE cohort had 3114 severe infections, of which 579 individuals (2%) had recurring severe infections with a range of 2–15 episodes. In the multivariable over-dispersed Poisson regression analysis for rate of severe infection, SLE was also associated with an increased risk of a greater total number of severe infections after adjusting for baseline covariates (age- and gender-adjusted RR = 3.28; 95% CI: 2.90, 3.72). The fully aRR was 2.07 (95% CI: 1.82, 2.36; Table 2).

Mortality related to infection

During follow-up, there were 539 deaths in SLE patients, of which 114 (21%) were related to severe infection (Table 2). In comparison, in the non-SLE cohort, we observed 1495 deaths in total and 269 (18%) deaths were related to severe infection (Table 2). The incidence rate ratio for infection-related mortality between the SLE and non-SLE cohort was 2.17 (95% CI: 1.76, 2.73; Table 2). The age- and gender-adjusted HR was 2.34 (95% CI: 1.88, 2.91; Table 2). After further adjustment for baseline covariates, the aHR of infection-related mortality for the SLE compared with the non-SLE cohort was 1.61 (95% CI: 1.24, 2.08; Table 2).

Sensitivity analyses

We performed three sensitivity analyses. First, multivariable Cox proportional hazard models were used to estimate the association of SLE with the first post-SLE-onset infection and infection-related mortality adjusting for baseline covariates and the unmeasured confounder, smoking history. Similarly adjusting for baseline covariates and smoking history, a Poisson count model was used to estimate the association of SLE with the total number of severe infections. Table 3 reports the comparison of the results from the primary analysis with sensitivity analyses. The aHR of first severe infection and infection-related mortality for SLE and aRR of total number of infections for SLE remained significant, but attenuated at values of 46% smoking prevalence in the SLE cohort and OR of 2.60 for the association between smoking and infection (Table 3). Secondly, after accounting for the competing risk of death due to causes unrelated to infection, CIFs (Figs 1 and 2) and Gray’s tests show patients in the SLE cohort had a statistically significant faster rate to their first severe infection and infection-related death than individuals in the non-SLE cohort (P-value < 0.001). Using subdistribution models, the aHR also remained significant, but the effect sizes were slightly attenuated for infection-related mortality (Table 3). Last, the aHR and aRR remained statistically significant for severe infection and infection-related mortality when using individuals with index date after 1 January 1997 only (Table 3).

Fig. 1

Cumulative incidence functions of first severe infection in SLE and non-SLE cohorts

Cumulative incidence was estimated adjusting for other causes of death as competing events.

Fig. 2

Cumulative incidence functions of infection-related death in SLE and non-SLE cohorts

Cumulative incidence was estimated adjusting for other causes of death as competing events.

Table 3

Sensitivity analyses for the risk of severe infection and infection-related mortality in SLE

Analyses	Post-SLE first severe infection, aHR (95% CI)a	Post-SLE total number of severe infections, aRR (95% CI)a	Infection-related death, aHR (95% CI)a
Primary analyses	1.82 (1.66, 1.99)	2.07 (1.82, 2.36)	1.61 (1.24, 2.08)
Sensitivity analyses modelling smoking with prevalence = 42% and OR = 2.2	1.68 (1.54, 1.84)	1.93 (1.81, 2.07)	1.56 (1.20, 2.01)
Sensitivity analyses modelling smoking with prevalence = 42% and OR = 2.6	1.64 (1.50, 1.80)	1.89 (1.77, 2.02)	1.53 (1.19, 1.99)
Sensitivity analyses modelling smoking with prevalence = 46% and OR = 2.2	1.65 (1.51, 1.81)	1.90 (1.78, 2.04)	1.54 (1.19, 2.00)
Sensitivity analyses modelling smoking with prevalence = 46% and OR = 2.6	1.60 (1.47, 1.76)	1.85 (1.73, 1.98)	1.52 (1.18, 1.97)
Sensitivity analyses accounting for competing risk of death due to causes unrelated to infection	1.85 (1.68, 2.03)	NA	1.51 (1.16, 1.96)
Sensitivity analyses excluding cases with index date earlier than 1 January 1997	1.76 (1.59, 1.94)	1.95 (1.81, 2.10)	1.57 (1.19, 2.08)

a

Adjusted for covariates listed in Table 1. aHR: adjusted hazard ratio; aRR: adjusted rate ratio; NA: not applicable; OR: odds ratio.

Discussion

To our knowledge, this is the first study to evaluate the risk of severe infections in a large population-based and incident SLE cohort. We observed that almost one in five SLE patients developed severe infection. Compared with the general population, SLE patients demonstrated significantly increased risks for first severe infection (1.8-fold), total number of severe infections (2.1-fold) and infection-related mortality (1.6-fold). These risks were independent of traditional risk factors for infection and the results remain robust in the presence of an unmeasured confounder (smoking) and competing risk of death.

The observed cumulative incidence of infection in 19% of all 5169 SLE patients is consistent with previous studies using prevalent cohorts [4–7]. We also observed that 21% of overall mortality was related to severe infection, a percentage which is very close to a US study conducted in 1995 using a prevalent cohort over a study period of 11 years [35]. In terms of risk difference, compared with the general population, there was an increased risk for infections among patients with SLE. These findings are in agreement with previous studies of severe infections in SLE patients [9, 10]. We deem that our findings are generalizable to the general SLE population due to the large population-based incident SLE cohort, as compared with previous studies that had a small sample size from selected samples (< 150 hospitalized patients, for example [7, 8]) Our study also has the advantage of being able to adjust for important infection risk factors such as comorbidities, income level, medications, prior hospitalized infection, unmeasured confounders and competing risks. For example, we observed a greater proportion of NSAIDs (39.3% vs 14.3%) and glucocorticoids (24.8% vs 2.9%) users in the SLE cohort than in the non-SLE cohort during the year prior to the index date. The difference can be explained by the fact that NSAIDs and glucocorticoids are used extensively for the treatment of ongoing inflammation preceding the time of SLE diagnosis.

The observed increased risk of infection in SLE patients may be a result of both intrinsic and extrinsic factors. Intrinsic factors include the immune system dysfunctions, with more active SLE with impaired chemotaxis and phagocytosis of macrophages and polymorphonuclear cells diminishing the body’s immune complexes and abnormal T-cell production [36]. On the extrinsic side, the use of immunosuppressive medications and glucocorticoids has been studied previously [37–39]. These medications inhibit the immunologic network and therefore decrease the resistance to a wide variety of bacterial, viral, protozoal and fungal agents [38]. Conversely, the elevated risk for infection due to the immunosuppressive actions may be counterbalanced by the benefit of these medications in controlling inflammation [25]. This work examines the total effect of these intrinsic and extrinsic factors on infections. Thus, we did not adjust for the medication uses (e.g. glucocorticoids use) during follow-up because this would mean adjusting for mediators, which is inappropriate for studying the total risk of having SLE on infections. Future research can focus on quantifying the relative contributions of these intrinsic and extrinsic factors on the increased infection risk in SLE patients. We note that in such analysis, simply entering the use of medications during follow-up as time-varying covariates in a traditional time-dependent multivariable Cox model can yield biased effect estimates because medication use is both a time-dependent confounder and a mediator.

Our study has limitations common to observational studies that use administrative data. Although the wealth of information contained in administrative data can be used to assess the SLE-infection association, it was not collected for research purposes and often lacks validation and clinical details. Consequently, data quality may cause uncertainty in our research validity. First, uncertainty around the diagnostic accuracy of SLE cannot be completely ruled out even though a strict algorithm with high positive predictive value (97.5%) for SLE diagnostic accuracy was used. There may be discrepancies in SLE accuracy resulting from population heterogeneity. Nevertheless, 99.4% of SLE patients in this study were diagnosed by rheumatologists or from hospitalizations and misclassification would be a conservative bias where the observed effect would bias the estimates towards the null. Measuring the timing of SLE onset using administrative data may also be imprecise; for example, there would be no record if patients did not have any healthcare utilization during the first SLE flare. Second, due to inaccuracy in prescription data (including glucocorticoid use at baseline) before 1996, we conducted a sensitivity analyses on individuals with index date on or later than 1 January 1997 only. The corresponding results remained robust. Third, although we adjusted for all known risk factors for infections available in our data, there are other risk factors such as smoking for which data are currently unavailable. Nonetheless, in our sensitivity analyses adjusting for plausible unmeasured confounders, the results remained statistically significant for each of the outcomes using values of 46% prevalence of smoking in the SLE cohort and an odds ratio of 2.60 for the association between smoking and the infection. Last, because there is a lack of details for non-hospitalized infection in administrative data, there may have been severe infections (e.g. endemic mycoses) that did not result in hospitalization [40]. As a result, our results may have underestimated the risk of infections.

Despite the limitations, our study has notable strengths. First, we used a large Canadian administrative dataset with a substantial timespan, from 1997 to 2015, based on the entire SLE population in BC, which makes our results more generalizable. To the best of our knowledge, this is the largest SLE cohort assembled to date to study the relationship between SLE and infection. Second, using an incident cohort can avoid the survival bias associated with prevalent cohorts [41]. Finally, unlike previous studies, we performed sensitivity analyses to account for the effect of unmeasured confounders, competing risk of death and inaccuracy of the prescription data before 1996, which make our results robust and less biased.

Our findings highlight the risk for severe infection and shed light on important implications for SLE patients and their treating physicians. Increased awareness of the risk of infections can identify their early signs and potentially prevent hospitalizations. We suggest that in the clinical setting, physician visits provide an opportunity to promote infection prevention behaviour for SLE patients. For instance, in some cases, infections may be prevented with vaccinations [42], and regular physician consultations could be valuable for awareness and promotion of appropriate vaccination strategies.

There is a need for additional research on the risk of infection in SLE patients given the large burden and possibility for prevention. Future studies should aim to comprehensively examine risk factors for severe infection in SLE patients to develop and implement strategies for the prevention of severe infection and infection-related mortality. One plausible reason for the increased risk of infection in SLE patients is the inflammation that may lead to the use of glucocorticoids for disease management [25]. Appropriate and opportune management of disease activity in SLE can decrease inflammation and potentially mitigate the risk of severe infections while minimizing the use of glucocorticoids. To reduce the infection-related morbidity and mortality in SLE, evidence on the risk factors for and burden of inflammation in SLE is required.

In summary, this is the first comprehensive population-based study assessing the SLE–infection association. Our study demonstrates that one in five SLE patients developed severe infections and 21% of overall mortality was related to severe infection. SLE patients have 82%, 107% and 61% increased risks of developing the first severe infection, a greater total number of severe infections and infection-related mortality compared with the general population, demonstrating that SLE is an independent risk factor for severe infection and infection-related mortality. This result expands on the findings of previous studies and has important implications for the prevention, screening and treatment of infections. We recommend a closer surveillance for severe infections in SLE patients and risk assessment for severe infections for SLE patients after diagnosis. Further studies are warranted to establish the identity of risk factors for infections in SLE patients to develop personalized treatment regimens and to select treatment in practice by synthesizing patient information.

Acknowledgements

The authors thank Shelby Marozoff for her editorial assistance in the preparation of this manuscript. We would like to thank the Ministry of Health of British Columbia and Population Data BC for providing access to the administrative data. Dr J. Antonio Avina-Zubieta is the BC Lupus Society Research Scholar and the Walter and Marilyn Booth Research Scholar.

Funding: Canadian Institutes for Health Research (CIHR team grant THC-135235) and  Natural Sciences and Engineering Research Council of Canada (NSERC discovery grant RGPIN-2018–04313).

Disclosure statement: The authors have declared no conflicts of interest.

Data availability statement

All the data are made available via Population Data BC (https://www.popdata.bc.ca/).

Supplementary data

Supplementary data are available at Rheumatology online.

Author notes

Hui Xie and J. Antonio Aviña-Zubieta contributed equally to this study.