Bacteremia in Systemic Lupus Erythematosus in Patients from a Spanish Registry: Risk Factors, Clinical and Microbiological Characteristics, and Outcomes

Objective. To describe the incidence of bacteremia in a large multicentric cohort of patients with systemic lupus erythematosus (SLE) and their clinical characteristics and to identify risk factors. Methods. All bacteremic episodes from the Spanish RELESSER registry were included. Clinical and laboratory characteristics concerning bacteremia and SLE status, as well as comorbidities at the time of infection, were retrospectively collected. A comparison with sex- and age-matched SLE controls without bacteremia was made. A logistic regression was conducted. Results. The study included 114 episodes of bacteremia in 83 patients. The incidence rate was 2.7/1000 patient-years. At the time of bacteremia, the median age was 40.5 (range: 8–90) years, and 88.6% of patients were female. The Safety of Estrogens in Lupus Erythematosus National Assessment–Systemic Lupus Erythematosus Disease Activity Index was 4 [interquartile range (IQR) 8]; 41% had an SLE flare (66% severe); Systemic Lupus International Collaborating Clinics/American College of Rheumatology Damage Index was 3 (IQR 4). A comorbidity was recorded in 64% of cases. At the time of bacteremia, 88.6% received corticosteroids (68.6% > 10 mg/day) and 57% immunosuppressors. Gram-negative bacilli, most frequently Escherichia coli (29.8%), caused 52.6% of the episodes. The bacteremia-related mortality was 14% and bacteremia was recurrent in 27.2% of cases. A dose-response relationship was found between corticosteroids and bacteremia risk. In the multivariate analysis, these factors were associated with bacteremia: elevated creatinine (OR 1.31, 95% CI 1.01–1.70; p = 0.045), diabetes (OR 6.01, 95% CI 2.26–15.95; p < 0.001), cancer (OR 5.32, 95% CI 2.23–12.70; p < 0.001), immunosuppressors (OR 6.35, 95% CI 3.42–11.77; p < 0.001), and damage (OR 1.65, 95% CI 1.31–2.09; p < 0.001). Conclusion. Bacteremia occurred mostly in patients with active SLE and was frequently associated with severe flares and corticosteroid use. Recurrence and mortality were high. Immunosuppressors, comorbidities, and disease-related damage were associated with bacteremia.

Infections remain a major cause of morbidity and mortality in patients with systemic lupus erythematosus (SLE), and 30% will have at least 1 serious infection at some time 1 .
Although respiratory infections are the most common severe infections in SLE, according to RELESSER (Spanish Society of Rheumatology Lupus Registry) data, bacteremia has a greater effect on mortality 2 . However, the prevalence of bacteremia in patients with SLE is not well known, with reports ranging between 7% and 49% across several studies 3,4,5,6 . In fact, according to a recent population-based study, the prevalence of this life-threatening complication is probably increasing 7 .
Additionally, the incidence of bacteremia of unknown origin was significantly greater in patients with SLE than in non-SLE controls in 1 study 3 , and the longterm survival rate of patients with SLE following a bacteremic episode was lower 6 . Despite the relevance of bloodstream infections in SLE, few studies have provided detailed information concerning the nature and significance of bacteremia in patients with SLE and none has been conducted in a European country.
The aim of this retrospective, case-control study is to describe the cumulative incidence, microbiology, and outcomes of bloodstream infections in a wide national SLE cohort from the RELESSER registry, and to examine risk factors associated with bacteremic events.

MATERIALS AND METHODS
All centers with bacteremic episodes registered with RELESSER were invited to participate in the study. The RELESSER registry includes data from 3679 patients with SLE [American College of Rheumatology (ACR) 1997 criteria] 8 from 45 Spanish hospitals. The methodologic and general characteristics of the RELESSER registry have been published previously 9 .
New information (i.e., not available in the RELESSER registry) concerning both bacteremia and SLE status at the time of the infection was retrospectively collected, including the etiologic agent, potential sources of bacteremia, antibiogram testing, treatment, and bacteremia-related outcomes. SLE activity at the time of bacteremia was estimated using Safety of Estrogens in Lupus Erythematosus National Assessment-Systemic Lupus Erythematosus Disease Activity Index (SELENA-SLEDAI), while flares, and flare severity, were defined using SELENA-SLEDAI flare index criteria 10 . Damage at the time of bacteremia was measured using the Systemic Lupus International Collaborating Clinics (SLICC)/ACR damage index.
A comparison was made with SLE controls without bacteremia who were matched for sex and age at diagnosis (1 case/6 controls), using the last visit recorded in the RELESSER-T registry database as the reference visit 11 .
Only bacteriologically proven bacteremia with sufficient clinical data were ultimately included in the analysis. Clinically significant bacteremia was defined as a positive blood culture and any sign or symptom of sepsis or a systemic inflammatory response. The presence of coagulase-negative Staphylococcus sp., Streptococcus viridans group, Corynebacterium sp., Propionibacterium sp., or Bacillus sp. in just 1 blood culture bottle was considered the result of contamination 12 . Bacteremia was classified as polymicrobial if 2 microorganisms, not usually considered contaminants, were obtained from blood cultures.
Nosocomial bloodstream infections were defined according to US Centers for Disease Control criteria 13 . The source was considered accurately established if the microorganism was isolated both from blood and the focus at the same time.
The Pitt Bacteremia Score was used as a numerical measure of the bacteremia's severity, with values above 8 having been previously associated with mortality 14 .
Common definitions for the main comorbidities (e.g., diabetes, cancer, etc.) were used and a vital prognosis of the comorbidity was assessed using McCabe and Jackson's criteria 15 .
Empirical antimicrobial therapy was defined as the initial therapy prior to the availability of blood culture results. Appropriate antimicrobial therapy was defined as the administration of any antimicrobial agent to which the causative organism was considered susceptible according to antibiogram results. Antibiotic multiresistance was defined as resistance or intermediate sensitivity to 1 or more antibiotics from 3 different categories in susceptibility testing.
The length of antimicrobial therapy was defined as the time from the first to the last day of an appropriate antimicrobial regimen. Total antimicrobial days were calculated on the basis of the length of therapy with each appropriate antimicrobial agent (e.g., 7 days of gentamicin and 7 days of carbapenem would represent 14 total antimicrobial therapy days).
We defined bacteremic-related mortality as any death occurring in a patient without previous disease-related severity and that was temporally related to a bacteremic event.
Statistical analysis. Each bacteremic episode was considered for analysis, and descriptive analyses were carried out. Numerical variables are expressed as the mean and SD for those having a normal distribution, and as median and interquartile ranges for non-normal distributions (Kolmogorov test). The categorical variables are described by absolute frequencies and percentages.
A bivariate analysis was performed to identify any differences between patients with and without bacteremia, using the chi-square test for quali-tative-independent variables (or Fisher's exact test when necessary), and a Student t test for quantitative-independent variables (or the nonparametric Mann-Whitney U test in the case of non-normal distributions).
A logistic regression was carried out as a multivariate analysis, using a stepwise approach, including variables sequentially, on the basis of likelihood ratio. The following variables were ultimately included in the model: SLE duration, creatinine, diabetes, cancer, immunosuppression use, cyclophosphamide (CYC), SLICC damage index, SELENA-SLEDAI, active lupus nephritis (LN), human immunodeficiency virus or hepatitis C seropositivity, splenectomy, hospitalization by SLE, corticosteroids > 10 mg/day, antimalarials, mycophenolate, renal transplant, and dialysis.
Given the low number of deaths, a multivariable analysis of mortality-associated factors was not considered appropriate.
The IBM-SPSS for Windows statistical software package (v.19.0) was used for all statistical analyses. Significance was defined as p < 0.05.
The study was approved by the Ethics Committee at Doctor Negrín Hospital (board approval number: RELES-SER-2009-01).

RESULTS
Comparative clinical and demographic characteristics between patients with and without bacteremia are shown in Table 1. The first bacteremic episode was recorded on April 1, 1980, and the last on January 3, 2015; however, 80.5% of the bacteremic events took place from 2000 to 2015.
A total of 114 episodes of bacteremia were recorded in 83 patients. The incidence rate was 2.7/1000 patient-years (N total of the cohort: 3658). At the time of the bacteremia, the median age was 40.5 (range: 8-90) years, and 88.6% were female. Median disease duration was 9.7 years [interquartile range (IQR) 16.7], median SELENA-SLEDAI 4 (IQR 8), 41.2% had a coincident SLE flare, and 66% of these flares were severe. SLE was serologically active in 50.9% of cases. Active nephritis was present in 19 (16.7%), median SLICC/ACR damage index: 3 (IQR 4). Some comorbidity was recorded in 64% of cases and proved rapidly or ultimately fatal in 28.1% (McCabe-Jackson criteria), the latter more often involving renal failure (15.8%) or diabetes (11.4%). The complete list of comorbidities recorded is provided in Table 2.
The bacteremia was nosocomial in 35.1% of cases and the source was more frequently urinary (27.2%), followed by respiratory tract (16.7%), intravascular catheter (11.4%), intestinal (8.8%), and cutaneous (7%). The origin remained undetermined in 25.5% of cases based on the predefined criteria. Fever was present in 78.9% of the episodes, 64% developed systemic inflammatory response syndrome, 3.5% endocarditis, and 35% required intensive care unit admission, with multiorgan failure in 22.8% of patients.
Sixty (52.6%) bacteremic episodes were caused by gram-negative bacteria. The most frequent microorganism isolated was Escherichia coli (29.8%), followed by Staphylococcus aureus (16.7%; 22% methicillin-resistant), and Salmonella sp. (10.5%). The bacteremia was polymicrobial in only 4 cases (3.5%). Sixteen percent of the gram-negative enteric bacilli were extended-spectrum b-lactamase-positive, while 17.5% proved to be multidrug-resistant. 236 The Journal of Rheumatology 2020; 47:2; doi:10.3899/jrheum.180882 Personal non-commercial use only. The Journal of Rheumatology Copyright © 2020. All rights reserved.  The distribution of bacteremia's causes, according to the area of acquisition, is shown in Table 3. There were greater proportions of Streptococcus pneumoniae and Salmonella sp. in community-acquired bacteremia compared to those acquired in hospital.
As expected, although the percentage of resistance to typical microorganisms was numerically greater in nosocomial bacteremia compared to community-acquired ones, the differences were not statistically significant in any examined cases (Table 4). However, the prevalence of multiresistance was significantly higher in nosocomial bacteremia (p = 0.005).
E. coli bacteremia was strongly associated with urinary sources (71.0% vs 14.5%; p < 0.001), although no associations between E. coli and active nephritis or E. coli and elevated creatinine were found.
In 68.4% of cases (78/114), antibiotherapy was started before blood culture results were available. This antibiotherapy was ultimately active in susceptibility testing in 56 cases (71.8%), indicating that the appropriate empirically based antibiotic therapy had been carried out in only 49% (56/114) of the episodes.
Bivariate analysis revealed several factors associated with bacteremia (114 bacteremias vs 688 controls; Table 5). Splenectomy was strongly associated with encapsulated microorganism bacteremia (OR 17.79, 95% CI 4.38-72.28). Antimalarials showed some protective effect. Interestingly, a dose-response relationship was found between corticosteroids and bacteremia; that is, the risk of bacteremia increased proportionally with the dose of corticosteroids ( Table 6).
The use of mycophenolate or CYC was not associated with neutropenia at the time of a bacteremic event (data not shown).
In the multivariate analysis (adjusted for disease duration), only these remained statistically significant: elevated creatinine (   per 100,000 per year have been estimated 16 . With the exception of Salmonella sp., the distribution of the most important etiologic agents in our study matched the data reported on community-acquired bloodstream infections in the general population 16 . This is consistent with the predominance of non-nosocomial bacteremia found in this cohort. As in our present study, gram-negative bacilli were the microorganisms most commonly responsible for bacteremia in Asian patients with SLE, which was also the case in a monocentric cohort in Spain 6,17 . The predominance of E. coli (29.8% in this cohort) as an etiologic agent of bacteremia in SLE has similarly been reported by other researchers 3,4 . As expected, in our study E. coli bacteremia was associated with the urinary tract; such infections remain the most common type among patients with SLE 18 . Although this finding could be put in the context of active nephritis as an independent factor associated with bacteremia, unfortunately this variable did not retain significance in the multivariable analysis in our study. This is in contrast to Lim, et al, who found that bacteremia was associated with LN relapse in their monocentric SLE cohort 19 . In another retrospective study, the frequency of LN was higher in urinary tract infection (UTI) cases than in SLE controls without UTI, as was a high frequency of bacteremia, affecting up to 25% of cases 20 . The high rate of Salmonella sp. isolates in our study is not a surprise, because SLE is a well-known risk factor for bacteremia in cases of Salmonella sp. infection 21 . In fact, Abramson, et al point out that SLE is the most common underlying disease for Salmonella sp. bacteremia in hospitalized patients. In their study, these authors demonstrated the inability of patients with SLE to confine the microorganism to the extravascular space 22 .
The association with damage, as measured by the SLICC/ACR damage index, in our multivariable analysis warrants further consideration. In our previous study, which took into account the total number of severe infections in the entire SLE-RELESSER cohort, we also found a significant association with damage 2 . It is tempting to speculate that renal damage could facilitate the dissemination of the microorganism throughout the bloodstream. The finding that elevated creatinine is also linked to bacteremia in the multivariable model, in the absence of "active nephritis" variable in the same model, reinforces this assumption.
The use of oral corticosteroids has been previously recognized as a risk factor in patients with SLE who developed bacteremia 6 . Our observations reinforce the relationship between corticosteroids and bloodstream infections, showing a strong, not previously reported, dose-dependent effect (i.e., the higher the corticosteroid dose, the greater the risk). It is possible that the combination of azotemia and high doses of corticosteroids favor the spread of infection, as has been previously suggested 23 . Unfortunately, in our multivariable analysis, neither of these variables retained statistical significance when included together in the same model. Further, a group of researchers found that prednisone doses during bacteremic episodes represented an independent risk factor for acquiring drug-resistant bacteria in patients with SLE 23 . A strong trend was found when the possible association between microbial resistance and corticosteroid doses was tested (p = 0.07).
It is worth noting that regarding immunosuppressors, only the use of CYC, and not mycophenolate, was ultimately linked to bacteremia in our multivariable analysis. In contrast, the other study that analyzed this topic in SLE patients with bacteremia did not note any differences 6 . Although most studies found that CYC produces more leukopenia compared with mycophenolate, most of these studies were unable to detect any differences in the prevalence of severe infection between the 2 drugs 24,25,26,27,28,29 . The dose of CYC, not usually recorded in studies of SLE-associated bacteremia, could be an important point to consider when addressing this question 30 .
Our study replicates the high rate of recurrent episodes of bacteremia in SLE that have been previously reported 6,17 , a rate notoriously higher than that observed in the general population 31,32 . These results probably reflect the chronic character of SLE, damage, immunosuppressive treatments, and associated comorbidities.
The bacteremia-related mortality rate exhibited in this cohort is an important finding, because it nearly duplicates that previously reported in a monocentric cohort from our country 17 . The inclusion of patients from a wider temporal spectrum could also have some effect, taking into account that the antibiotic armamentarium has improved in recent years, with a consequent reduction in sepsis-related mortality 3 . Moreover, our rate of bacteremia-related mortality was higher when comparing any type of organ transplant, including bone marrow, in patients who underwent these procedures and who were included in the Spanish national registry of severe infections 33 . One additional explanation for such high mortality rates could be that patients with SLE are less likely to receive appropriate antimicrobial therapy within the initial 72 h, when the first symptoms of bacteremia can be easily mistaken for SLE disease activity flares. The finding that antibiotherapy was started in only 68% of cases before blood culture results were available supports this hypothesis. In this sense, it is pertinent to remember that a delay in the starting of antibiotherapy in patients with bacteremia is perhaps the most important risk factor for mortality 34,35,36 . Interestingly, some researchers have reported a high mortality rate from Salmonella sp. infection (ranging from 8.7% to 28.5%) in patients with SLE, despite their having received appropriate antimicrobial treatment 37,38 . Further, reinfection, which frequently occurs in such patients, was the most important risk factor of SLE-related mortality in those having Salmonella infections in 1 study 39 .
Taking into account these data, including the association with high mortality, some recommendations for the empirical treatment of bacteremia in patients with SLE can be advanced. Regarding nosocomial bacteremia, at least for Spanish patients with SLE, we suggest using a carbapenem, perhaps meropenem or imipenem, to ensure proper coverage of Pseudomonas aeruginosa and extended spectrum betalactameses-producer Enterobacteriaceae sp. The use of β-lactamase inhibitors (such as the combination ceftazidimeavibactam) 40 could represent an interesting new alternative. An antibiotic active against methicillin-resistant Staphylococcus sp. should always be added. For community-acquired bacteremia, given the low prevalence of P. aeruginosa, we recommend ertapenem, avoiding the use of wide-spectrum carbapenems. Moreover, ertapenem is also active against S. pneumoniae, a microorganism that should always be considered in community-acquired bacteremia. Of course, before any decision can be taken, one must always consider the local epidemiology.
Our study has further limitations that need to be pointed out. The limitations of the RELESSER registry have been previously described 15 . The most important of these involves its retrospective design, which renders it susceptible to a higher likelihood of measurement mistakes and which lacks sufficient information regarding confounding variables.