Introduction

Chronic lymphocytic leukemia (CLL) is the most common leukemia in adults in the Western world with an incidence of approximately 5 per 100,000 person-years in Sweden [1]. It affects men approximately twice as frequently as women and is a disease of older people with a median age at diagnosis of approximately 71 years. Clinical management of CLL patients has improved considerably over recent years, to which an increased use of prognostic markers, introduction of new therapeutic agents and procedures, and improved supportive care measures all have contributed [2, 3].

Infectious complications have since long been recognized as a major cause for morbidity and mortality in patients with CLL [4, 5]. Based on studies from 1970s to 1990s, the overall mortality from infections ranges from 30 to 50 %; however, large studies on CLL patients diagnosed in more recent years are lacking [6]. The pathogenesis of infections in CLL patients is multifactorial and is related to inherent immune defects and therapy-related immunosuppression [7]. The inherent immune dysfunction is characterized by defects in both humoral and cell-mediated immunity. Cytotoxic chemotherapy also targets normal immune cells, thus exacerbating this immunosuppression. Splenectomy or hyposplenism (secondary to irradiation to spleen) performed for complications of CLL including autoimmune hemolytic anemia or thrombocytopenia, hypersplenism, and symptomatic splenomegaly is associated with complex innate and adaptive immune responses [8]. Overwhelming post-splenectomy infection is a rare medical emergency [9, 10]. Use of alemtuzumab is associated with neutropenia and reductions in B, T, and NK cells and rituximab with transient reduction of B cell counts; both antibodies are associated with increased risk of infections [11].

Apart from data based on clinical trials and smaller case studies, there is a lack of detailed knowledge on bloodstream infections (BSIs) in CLL patients, their impact on survival, and the effect of the anti-tumoral treatments on the distribution of causative pathogens over time. We thus performed a study, based on blood cultures sampled from CLL patients which were analyzed 1988–2006 at the Clinical Microbiology Laboratory, Karolinska University Hospital Solna, Sweden. The aims were to identify all culture-verified blood cultures, to identify potential changes of causative agents over time, and to record changes in patient survival.

Patient and methods

Data sources and study population

In Sweden, more than 96 % of all new cancers are reported to the national Swedish Cancer Registry, which was established in 1958 and has a high coverage for hematological malignancies [1, 12, 13]. Every physician and pathologist/cytologist is obliged by law to report each occurrence of cancer to the registry.

All blood cultures that were analyzed at the Karolinska University Hospital’s Clinical Microbiology Laboratory were identified, and individuals who had at least one blood culture analyzed between 1988 and 2006 were registered (Fig. 1). Individuals with a drawn blood culture were linked to the nationwide Swedish Cancer Registry to identify all persons diagnosed with CLL. Only CLL patients with a blood culture drawn at the same date or after CLL diagnosis were included in the analyses. A positive blood culture included growth of any microorganism in blood culture. A BSI was defined as growth of bacteria in blood culture excluding commonly found contaminants [14]. Growth of the same microorganism within 7 days was considered as the same BSI episode.

Fig. 1
figure 1

Flowchart describing the study cohort. aUsing the national registration number unique to all Swedish citizens, individuals with a drawn blood culture were linked to the nationwide Swedish Cancer Registry to identify all persons diagnosed with CLL (ICD-7 204.1 with morphology codes 98223, 98231, 98233, 982335, 982336, or missing morphology code (not 80003)) between January 1, 1988 and June 30, 2006. bOne negative blood culture is defined as a negative culture with no previous cultures on the same date. One positive blood culture is defined as a positive culture with no previous positive cultures with the same finding within the same week. cThe centralized Swedish Patient Registry contains information on individual patient-based discharge diagnoses from inpatient (since 1964, with nationwide coverage since 1987) and outpatient (since 2000) care. dInformation from the nationwide Cause of Death Registry was used to retrieve dates of death. eError in dataset (patient registered dead before blood culture date)

Full size image

Information from the nationwide Cause of Death Registry (with data from 1954) was used to retrieve dates of death, and the Swedish Patient Registry (with data from 1964) was used to retrieve information on individual patient-based discharge diagnoses from inpatient and outpatient care [15].

Approval for this study was obtained from the Regional Ethical Review Board in Stockholm (record 2007/1307-31/5).

Statistical methods

Patient characteristics as well as bacterial findings were analyzed for three separate time periods of CLL diagnosis (1988–1993, 1994–1999, and 2000–2006; June 30). To evaluate survival after diagnosis of CLL, by blood culture pattern, data was analyzed using survival analysis methods with the event of interest being all-cause mortality during follow-up. Risk time was accumulated from date of first blood culture following diagnosis until death or December 31, 2009, whichever occurred first. Time since CLL diagnosis was used as the underlying timescale in all analyses. Crude mortality rates (and 95 % confidence intervals [CIs]) were calculated as number of events per 1000 person-years. The exposure of interest was a BSI following CLL diagnosis and was analyzed both as a dichotomous variable (“ever” vs “never”) and as time-varying (“after” vs “before” or “never”). The effect of having had a splenectomy on survival was evaluated in a separate model. To evaluate possible associations, Cox proportional hazard models were used, yielding hazard ratios (HRs) with 95 % CIs. The proportional hazard assumption was formally assessed in all models, and no evidence suggested that this assumption was violated. Analyses were adjusted for time since diagnosis, sex, and age at diagnosis. The effect of the exposure was estimated separately for the three time periods of diagnosis by including an interaction term in the models. As a sensitivity analysis, to obtain similar time to first culture between different calendar periods, the cohort was restricted to patients diagnosed before June 30, 2005 who had their first culture within 1 year following diagnosis. Difference in proportion of BSI between time periods was tested using a chi-squared test or Fisher’s exact test. Difference in mean time to culture was tested using a t test.

Stata v.12 software was used for the statistical analyses (StataCorp, College Station, TX, USA).

Results

A total of 275 CLL patients, diagnosed between January 1, 1988 and June 30, 2006, were included in the study. Among these CLL patients, a total of 1092 blood culture episodes were identified. The mean time between CLL diagnosis and first drawn blood culture was significantly shorter in the most recent time period (6.2, 4.7, and 2.4 years for the time periods 1988–1993, 1994–1999, and 2000–2006, respectively (p value <0.001); Table 1).

Table 1 Demographic and infection characteristics among CLL patients having at least one blood culture episode at/following diagnosis, by time period of diagnosis
Full size table

In the last time period, patients had a significant lower percentage of positive blood cultures (51.5, 45.9, and 28.2 %, respectively (p value 0.007)). Similarly, the percentage of patients with at least one BSI decreased significantly (38.2, 38.5, and 22.4 %, respectively (p value 0.033)). In a sensitivity analysis including only patients diagnosed until 2005 with their positive blood culture within a year of diagnosis, the decline remained statistically significant both for BSI (71.4, 36.4, and 13.8 %; p value 0.008) and for all positive blood cultures (71.4, 50.0, and 24.1 %; p value 0.033; Fig. 2).

Fig. 2
figure 2

Proportion positive blood cultures (BSI and all positive blood cultures) for the whole study population (diagnosed 1988–2006) as well as for the sensitivity analysis (diagnosed 1988–2005 with a blood culture within a year), at/following diagnosis by time period of diagnosis

Full size image

Positive blood cultures were observed in 115 CLL patients, with 215 positive blood cultures for bacteria and two episodes of candidemia. Dominating bacterial species were coagulase-negative staphylococci (CoNS) (22 (31 %), 23 (22 %), and 5 (12 %), for the three time periods, Escherichia coli (11 (16 %), 15 (14 %), and 9 (22 %)), Streptococcus pneumoniae (7 (10 %), 13 (12 %), and 6 (15 %)), Pseudomonas aeruginosa (2 (3 %), 8 (8 %), and 3 (7 %)), Staphylococcus aureus (1 (1 %), 6 (6 %), and 6 (15 %)), and Viridans streptococci (5 (7 %), 6 (6 %), and 2 (5 %)); Table 2). Excluding common contaminants (CoNS, Micrococcus spp., Corynbacterium spp., Bacillus spp., Proprionbacterium acnes, and Gram-positive bacillus (not further specified), 152 bacterial BSI (hereafter BSI) episodes (42, 77, and 33 in respective time period) and two episodes of candidemia were found. Excluding strictly anaerobes and fungi, the percentage of Gram-positive vs Gram-negative bacteria remained stable in the three time periods, respectively (61.4 vs 31.4 %, 54.7 % vs 34.0 %, and 65.9 vs 31.7 %; Table 2).

Table 2 Distribution of positive blood cultures (n = 217 among 115 patients), and BSIs (in brackets), by time period of diagnosis
Full size table

Mortality rates and relative risks of death among CLL patients with at least one positive blood culture or BSI at or following diagnosis, compared to patients with only negative blood culture episodes, or negative blood culture episodes including cultures with finding of contaminants, are presented in Table 3. In the last time period, mortality among patients with a BSI was significantly higher compared to patients with either a contaminant or a negative blood culture (HR = 2.52, 95 % CI 1.44–4.41), as well as compared to patients with negative blood cultures alone (HR = 2.44, 95 % CI 1.38–4.30), when adjusting for time since diagnosis, age at diagnosis, and sex. Being splenectomized did not significantly affect mortality during any of the three time periods (HR = 0.97, 95 % CI 0.38–2.48; HR = 0.45, 95 % CI 0.19–1.06; and HR = 0.64, 95 % CI 0.25–1.60; Table 3).

Table 3 Frequencies, mortality rates, and hazard ratios (HRs) with 95 % confidence interval (CI) among patients with at least one positive blood culture at/following diagnosis, by time period of diagnosis
Full size table

In a sensitivity analysis, including only patients diagnosed until 2005 with their first culture within a year of diagnosis, the mortality rate was higher during 1994–1999 among patients with a BSI, as compared to patients with a contaminant or a negative blood culture (HR = 2.94, 95 % CI 1.04–8.33), as well as compared to patients with a negative blood culture alone (HR = 5.72, 95 % CI 1.74–18.7). During the same time period, the mortality rate was also higher comparing all positive blood cultures to negative blood cultures (HR = 3.55, 95 % CI 1.34–9.44; Table 3, lower panel).

When analyzing positive blood culture as a time-varying exposure, mortality was significantly higher among patients with a BSI both during 1994–1999 (HR = 2.59, 95 % CI 1.74–3.85) and 2000–2006 (HR = 3.21, 95 % CI 1.83–5.62; Table 4). In a sensitivity analysis, the mortality among patients with BSI was significantly higher (HR = 5.07, 95 % CI 1.33–19.3) in the middle time period (data not shown).

Table 4 Frequencies, mortality rates, and hazard ratios (HRs) with 95 % confidence interval (CI) comparing patients before or without BSI to after a BSI, by time period of diagnosis
Full size table

Discussion

In this large study based on 1092 blood culture episodes from 275 CLL patients analyzed between 1998 and 2006 at Karolinska University Hospital Solna, we found a stable proportion of Gram-positive to Gram-negative bacteria and no changes in distribution of bacterial species were observed, with the dominating BSI pathogens being E. coli, S. pneumoniae, P. aeruginosa, S. aureus, and V. streptococci. CoNS was the most frequently detected microorganism in blood cultures but is a frequent contaminant. Our study is of importance given concerns of infectious complications using new treatment options and highlights real-life data in a large cohort of CLL patients.

We found a significant decrease in proportion of positive blood cultures and BSI between 1994 and 1999 and between 2000 and 2006. This is in contrast to earlier experiences from blood cultures collected from the hematology ward 1988–2008 in the same institution, where we observed stable percentage of positive blood cultures over time [16]. A possible explanation could be the introduction of combination treatments including monoclonal antibodies [17–19]. We speculate that more effective therapy given in recent years, which leads to deeper responses and longer time to next progression, results in the observed reduction of positive blood cultures and BSI. Our results are supported by a study from the UK where a trend to lower rates of BSI over three 12-month periods was observed [20]. On the other hand, in more aggressive hematological diseases, BSI seems to have increased in another UK center between 2004 and 2010 [21]. However, anti-bacterial prophylaxis, non-comparable use of anti-cancer therapy, and different overall survival make comparisons difficult. Moreover, stated limitations preclude us from interpreting any trends from our data in proportion of positive blood cultures or BSI.

The stable distribution between Gram-positive and Gram-negative bacteria observed in this study is similar to an observation by Francis et al. [22]. Nevertheless, in two reviews of BSI in hematology and cancer patients and a questionnaire on the etiology of BSI in conjunction with the Fourth European Conference on Infections in Leukemia (2011), a recent trend in increased number of infections caused by Gram-negative bacilli was found [23, 24]. The proposed explanation is decreased use and shorter duration of anti-bacterial prophylaxis and, to a lesser extent, decreased use of indwelling catheters and cytotoxic chemotherapy. Trimethoprim/sulfamethoxazole effectively reduces risk of Pneumocystis jirovecii pneumonia and also reduces the incidence of BSI [25–27]. Considering the scarce use of fluoroquinolone prophylaxis and indwelling catheters in CLL, comparisons of trends in BSI with other malignancies must be done with caution.

During the study period, the anti-tumoral armamentarium has expanded and treatment has remained reserved for patients with systemic symptoms or progressive disease [28]. The monoclonal anti-CD-20 antibody rituximab in combination with FC was shown in 2005 to further improve response rates [18, 19]. For rituximab and other anti-tumoral agents, only scarce information on BSI in CLL exists [6, 11, 18, 19, 29–32]. Furthermore, major infectious complications increase with previously received anti-tumoral therapies [33, 34]. Our results on the most common pathogens in BSI might affect anti-biotic choice when empirical therapy in CLL patients with signs and symptoms of BSI is considered.

We found BSI to be associated with worse prognosis especially during the latter time periods, but limitations of our study may bias our results. The effect was seen both when treating blood cultures as a dichotomous exposure and as time-varying. This has to our knowledge not been shown previously in CLL patients. To evaluate this further, we did sensitivity analysis on a homogenous group of more high-risk patients, by only including patients diagnosed 1988 to 2005 who had a culture within 1 year of diagnosis. The point estimates during 1999–2005 remained of similar magnitude, although slightly smaller as compared to those from the full cohort analysis, and they were no longer statistically significant. However, fewer deaths in this analysis made power to detect potential differences small, which is reflected in the wide confidence intervals. During the middle period, where event counts were larger, a significantly higher mortality was observed among patients with BSI. This, together with the non-significant risk increase in the last time period, support the hypothesis that among more high-risk patients, a positive blood culture results in higher mortality. In a retrospective study of BSI 1988–2008, from the same institution, anti-biotic resistance was stable and should therefore not influence our results on temporal trends in mortality rate [16]. We speculate if CLL patients in later time periods, due to introduction of immunotherapy, had developed therapy-related immunodeficiencies in their adaptive immune response, with less favorable outcomes of BSI. Our study includes a stable number of splenectomies among CLL patients during the study period. No statistically significant differences in mortality were seen when comparing patients with splenectomy to those without. However, the numbers of splenectomies were small, and thus, the results are limited by power.

To our knowledge, this is the first study to examine the risk for and distribution of BSI in patients with CLL. Previous reports have presented BSI in all hematological diseases as a group [16, 35], presented mortality rates for CLL with BSIs or major infections without displaying CLL-specific pathogen distribution [36, 37], or only presented very few BSIs among CLL patients [17]. The main strengths are the large sample size from a single institution, the use of high-quality registries, and the uniform treatment according to national guidelines. Our study has some important limitations. Generalizability is impeded by the retrospective data and that data consists of CLL patients with drawn blood culture, rather than all CLL patients. Information on important clinical descriptors were not available as, for example, anti-tumoral treatment, anti-microbial prophylaxis, anti-microbial resistance, and site of drawn blood culture. For clinical purpose, we therefore excluded pathogens known to be possible contaminants even though delineation of clinical significance can be difficult in the immunocompromised host. For the latter time period, due to administrative censoring, time window for having a blood culture drawn was shorter as compared to that for patients diagnosed during the earlier time periods. Low counts of exposure and events further hampered our possibilities to draw strong conclusions on temporal differences in the sensitivity analyses of BSI and mortality. Lastly, the design of the study, organizing 1.8 million microbial analyzes and then cross-matching data against three national registries, rendered delays in presentation of results.

In summary, based on the largest study to date on BSI in CLL patients, we found a stable proportion of Gram-positive to Gram-negative bacteria for BSIs 1988–2006. The most common pathogens of BSI were E. coli, S. pneumoniae, P. aeruginosa, S. aureus, and V. streptococci, and no temporal change of distribution was observed. Surveying treatment-related infections and optimizing prophylaxis will be important objectives when improving anti-tumoral treatments for CLL.