Patients and Methods

Patient selection and design. Emory University Hospital is a 587-bed tertiary-care teaching hospital in Atlanta. The first case of VRE BSI in our institution was diagnosed in November 1994 in a patient with neutropenia. We reviewed all episodes of enterococcal BSI among patients with neutropenia from November 1994 to January 2001 at Emory University Hospital. A systematic procedure was used for patient selection. Initially, a list of patients having at least 1 blood culture that grew an Enterococcus species was obtained from computerized clinical microbiology laboratory records. Absolute neutrophil counts (ANCs) for each patient on the list were obtained from electronic medical records, and those patients with an ANC >500 neutrophils/mm³ at the time of the onset of bacteremia were included in the study. Patients with concomitant VRE and VSE bacteremia and patients with BSI caused by enterococcal isolates with intermediate susceptibility to vancomycin were excluded. Most patients were enrolled before the approval of quinupristin/dalfopristin and linezolid; therefore, these agents were not widely available for initial therapy during the study period.

Medical records for each study patient were reviewed. The data obtained for each patient included age, sex, race, hospital location, admission date, discharge date, date of the onset of neutropenia, date of the resolution of neutropenia, date of the onset of bacteremia, ANC within 24 h of the onset of bacteremia, the presence or absence of documented VRE colonization, results of all blood cultures performed during the hospital stay, the result of all catheter-tip cultures performed during the hospital stay, survival status, date of death (if applicable), underlying diseases and comorbidities, administration of chemotherapy during the hospital stay or within 15 days of admission, administration of other immunosuppressive therapies during the hospital stay (corticosteroids, cyclophosphamide, and other cytotoxic agents), administration of antimicrobial agents before and after the onset of bacteremia, history of bonemarrow transplantation, the presence of a central line at the onset of bacteremia, and severity of illness scores (APACHE II) calculated before (as close as possible to the fifth day before the onset of bacteremia) and at the onset of bacteremia.

Definitions. Enterococcal BSI was defined as the isolation of an Enterococcus species from ⩾ 1 blood culture obtained from a patient who met the criteria for BSI as defined by the Centers for Disease Control and Prevention (CDC; Atlanta) [21]. Enterococcal bacteremia occurring >60 days after a previous episode was considered to be a separate BSI. Nosocomial BSI was defined as a BSI in which the first positive blood culture was obtained >72 h after admission. VRE colonization was defined as the isolation of VRE in surveillance or clinical cultures obtained before or within 5 days of the onset of the enterococcal BSI. A copathogen was defined as any bacterial pathogen other than enterococci isolated from ⩾1 blood culture within 60 days of the onset of enterococcal BSI. Polymicrobial bacteremia was defined as the isolation of a copathogen from a blood culture performed on the same day as that of the onset of enterococcal BSI. An active antimicrobial agent was defined as one with activity against the blood-culture isolate, as indicated by standard antimicrobial-susceptibility methods. The presumptive source of bacteremia was defined by clinical judgement. When there was no clinical suggestion of localized infection or when the source was a catheter, the bacteremia was considered primary. The duration of bacteremia was defined as the number of days from the first positive blood culture to the last positive blood culture. Patients for whom blood cultures were positive only on a single day were assigned a duration of bacteremia of <1 day.

Statistical analysis. Sample-size calculations were performed by use of EpiInfo (version 6; CDC). On the basis of figures from previous studies, we assumed a frequency of death among unexposed patients of 40% [14], an attributable mortality to vancomycin resistance of 37% [6], and a ratio of nonexposed to exposed patients of 3:1 [11, 14, 15]. With these assumptions, the sample sizes required to detect a significant difference in mortality, with a power of 80% in a 2-sided test with a probability of .05 between VRE and VSE BSI were 21 and 63, respectively. Statistical analysis was performed by use of SAS software (version 8.2; SAS Institute). Univariate analysis was done by use of the Wilcoxon rank sum test for quantitative variables and the x² statistic or Fisher's exact test for categorical variables.

Survival analysis was performed by use of Kaplan-Meier curves and Cox proportional hazards models to determine predictors of mortality and the duration of bacteremia. Variables included in the multivariate Cox proportional hazards models were those considered to be biologically, clinically, or statistically relevant (P < .15 in univariate analysis). For measures of association, a 2-tailed P < .05 was considered to be significant.

Results

Descriptive statistics. Eighty-three enterococcal BSIs were identified among 77 patients. Most demographic and clinical features were similar among episodes of VRE and VSE BSI (table 1). Follow-up information was available for at least 60 days after the onset of bacteremia for all except 4 patients with BSI (3 VSE and 1 VRE).

Seventy-seven (92.8%) episodes were nosocomial (all VRE and 55 [90.2%] VSE episodes). Seventy-seven episodes occurred in patients with hematologic malignancies (22 VRE and 55 VSE); 3 VSE BSIs occurred in patients with breast cancer who underwent bone-marrow transplantation, 2 in patients with agranulocytosis, and 1 in a patient after orthotopic liver transplantation.

VRE BSI was associated with black race (P = .017) and prior VRE colonization (P = .0001). The number of days of receiving antimicrobial agents before the onset of bacteremia (P = .0007)—particularly glycopeptide (P = .0004), quinolone (P p.0036), and cephalosporin (P = .0007)—was significantly higher among patients with VRE BSI (table 1).

Patients with VSE BSI were more likely to have received active antimicrobial agents within 48 h after the onset of bacteremia than were those with VRE BSI (P = .017), whereas invasive devices were more likely to have been removed during the BSI of patients with VRE BSI (P = .0001). The duration of both hospital stay (P = .002) and neutropenia (P = .013) before the onset of bacteremia and the total duration of bacteremia (P = .0001) were significantly longer for VRE than for VSE BSIs. The severity of illness scores before the onset of bacteremia were based on data collected, on average, 5 days before the onset of bacteremia (median, 5 days; range, 2–18 days).

Follow-up blood cultures were obtained within 24–72 h after the onset of bacteremia in 75 (90.4%) of the BSIs. Of the 8 BSIs for which no follow-up blood cultures were available, 4 (1 VRE and 3 VSE) were followed by death of the patient within 48 h of onset. Blood cultures were obtained with similar frequency during the first 10 days of bacteremia for both VRE and VSE BSIs (0.56 blood cultures/day of VRE bacteremia and 0.46 blood cultures/day of VSE bacteremia; P = .08).

In 2 episodes of VRE BSI, the patients died before they received an active antimicrobial agent. For the remaining 20 episodes of VRE BSI, all were treated initially with chloramphenicol, to which all isolates were susceptible. In 9 patients, chloramphenicol was subsequently replaced with another agent (quinupristin/dalfopristin in 7 patients, linezolid in 1, and oritavancin in 1). The switch from chloramphenicol to a subsequent agent was made after a median of 5.5 days of therapy (range, 3–20 days).

Three patients with VSE BSI died before they received an active antimicrobial agent. Twenty-six VSE BSIs were treated with vancomycin alone, 13 with the combination of vancomycin and gentamicin, 14 with ampicillin after the initial administration of vancomycin, and 2 with ampicillin and gentamicin after the initial administration of vancomycin. Additionally, 1 patient each received ampicillin/sulbactam, ampicillin alone, or ampicillin/sulbactam combined with gentamicin after the initial administration of vancomycin.

Overall mortality during the 60 days after the onset of bacteremia was 47% (64% for VRE and 41% for VSE; relative risk [RR], 1.9; P = .068). Mortality was significantly associated with male sex (RR, 1.5; P = .03) and APACHE II score (P = .01), and a trend toward significant association was also observed for gram-negative copathogen (RR, 2.5; P = .08) and the number of days of neutropenia before the onset of bacteremia (P = .07).

Outcome analysis. Kaplan-Meier curves comparing the survival of patients with VRE BSI and those with VSE BSI are shown in figure 1. We evaluated the effect of vancomycin resistance on survival, using an extended Cox model with 2 Heaviside variables (time dependent), to distinguish the effect of vancomycin resistance before and after the 10 days following the onset of bacteremia. The 10-day cutoff point was selected on the basis of characteristics of the Kaplan-Meier survival curves. Independent predictors of mortality included APACHE II score (hazard ratio [HR], 1.096; 95% confidence interval [CI], 1.029–1.167 for each score point) and vancomycin resistance. The effect of vancomycin resistance was significant only 10 days after the onset of bacteremia (HR, 4.969; 95% CI, 1.208–20.437). The model also controlled for sex, race, gram-negative copathogen, number of days of neutropenia before the onset of bacteremia, and enterococcal species (table 2). An interaction was found between days of neutropenia before the onset of bacteremia and race. When the duration of neutropenia before the onset of bacteremia was ⩽10 days, black race was a significant predictor of death (HR, 12.92; 95% CI, 2.504–66.685), but this effect was absent when the duration of neutropenia before the onset of bacteremia was <10 days. Because the control for days of neutropenia before bacteremia was done by stratification, no estimation of the HR for this variable was obtained.

Predictors of mortality were further evaluated by use of additional Cox models that controlled for the duration of bacteremia. This variable was converted into a bivariate categorical variable, and cutoff points of 2, 3, 4, 5, and 6 days' duration were tested. For the models in which the cutoff points were 2 and 3 days, the duration of bacteremia was associated with mortality, but the relationship did not reach statistical significance. A duration of bacteremia of ⩾4 days was significantly associated with mortality (HR, 6.665; 95% CI, 1.3–34.4), but the difference was observed only 10 days after the onset of bacteremia. The HR observed for different cutoff points for the duration of bacteremia suggests a dose-response relationship between the duration of bacteremia and mortality (table 3). APACHE score (HR, 1.093; 95% CI, 1.03–1.17 for each score point) was also a significant independent predictor of mortality in this model. The presence of gram-negative copathogens showed a trend toward significance (HR, 2.447; 95% CI, 0.946–6.328). Vancomycin resistance (VRE BSI group) was not a significant predictor of mortality in this model (table 4). An interaction similar to the one described for the initial model was present between race and the number of days of neutropenia before the onset of bacteremia.

Figure 2 shows the Kaplan-Meier curves for the duration of bacteremia for VRE and VSE BSIs (P = .0001, log-rank test). We designed a Cox model to assess determinants of the duration of bacteremia. In a model that controlled for race, sex, the removal of invasive devices, ANC and APACHE II score at the onset of bacteremia, the duration of neutropenia after bacteremia, the presence of comorbidities, the absence of treatment with an active antibiotic within 48 h after the onset of bacteremia, outcome, and enterococcal species, vancomycin resistance was the only independent predictor of duration of bacteremia (HR, 3.863; 95% CI, 1.15–12.93).

Discussion

VRE have emerged as an important pathogen among patients with neutropenia, but the impact on patient outcome has not been well established [16, 19, 22]. The present study is the largest to date to have examined the clinical outcomes of enterococcal bacteremia in patients with neutropenia, and our results suggest that vancomycin resistance is an important independent predictor of mortality in this population.

Other studies that have examined the relationship between vancomycin resistance and outcome in patients with enterococcal bacteremia have yielded varying results [7–15]. Studies that have found an increased risk of death among patients with VRE bacteremia tended to focus on specific high-risk patient populations, such as liver-transplant recipients [7], trauma patients [8], or, in the case of our study, patients with neutropenia. In studies that have examined outcome in a study population composed of diverse patient types, a significant association between vancomycin resistance and mortality was observed only in multicenter studies [9, 10]. This apparent inconsistency may be explained by variations in the magnitude of the impact of vancomycin resistance among different patient populations. If the risk of death due to VRE bacteremia in special populations, such as immunocompromised or trauma patients, is higher than in other patient populations, studies that have focused on these populations may have more success in demonstrating the true risk associated with vancomycin resistance. Those that analyze data from heterogeneous study populations may lack statistical power to detect overall differences unless the study populations are large, such as those reported by Bhavnani et al. [9] and Vergis et al. [10].

Although the immune responses to enterococci are poorly understood [23], neutrophils are known to be a critical component of the host immune response to bacterial infections, and virulence factors conferring resistance to phagocytosis may play a particularly important role in the pathogenesis of enterococcal infections [24, 25]. We therefore hypothesize that immunocompromised patients, particularly those who have neutropenia, might be at a particularly high risk for adverse outcomes after infection by antimicrobial-resistant enterococci, for which antimicrobial therapeutic options may be suboptimal.

Controlling for differences in the baseline severity of illness is a critical step when evaluating the impact of antimicrobial resistance on the outcome of bacterial infections. Our data suggest that the observed difference in the outcome between patients with VRE BSI and those with VSE BSI was not confounded by differences in severity of illness between the 2 groups. The groups were very similar with regard to underlying diagnoses and treatment regimens. In addition, we controlled for severity of illness using APACHE II scores. We calculated APACHE II scores both before and at the time of bacteremia onset for both VRE and VSE BSIs, and the median scores for both groups were nearly identical. In our multivariate analyses, we used the scores calculated before the onset of bacteremia (mean and median, 5 days before) to control for severity of illness, but the results were not significantly different when scores calculated at the onset of bacteremia were used. Although APACHE II scores were developed for application in intensive-care units [26], they were successful at predicting the risk of mortality in univariate and multivariate analyses of our patient population.

The increased risk of mortality associated with VRE BSI was not constant over time; it became apparent only 10 days after the onset of bacteremia. This observation is most likely explained by the tendency for VRE BSI to have been prolonged in this cohort of patients. The median duration of VRE bacteremia was 4.5 days, with a range of <1–30 days. Given the relatively low virulence of the organism [27], life-threatening effects may not manifest until many days after onset of prolonged bacteremia. Although studies of causation conventionally exclude variables that might be in the causal pathway, we performed an additional multivariate analysis that controlled for the duration of bacteremia, to better explain why VRE infection was associated with mortality. The observation that the increased risk of death associated with vancomycin resistance disappeared in multivariate models that controlled for the duration of bacteremia suggests that a delay in clearing bacteremia may be an important intervening variable in the association between vancomycin resistance and mortality. The importance of the duration of bacteremia as a possible causal mechanism for increased mortality among patients with enterococcal BSIs is further supported by our finding of a dose-response relationship between the duration of bacteremia and risk of death. These data are consistent with a previous study that suggested that patients with VRE BSI and persistently positive blood cultures (defined as ⩾1 positive follow-up blood culture) are more likely to die [9].

The distribution of enterococcal species between patients with VRE and VSE BSI was different; Enterococcus faecium accounted for almost 90% of VRE BSIs but only #x007E;30% of VSE BSIs. Some in vitro studies have suggested that enterococcal virulence determinants (gelatinase, aggregation substance, cytolysin/ hemolysin, lipase, extracellular superoxide, and extracellular surface protein) are found more frequently in Enterococcus faecalis isolates than in E. faecium isolates [23, 28, 29]. On the other hand, some studies have reported that E. faecium is more often resistant to phagocytosis than in E. faecalis [25]. Whether there is a clinically relevant difference in virulence between VRE and VSE or between different enterococcal species is unknown. Previous investigators have postulated that interspecies differences may confound estimates of mortality risk for patients with enterococcal bacteremia [14]. However, we included enterococcal species in our multivariate models and found that species was not associated with mortality, whereas vancomycin resistance remained a significant independent risk factor for death.

The CDC definition of BSI that we used considers Enterococcus to be a recognized pathogen rather than a skin commensal; therefore, some of the BSIs in the cohort were included on the basis of a single positive blood culture [21]. It is possible that some of the cases represented skin contamination of the culture rather than true bacteremia. If such patients were more heavily distributed to the VSE group, the effect of vancomycin resistance on mortality may have been overestimated. However, our results do not appear to be biased in this way, because patients in whom BSI was defined by use of a single blood culture were evenly distributed among the VRE and VSE groups (1/22 and 7/61; P = .4). Furthermore, our findings were not significantly changed when we repeated the analyses using a more strict BSI definition that required the isolation of an enterococcal species from at least 2 blood cultures or the isolation of an enterococcal species from 1 blood culture plus the presence of fever, chills, or hypotension.

Invasive devices were removed much more frequently from patients with VRE BSI than from those with VSE BSI. We believe that this observation is explained by the fact that many patients with VRE BSI had prolonged bacteremia, which, in the clinical setting, usually dictates the removal of catheters and other devices.

As opposed to a previous investigation [10], we were unable to demonstrate an association between survival and the administration of an active antimicrobial agent. Many of our patients had prolonged durations of VRE bacteremia, despite early administration of an antimicrobial agent to which the bloodstream isolate was susceptible (usually chloramphenicol). Although there have been reports of successful treatment of VRE infection with chloramphenicol, no controlled trials to demonstrate its efficacy are available, and there are few studies that have specifically examined its efficacy in patients with neutropenia and in VRE bacteremia [30]. Two recently US Food and Drug Administration-approved antimicrobial agents with activity against VRE, linezolid and quinupristin/dalfopristin, and an investigational drug, oritavancin, were used in some of our patients with VRE BSI. Only 9 patients received 1 of these newer agents, 7 of whom died. However, none of these patients received the newer agents within 48 h of the onset of bacteremia, and they were used in most cases for patients who failed therapy with chloramphenicol and likely represented a group with very poor prognosis. Therefore, we were unable to assess the efficacy of these agents in the treatment of VRE bacteremia. The increased risk of mortality observed among patients with VRE BSI may have been reduced if newer antimicrobial agents were routinely used as part of the initial therapy. Nevertheless, the observed association between mortality and resistance highlighted in the present study has important implications for the future regarding this and other resistant organisms. It provides additional evidence that the emergence of antimicrobial resistance can have an important and deleterious impact on patient outcome. Although new drug development may provide a temporary solution to such problems, the longevity of any advantage offered by newer agents may be limited by the continued emergence of resistance. Resistance to quinupristin/dalfopristin and linezolid is already being reported [31–35], and there are few new drugs in development to replace them [36]. These observations suggest that the health-care community should continue to seek more effective ways of preventing the continued emergence of resistance, rather than relying solely on new drug development to address resistant pathogens as they emerge.

In conclusion, VRE bacteremia was associated with increased mortality in this cohort of patients with neutropenia and was likely secondary to prolonged duration of bacteremia. Further research is needed to evaluate the impact of newer antimicrobial agents.

Acknowledgements

We thank Mitchel Klein (Emory University); Holly Hill (Centers for Disease Control and Prevention [CDC]/Emory University); Jonathan Edwards and Juan Alonso-Echanove (CDC); and the Microbiology Laboratory, Infection Control Office, and Medical Records, Emory University Hospital.