Introduction

Cancer patients with chemotherapy-induced neutropenia are highly susceptible to bacterial infection [1], and Gram-negative bacilli (GNB) bacteremia still causes high mortality and morbidity in neutropenic patients. Identifying patients at high risk for severe bacterial infection at the onset of febrile neutropenia would be useful, especially since clinical signs of infection in these patients are often minimal [2, 3]. Patients with a high probability for a severe GNB infection could receive early therapy.

The acute-phase protein lipopolysaccharide (LPS) binding protein (LBP) plays an important role in innate host defense. Mice lacking the gene encoding for LBP demonstrate an impaired immune response, especially to Gram-negative infection [4]. LPS, a component of the outer cell wall of GNB, known as the principal activator of host response, is bound to LBP and transported to soluble CD14 or CD14-expressing cells. Recently it was demonstrated that the LPS-LBP-CD14 complex binds to the Toll-like receptor 2 and 4 on effector cells, leading to production of inflammatory cytokines [5, 6]. In addition, LBP was found to neutralize endotoxin by enhancing incorporation of LPS in serum lipoproteins [7]. Both these processes seem to be essential for an adequate innate host response to GNB. Whether LBP has an active role in Gram-positive infection still needs to be elucidated, although it has been shown that plasma LBP levels are elevated during both Gram-negative and Gram-positive bacterial infection as well as other inflammatory diseases such as pancreatitis and the acute respiratory distress syndrome [8, 9].

The aim of this study was to investigate whether plasma LBP levels increase during bacterial infections in neutropenic cancer patients, and whether determination of plasma LBP level at the onset of febrile neutropenia can be used as an early diagnostic parameter for the presence and/or severity of a Gram-negative infection.

Materials and methods

This prospective study was performed at the Departments of Pediatric Oncology and Internal Medicine of the University Hospital in Groningen between March 1998 and May 2000. The medical ethics committee approved the study protocol, and all patients gave written informed consent. To evaluate whether neutropenic patients respond to bacterial infection by increasing LBP production, serial LBP measurements during chemotherapy-induced neutropenia (defined as granulocytes <0.5×109/l or leukocytes <1.0×109/l) were performed three times weekly in two patients. To study the potential diagnostic role of LBP at the onset of febrile neutropenia, patients with fever (defined as >38.5°C once or twice >38.0°C during an observation period of 6 h) and neutropenia were considered eligible. Patients already receiving antibiotics were excluded. The use of selective gut decontamination or Pneumocystis carinii pneumonia prophylaxis was not an exclusion criterion. Clinical sepsis was defined by the following criteria: (a) systolic blood pressure below 90 mmHg in adults or less than 2 SD decreased for age in children or (b) both heart rate less than 90/min and breath rate higher than 20/min in adults or both more than 2 SD elevated for age in children according to the criteria for systemic inflammatory response syndrome [10]. Patients were divided into four groups: those with GNB bacteremia, Gram-positive bacteremia, clinical sepsis, or fever of unknown origin (FUO). There were no significant differences between the four groups in age, sex or leukocyte count.

Blood samples for LBP and CRP determination and blood cultures were collected at presentation before antibiotics were started. Plasma was separated from the cells within 30 min after collection and stored at −80°C. Plasma LBP concentrations were determined using a chemiluminescence immunoassay (Diagnostic Products, Los Angeles, Calif., USA); reference values vary between 2 and 15.2 mg/l according to the manufacturer. Serum CRP concentrations were measured on a nephelometer (Dade-Behring, Amersfoort, The Netherlands); reference values are below 10 mg/l.

Data were analyzed using SPSS for Windows. Median LBP and CRP levels were compared by the exact Mann-Whitney test. Spearman’s correlation coefficient was used to assess relationships between LBP and CRP levels and betweenrespectively LBP level and age. To evaluate the diagnostic test properties of LBP we determined the sensitivity, specificity, and predictive values with the corresponding 95% confidence intervals (CI, calculated by the exact method). Since patients could be enrolled several times, we analyzed both the complete data set and the data restricted to the initial episode of each patient. Unless otherwise stated, only the data of the complete set are presented. A two-tailed p value less than 0.05 was considered statistically significant.

Results

Serial measurements of plasma LBP levels during a chemotherapy-induced neutropenic episode in two patients with acute myeloid leukemia are shown Fig. 1. LBP levels rose during bacterial infection (documented Gram-positive and clinically defined infection, respectively), confirming that neutropenic patients are able to mount a LBP response during bacterial infection. Subsequently to evaluate the diagnostic value of one LBP level at the onset of febrile neutropenic episodes in cancer patients plasma LBP concentrations were determined in 66 episodes in 57 patients. Three patients were studied twice, and three others studied during three episodes. Patients characteristics are shown in Table 1. Eighteen (27%) episodes were bacteremic, four with GNB (one Escherichia coli, one Klebsiella pneumoniae, two Pseudomonas aeruginosa; GNB bacteremia group) and 14 with a Gram-positive organism (eight coagulase-negative staphylococci, six streptococci; Gram-positive bacteremia group). Clinical sepsis was recorded in 8 (12%) episodes (four in the GNB bacteremia group, two in the Gram-positive bacteremia group, and two in the clinical sepsis group). No bacterial origin was detected in 46 (70%) episodes (FUO group). Of the nine next (second or third) episodes in the same patients seven were in the FUO group and two in the Gram-positive bacteremia group groups.

Fig. 1
figure 1

Plasma Lipopolysaccharide-binding protein (LBP) and C-reactive protein (CRP) levels during neutropenic periods in two cancer patients following chemotherapy (ended on day 0): A: Documented bacterial infection (neuropenia from day 4, developement of fever at day 11, Streptococcus gr. B in blood culture at day 13), B: Clinical infection (neutropenia from day 9, development of fever at day 16, day 17 until 34 clinical signs of bacterial infection)

Full size image
Table 1 Patient characteristics (brackets no. of episodes)
Full size table

Median CRP levels at the onset of febrile neutropenia in episodes with GNB bacteremia did not differ significantly from than those in the FUO group (185 mg/l, range 20–213, vs. 52 mg/l, 4–284; p=0.10) or those in the Gram-positive bacteremia (122.5 mg/l, 10–357; p=0.65). Similar results were found when considering initial episodes only (respectively, p=0.08 and p=0.77). In contrast, median LBP levels were significantly higher in episodes with GNB bacteremia (54.2 mg/l, range 47–72) than in patients with Gram-positive bacteremia (21.1 mg/l, 10.9–49.2; p=0.003) and those with FUO (21.2 mg/l, 0.5–60.0; p<0.0005; Fig. 2). Again, this was found in only the initial febrile neutropenic episodes as well (respectively, p=0.004 and p=0.001). In addition, no correlation was found between LBP levels and age. A significant correlation was found between plasma LBP levels and CRP levels at the onset of febrile neutropenia (r=0.59, p<0.0005).

Fig. 2
figure 2

Plasma concentrations of Lipopolysaccharide-binding protein (LBP) (A) and C-reactive protein (CRP) (B) in neutropenic cancer patients with fever of unknown origin (FUO; n=46), clinical sepsis (n=2), gram-positive bacteremia (n=14) and gram-negative bacteremia (n=4)

Full size image

To identify patients with GNB bacteremia reliably a cutoff value for LBP should be 100% sensitive. Choosing the best cutoff value of 46.3 mg/l, sensitivity, specificity, and positive and negative predictive values were 100% (CI 47–100%), 92% (CI 82–98%), 44% (CI 18–79%), and 100% (CI 94–100%), respectively. For CRP the best cutoff value on the basis of our results was 19.5 mg/l, with sensitivity for GNB bacteremia of 100% (CI 47–100%). However, the specificity was only 19% (CI 12–32%). Positive and negative predictive values were 7% (CI 2–18%) and 100% (CI 77–100%), respectively. These results indicate that a LBP level higher than 46 mg/l at the onset of febrile neutropenia predicts GNB bacteremia more accurately than the CRP level. Results of the analysis of initial episodes were comparable.

Discussion

Onset of fever in neutropenic patients often heralds a severe bacterial infection, especially one caused by GNB bacteremia. However, clinical signs of a bacterial infection are often limited because of neutropenia. Therefore an early diagnostic marker of the presence and/or severity of a GNB infection at the onset of a febrile episode would be very useful. Several groups have investigated the use of clinical features in risk assessment models [2]. Plasma concentrations of acute-phase proteins and cytokines have also been proposed as predictive diagnostic tools in febrile neutropenia. The diagnostic value of CRP has been extensively studied. Serial measurements of CRP levels can be helpful to monitor the response to antibiotic therapy, but an initial CRP level was found not to be a useful predictor of bacterial infection in febrile neutropenia [11]. Likewise, procalcitonin was not found an adequate marker for bacterial infection in febrile neutropenic patients, most likely because neutropenic patients are unable to produce procalcitonin adequately [12, 13]. More recently several groups have shown that plasma levels of interleukins 8 and 6 are significantly higher in Gram-negative and in Gram-positive infections than in cases of FUO [14, 15].

Using serial measurements of plasma LBP we found that LBP levels increase during bacterial infection in neutropenic cancer patients. In addition, LBP was measured in a large group of patients at the onset of febrile neutropenia. LBP levels at the onset of febrile neutropenia due to GNB bacteremia were significantly higher than in patients with FUO or Gram-positive bacteremia. This difference in LBP levels between Gram-negative and Gram-positive infection may be explained by a more pronounced acute-phase response in patients with Gram-negative than Gram-positive bacteremia, resulting in faster and higher LBP production at the onset of the infection. The recognized role of LBP in the host response to GNB, which acts as a transport protein for LPS, may explain in large part our findings whereas in Gram-positive infection no function for LBP has yet been found.

Our results indicate that a high LBP level at the onset of febrile neutropenia can predict Gram-negative bacteremia. Due to the small numbers in this study these results need be confirmed in larger series. In view of the increased tendency for treating a subset of febrile neutropenic patients with oral antibiotics in an outpatient setting it is even more important to identify patients with Gram-negative bacteremia in an early stage [16, 17]. The ability to identify patients with a high probability of GNB bacteremia at the onset of febrile episode may enable physicians to anticipate clinical deterioration and to improve early management.