A Randomized, Open-Label, Multicenter Comparative Study of the Efﬁcacy and Safety of Piperacillin-Tazobactam and Cefepime for the Empirical Treatment of Febrile Neutropenic Episodes in Patients with Hematologic Malignancies

Background. The empirical treatment of febrile, neutropenic patients with cancer requires antibacterial regimens active against both gram-positive and gram-negative pathogens. This study was performed to demonstrate the noninferiority of monotherapy with piperacillin-tazobactam, compared with cefepime. Methods. We conducted a randomized-controlled, open-label, multicenter clinical trial among high-risk patients from 34 university-afﬁliated tertiary care medical centers in the United States, Canada, and Australia who were undergoing treatment for leukemia or hematopoietic stem cell transplantation and were hospitalized for empirical treatment of febrile neutropenic episodes. Patients received piperacillin-tazobactam (4.5 g every 6 h) or cefepime (2 g every 8 h) intravenously. The primary outcome was success (deﬁned by defervescence without treatment modiﬁcation) at 72 h of treatment, end of treatment, and test of cure in the modiﬁed intent-to-treat analysis. Secondary outcomes included time to defervescence, microbiological efﬁcacy, the additional use of glycopeptide antibiotics, emergence of resistant bacteria, and safety. test-of-cure piperacillin- P p .11 tazobactam at all time points ( ). Treatment with piperacillin-tazobactam was independently associated with analysis ratio, 1.65; 95% conﬁdence interval, 1.04–2.64; ). P p .035 Both regimens tolerated.

. Randomization scheme with the disposition of the modified intent-to-treat and clinically evaluable patient populations. More cefepime patients discontinued taking the study agent than piperacillin-tazobactam patients ( ). TOC, test-of-cure. 2 x p 2.815 P p .093 therapy for the management of high-risk febrile neutropenic patients with cancer. Recommendations for combination therapy based on an extended spectrum antipseudomonal b-lactam agent plus an aminoglycoside remain prominent in the American [2,4] and German [3]-but not in the Spanish [1]guidelines. Despite this, the weight of clinical trial experience supports the adoption of monotherapy as the standard of treatment of febrile neutropenic patients with cancer [5,6].

Patients.
Eligible subjects included those patients at high risk for medical complications [38][39][40] who were у18 years old, severely neutropenic (defined as having an absolute neutrophil count [ANC] of ! cells/L or a count of ! 9 9 0.5 ϫ 10 1.0 ϫ 10 cells/L with the expectation that it will decrease to ! 9 0.5 ϫ 10 cells/L after cytotoxic therapy), hospitalized for the management of a febrile episode complicating the course of cytotoxic therapy for a hematological malignancy or for a hematopoietic stem cell transplant (HSCT), and who had provided written, informed consent according to institutional protocol. The protocol was approved through the ethics review process of each Downloaded from https://academic.oup.com/cid/article-abstract/43/4/447/388596 by guest on 16 March 2020 of the participating university-affiliated institutions. Exclusion criteria included a history of hypersensitivity to b-lactam antibacterial agents, evidence of hepatic dysfunction (defined as a total serum bilirubin level 13 times the upper limit of the normal range or as a serum transaminase level 15 times the upper limit of the normal range), severe renal insufficiency requiring dialysis, or a positive test result for HIV antibodies.
Antibacterial regimens. Patients were randomly allocated to receive piperacillin-tazobactam (Zosyn, Wyeth Ayerst; 4.5 g intravenously every 6 h) or cefepime (Maxipime, Bristol-Myers Squibb; 2 g intravenously every 8 h). Regimens were assigned using a centralized randomization procedure in a 1: 1 ratio and were stratified according to receipt of prophylactic antibacterial agents. Antibacterial prophylaxis was discontinued at the time of study entry. Study regimens were administered for up to 21 days and were modified at the investigator's discretion. Adherence to published management guidelines was encouraged [41].
Study design. The study was conducted as an open-label, randomized-controlled, multicenter, noninferiority clinical trial. Calculation of the sample size was based on the assumption that the study agents were equally effective. Assuming an evaluable rate of at least 50%, ∼528 patients were to be enrolled to obtain 264 evaluable patients. If the 2 treatments were equally effective, with success rates of 50% at the test-of-cure visit, ∼132 clinically evaluable patients per treatment group would be required.
Definitions. The primary outcome was treatment success without regimen modification assessed at 72 h of therapy, end of therapy, and the test-of-cure review among the modified intent-to-treat population (patients receiving at least 1 dose of the study agents), in accordance with recommendations [42][43][44]. Treatment success was assessed at the test-of-cure assessment (defined as at least 7 days posttreatment). Fever and febrile episodes were defined and classified in accordance with current guidelines [2,45]. Defervescence was defined as a reduction in body temperature to !38ЊC when measured orally, sustained over at least 48 h. Treatment success was defined as resolution of all signs and symptoms of infection without modification of the initial empirical antibacterial regimen. Initial treatment response but with a modified regimen was defined as recrudescence of fever within 48 h of defervescence because of a viral, fungal, or parasitic pathogen outside the spectrum of activity of the antibacterial study agents that required regimen modification. Treatment failure was defined as death due to infection or the administration of any additional antibacterial agent for persistent fever, lack of improvement, progressive infection, or new bacterial infection. Indeterminate status was defined as loss to follow-up; an instance in which the etiology of the initial febrile episode was determined to be due to a viral, fungal, parasitic, or mycobacterial pathogen; or an in-stance in which the patient received concomitant antibacterial therapy for reasons other than treatment failure. Statistical analysis. The primary objective was to demonstrate piperacillin-tazobactam to be noninferior to cefepime in the modified intent-to-treat analysis. One-sided 95% CIs (corrected for continuity) were calculated for the difference in treatment success to evaluate noninferiority between treatment groups. Noninferiority was concluded if the lower bound of the 95% CI for the difference in success was уϪ0.20. Secondary outcome analyses included microbiological success; treatment success by prestudy use of antibacterial prophylaxis, by classification of infection, or by neutrophil recovery profile; time (in days) to defervescence; emergence of vancomycin-resistant enterococci or extended-spectrum b-lactamase-producing gram-negative bacteria in rectal surveillance cultures; and the emergence of Clostridium difficile-associated diarrhea. Success rates with 95% CIs were compared between treatment groups using the Cochrane-Mantel-Haenszel statistic stratified across study centers. Time-to-event analyses were conducted using Kaplan-Meier product limit estimates. Differences by allocation were examined using the log-rank test.
Logistic regression was used to assess the influence of potential prognostic factors on treatment success. In univariate analysis, variables associated with treatment success, including demographic covariates, such as age and country of residence ( ), were entered into a stepwise multivariate logistic P р .05 regression model. Those variables with a P value р.15 were subsequently entered into the final model. The final model was developed with variables ( ) retained from the first model. P р .1 Similarly, variables associated with defervescence and with prolonged neutropenia were entered into multivariate Cox regression models in order to identify predictors. Such variables included age (у65 years vs. !65 years), sex, underlying diagnoses (acute leukemia, lymphoma, or other), treatment (HSCT vs. chemotherapy), characteristics of the HSCT (autologous vs. allogeneic; stem cell source [peripheral blood vs. bone marrow]), use of hematopoietic growth factors either prior to or concomitantly with the initiation of empirical antibacterial therapy, neutrophil recovery profile from baseline to end of treatment (ANC ! cells/L at baseline to 1 Safety was evaluated by analyzing adverse events, vital signs,     clinical laboratory results, and physical examination. Treatment-related adverse events were classified by body system and by allocation. Adverse events with an overall incidence of у3% were compared using Fisher's exact test.

RESULTS
A total of 528 eligible subjects were enrolled from 34 institutions in the United States, Canada, and Australia ( figure 1 and table  1). Of these, 265 subjects were randomized to receive at least 1 dose of piperacillin-tazobactam, and 263 subjects were randomized to receive at least 1 dose of cefepime over a mean (‫ע‬SD) of 9 ‫)5ע ( x p 4.865 P p .027 sponse rate for clinically documented infections was marginally higher in the piperacillin-tazobactam group (25.9%) than in the cefepime group (7.4%; ; ). At each of 2 x p 3.333 P p .068 the response assessment time points of 72 h, end of treatment, and the test-of-cure visit, piperacillin-tazobactam was noninferior to cefepime (table 3). Superiority was noted for piperacillin-tazobactam at the 72-h assessment. Table 4 shows the results of the logistic regression analyses of covariates predictive of treatment success. Allocation to piperacillin-tazobactam predicted treatment success (OR, 1.65; 95% CI, 1.04-2.64; ); however, having an indwelling P p .034 central venous catheter, experiencing failure of neutrophil recovery, having documented infection, and undergoing allogeneic HSCT were associated with poorer outcomes. Treatment success was almost one-half as likely among subjects for whom hematopoietic growth factors were initiated with the empirical antibacterial therapy.
Overall, defervescence occurred at a median of 7 and 10 days for the piperacillin-tazobactam and cefepime groups, respectively ( ; figure 2A). Among patients classified as ex-P p .1058 periencing treatment success, defervescence at 5 days was observed in each study group ( ; figure 2B), versus 9 P p .9649 days and 14 days among subjects in the piperacillin-tazobactam and cefepime groups, classified as experiencing treatment failure ( ; figure 2C). P p .0202 Cox proportional hazard models were used to examine factors associated with time to defervescence. Treatment allocation was included as a forced variable. Glycopeptide antibiotic treatment modification (with vancomycin or teicoplanin) (hazard ratio [HR], 0.71; 95% CI, 0.58-0.87; ) and study en-P p .001 rollment in the United States (HR, 0.54; 95% CI, 0.42-0.70; ) were less likely to be associated with earlier defer-P ! .0001 vescence, and a diagnosis of lymphoma (HR, 1.25; 95% CI 1.01-1.54; ) and receipt of piperacillin-tazobactam P p .0452 (HR, 1.24; 95% CI, 1.02-1.51; ) were more likely to P p .0332 be associated with earlier defervescence.

DISCUSSION
This study achieved its primary objective of demonstrating noninferiority of piperacillin-tazobactam, compared with cefepime, for the empirical treatment of high-risk febrile neutropenic patients with cancer. Furthermore, randomization to receive piperacillin-tazobactam was an independent predictor of treatment success and defervescence in multivariate analyses, which is consistent with the results of recent systematic reviews [46].
The multivariate analyses identified a treatment effect for piperacillin-tazobactam that was not observed in the primary analysis, but was consistent with previous reports from the Figure 2. A, Time to defervescence for all patients in the modified intent-to-treat analysis. Median times were 7 days and 10 days for the piperacillin-tazobactam and cefepime groups, respectively ( ). B, P p .1058 Time to defervescence for modified intent-to-treat patients classified as experiencing treatment success without modification. Median times were 5 days in both groups ( ). C, Time to defervescence for modified P p .9649 intent-to-treat patients classified as experiencing treatment failure. The median times were 9 days and 14 days for the piperacillin-tazobactam and cefepime groups, respectively ( ). P p .0202 European Organization for Research and Treatment of Cancer [47]. This was likely obscured in our study by other variables. Furthermore, our analyses confirmed previously reported relationships between outcome and neutrophil recovery, presence of indwelling central venous catheters, underlying diagnosis, and HSCT. The analyses also identified some unexpected relationships. First, the parallel use of hematopoietic growth factors with empirical antibacterial therapy was associated with a lower likelihood of treatment success (OR, 0.53; 95% CI, 0.31-0.91; ). Although this may represent P ! .02 a selection for a subgroup of subjects with more-severe infection, it may also be a surrogate, given our study design, for physicians more inclined to prescribe additional interventions in the circumstance of persistent fever, despite current guidelines [48][49][50][51]. Second, we noted that US subjects had a lower likelihood of earlier defervescence (HR, 0.65; 95% CI, 0.53-0.78; ), which suggests that either there may P ! .0001 have been more patients at higher risk of nonresponse enrolled in US centers or, alternatively, US physicians may have been more likely to initiate salvage treatments for perceived nonresponse due to persistent fever.
Gram-positive bacteria was the predominant cause of microbiologically documented infections, causing two-thirds of the bloodstream infections. Piperacillin-tazobactam and cefepime both have a spectrum of activity that is well suited for the management of patients at risk for gram-positive infections. Despite this, we noted that second-line administration of glycopeptides for persistent fever was a common phenomenon with a regional time-to-prescription effect. Canadian investigators tended to prescribe glycopeptides later than either their US or Australian counterparts, suggesting a differential tolerance for persistent fever. Moreover, more than three-quarters of the subjects who experienced treatment failure received glycopeptides as empirical second-line therapy after a median of only 3 days of initial therapy, whereas defervescence among responders was observed after a median of 5 days of initial therapy, suggesting that modification prior to 5 days may have been unnecessary in many patients, as reported previously [37,52,53]. Furthermore, we noted a relationship between prolonged neutropenia and glycopeptide administration not heretofore reported, except in anecdotal reports [37,[54][55][56][57][58]. These and other observations [14,59,60] suggest that in the interest of reducing excess glycopeptide-related adverse events and emergence of resistant pathogens, continued observation of empirical piperacillin-tazobactam monotherapy recipients without regimen modification in hemodynamically stable patients with persistent but unexplained fever for 15 days should be considered.
We noted a differential median time to response of 9 days in the piperacillin-tazobactam group and 14 days in the cefepime group among patients classified as experiencing treatment

failure (
). The distributions of second-line antibiotic P p .02 modifications, breakthrough infections, or empirical antifungal therapy by allocation did not account for this difference. Whether this represents a function of a heretofore undetected delayed primary treatment effect cannot be discerned from the present data; however, this observation, to our knowledge, appears to be unique and should be explored further in future studies.
The study regimens were well tolerated in both arms of the trial. Rashes and diarrhea were reported relatively frequently, 26% and 33%, respectively, compared with other trials [34,47] and may have been a function of the high proportion of patients (53%) who underwent HSCT. Consistent with previous studies [61][62][63], we noted a reduced risk for C. difficile-associated diarrhea among piperacillin-tazobactam recipients (OR, 0.32; 95% CI, 0.12-0.81), an observation with important economic implications [64]. Although piperacillin-related myelosuppression has been reported previously [65,66], the prolonged neu-tropenia observed among piperacillin-tazobactam recipients in our study was associated with other factors, including underlying disease and use of other potentially myelosuppressive agents, such as glycopeptides [54][55][56][57][58]. The all-cause mortality among cefepime recipients was nonsignificantly higher in our study (3% in the piperacillin-tazobactam group vs. 5.7% in the cefepime group), a finding that is in keeping with a recent systematic review [46]. This study has several limitations. First, the study design would have been enhanced by double-blinding. The differing schedules of drug administration, however, made this logistically difficult. Second, like previous studies [67], early modification of the primary empirical antibacterial regimen with a glycopeptide was at the discretion of the investigator and may have impaired our ability to recognize a treatment effect. Third, our rates of treatment success were low but consistent with previous reports, which reflects higher-risk patient populations with similar rigorous definitions of treatment success and fail-ure [34,37,47]. Fourth, a treatment success difference of 20% between the 2 treatment regimen arms may be too large to be clinically acceptable. Because the piperacillin-tazobactam arm achieved response rates that were as good as or, in some instances, better than the cefepime arm, this a priori requirement had no impact on the analysis. Finally, the advantages for piperacillin-tazobactam over cefepime were observed only after controlling for other variables in the multivariate analyses, which suggests that the entry criteria permitted enrollment of a study population too heterogeneous to detect the treatment effect in the primary analysis. This not withstanding, the study achieved its primary objective of noninferiority.
This trial, taken together with other large clinical trials, firmly establishes the safety and efficacy of piperacillin-tazobactam monotherapy for the empirical treatment of the febrile neutropenic patients with cancer. Based on our observations and in consideration of the Spanish [1] and German [3] guidelines, it seems appropriate to revise the North American guidelines [2,4] to include piperacillin-tazobactam as an acceptable monotherapeutic option in the higher-risk febrile neutropenic population.