Systematic Review of the Short-Term versus Long-Term Duration of Antibiotic Management for Neutropenic Fever in Patients with Cancer

Simple Summary Empirical administration of broad-spectrum antibiotics during neutropenia has been shown to reduce mortality from bacterial infections. However, prolonged antibiotic exposure, in particular, promotes the development of antimicrobial resistance and the selection of resistant microorganisms, which are often more difficult to treat, and carry a higher risk of complications. Early antibiotic discontinuation has been proposed in patients with hematologic malignancy who have febrile neutropenia. Several studies have found that shorter duration of antimicrobial therapy have better clinical outcomes and lower the exposure to the broad-spectrum antibiotics, but this raises concerns about their implementation in clinical practice. Furthermore, their safety and efficacy have been questioned. In our study, a systematic review was conducted to compare the short-term and long-term durations of antibiotics for febrile neutropenia for the outcomes of clinical failure, mortality, and bacteremia. Abstract Early antibiotic discontinuation has been proposed in patients with hematologic malignancy with fever of unknown origin during febrile neutropenia (FN). We intended to investigate the safety of early antibiotic discontinuation in FN. Two reviewers independently searched for articles from Embase, CENTRAL, and MEDLINE on 30 September 2022. The selection criteria were randomized control trials (RCTs) comparing short- and long-term durations for FN in cancer patients, and evaluating mortality, clinical failure, and bacteremia. Risk ratios (RRs) with 95% confidence intervals (CIs) were calculated. We identified eleven RCTs (comprising 1128 distinct patients with FN) from 1977 to 2022. A low certainty of evidence was observed, and no significant differences in mortality (RR 1.43, 95% CI, 0.81, 2.53, I2 = 0), clinical failure (RR 1.14, 95% CI, 0.86, 1.49, I2 = 25), or bacteremia (RR 1.32, 95% CI, 0.87, 2.01, I2 = 34) were identified, indicating that the efficacy of short-term treatment may not differ statistically from that of long-term treatment. Regarding patients with FN, our findings provide weak conclusions regarding the safety and efficacy of antimicrobial discontinuation prior to neutropenia resolution.


Introduction
Fever due to chemotherapy-induced neutropenia is experienced in 10-50% of patients with solid tumors, and more than 80% of those with hematologic malignancies [1]. Patients with hematological malignancy are at high risk of febrile neutropenia (FN), and experience Gram-negative bacilli bloodstream infections. Broad-spectrum beta-lactam antibiotics should be administered, such as carbapenem, piperacillin/tazobactam, ceftazidime, or cefepime, according to several guidelines [1][2][3]. Determining the optimal shown in the Supplementary Materials. This systematic review was conducted according to the guidelines of the Preferred Reporting Items for Systematic reviews and Meta-Analyses (PRISMA) [14]. The review protocol was recorded on 2 November 2022 with PROSPERO with the CRD number 42022369590.

Selection of Studies
We included randomized control trials (RCTs) in any language that reported all-cause mortality, clinical failure, or bacteremia, comparing the short-term duration of antibiotics with the long-term duration in hematological FN. We excluded patients with clinically and microbiologically documented infections, as well as neonatal patients. Two investigators (K.I. and T.M.) independently assessed the full texts of the articles. Discrepancies were discussed with a third and fourth investigator (E.O. and N.M.). Regarding the selection of studies until the end of 2017, we referred to the results of a Cochrane review by Stern et al. [13].
Adults (older than 18 years) and children (younger than 18 years) with FN caused by cancer chemotherapy and treated with any antibiotic regimen were included in this study. We defined fever as a single oral temperature higher than 38.3 • C or a temperature higher than 38.0 • C sustained for more than 1 h, according to the guidelines [1,2]. Neutropenia was defined as an absolute neutrophil count of less than 500 cells/µL. Studies that used a different although similar definition to that in the guidelines were included in the review. The types of interventions in the RCT define protocol-guided antibiotic discontinuation prior to neutropenia resolution versus antibiotic continuation until neutropenia resolution. We recorded the criteria defined for antibiotic discontinuation, including the timing of discontinuation, definitions of defervescence, and neutrophil count defined for neutropenia resolution.

Outcomes
The primary outcomes in this systematic review were any cause mortality, clinical failure, and bacteremia. Clinical failure was assessed as defined in each study.

Data Extraction
Two investigators (K.I. and T.M.) independently extracted the following data: publication country, published year, sample size, type of cancer (solid tumor or hematological malignancy, including stem cell transplantation), type of beta-lactam antibiotics, follow-up period, mortality, clinical failure, and bacteremia in each study. Data were extracted to the data extraction sheet using Microsoft Excel and Google spreadsheets, and were easily checked by a reviewer, including study information (e.g., publication country, study years, single-center or multi-center study), participant baseline characteristics (type of population, inclusion and exclusion criteria, comorbidity, and type of cancer), information regarding the intervention (type of antimicrobials and planned antibiotic duration in each arm), information regarding risk of bias (e.g., randomization method, allocation concealment, blinding, discontinuation of study, and incomplete outcome reporting), and information regarding outcomes (mortality, clinical failure, and bacteremia). Two review authors (K.I. and T.M.) extracted data from the included trials independently and entered them into the data extraction sheet. We extracted data preferentially using the intention-to-treat method, which included all individuals who were randomly assigned to the study outcome. For dichotomous outcomes, we recorded the number of participants manifesting the outcome in each group, as well as the number of evaluated participants. For continuous outcomes, we documented values, as well as the measure used, to represent the data (including mean with standard deviation and median with interquartile range). Discrepancies were resolved through discussion or by other investigators (E.O. and N.M.). We asked study authors for any missing data so that we could include findings from any studies published after 2018 in our study. We also referred to the data from the Stern et al. study [13], which was published before 2018.

Risk of Bias Assessment
Two investigators (K.I. and T.M.) independently assessed the risk of bias. Disagreements were resolved via discussion with a third investigator (E.O. and N.M.). The risk of bias was assessed according to the scales of the Cochrane risk-of-bias tool (RoB) [15]. With the RoB, we evaluated seven domains of bias: random sequence generation, allocation concealment, blinding of participants and personnel, blinding of outcome assessment, incomplete outcome data, reporting selection, and others. We assessed the effect of allocation concealment on results based on the evidence of a strong association between poor allocation concealment and overestimation of effect [16], as defined below: • Low risk of bias (adequate allocation concealment); • Unclear bias (uncertainty regarding allocation concealment); • High risk of bias (inadequate allocation concealment).
The two review authors independently recorded methods of allocation generation, blinding, incomplete outcome data, selective reporting, the unit of randomization (patient or febrile episode), and publication status, in addition to the adequacy of allocation concealment.

Statistical Analyses
We analyzed dichotomous data by calculating the RR for each study, with the uncertainty in each result presented as 95% CIs. We assessed the percentage of variation across studies that could not be ascribed to sampling variation using the I 2 statistic. A fixed-effects model was used unless significant heterogeneity was observed (p < 0.1 or I 2 > 50%), in which case the random-effects model was used. We also visually inspected the forest plots to judge heterogeneity. We analyzed the data using Review Manager 5.4 (freely available software, released by Cochrane, London, UK).

Certainty of Evidence
We used the Grades of Recommendations, Assessment, Development, and Evaluations (GRADE) approach to interpret the findings and rate the certainty of evidence [17], grading the major outcomes (mortality, clinical failure, and bacteremia development). A certainty of evidence of review was evaluated using GRADEpro guideline development tool software (GRADEpro GDT, Evidence Prime Inc., Hamilton, ON, Canada) [18], using parameters such as study design, risk of bias, directness of outcomes, heterogeneity, precision within results, bias due to publication, estimate effect, and dose relationship with response and confounders. Thus, the overall GRADE obtained can be high, moderate, low, or very low certainty of evidence. We considered this analysis in our conclusions.

Sensitivity Analysis
We conducted a sensitivity analysis to assess the effect of allocation concealment on mortality to prevent the overestimation of effects of studies with inadequate or unclear allocation concealment. The studies with unclear risk were same as those identified by Stern et al. [13]; therefore, we only analyzed low risk allocation.

Selection Bias
Funnel plot analyses were performed for the three main comparisons: mortality, clinical failure, and bacteremia. The funnel plots for three main comparisons were symmetrical ( Figure 2). An indication that small trials are missing may be present for bacteremia in Figure 2.

Selection Bias
Funnel plot analyses were performed for the three main comparisons: mortality, clinical failure, and bacteremia. The funnel plots for three main comparisons were symmetrical (Figure 2). An indication that small trials are missing may be present for bacteremia in Figure 2.

Risk of Bias Assessment, GRADE, and Meta-Analyses
In our systematic review, no significant differences in mortality (RR 1.43, 95% CI, 0.81, 2.53, I 2 = 0), clinical failure (RR 1.14, 95% CI, 0.86, 1.49, I 2 = 25), and bacteremia (RR 1.32, 95% CI, 0.87, 2.01, I 2 = 34) were observed (Figure 3). The risk of bias assessment data are graphically presented in Table 2. We also evaluated the GRADE for mortality, treatment failure, and bacteremia in the RCTs and found a low certainty of evidence (Table 3). We also analyzed the mortality, clinical failure, and bacteremia for only hematological malignancy patients, including patients who underwent stem cell transplantation as reported by de Jonge et al. [12], Ram et al. [25], Aguilar-Guisado et al. [19], and Santolaya et al. [27], and found similar results (Supplementary Figure S1). Figure 3. Summary of findings of short-compared with long-term duration antibiotic therapy presented as Forest plots, including the results reported by de Jonge et al. [12], Ram et al. [25], and Kumar et al. [22], and the results of the systematic review by Stern et al. [13]. Mortality (above), clinical failure (middle), and bacteremia (below). Total RR across 1 (left favors short, right favors long). Abbreviations: RCT, randomized control study, CI, confidence interval; RR, risk ratio [12,[19][20][21][22][23][24][25][26][27][28]. . Summary of findings of short-compared with long-term duration antibiotic therapy presented as Forest plots, including the results reported by de Jonge et al. [12], Ram et al. [25], and Kumar et al. [22], and the results of the systematic review by Stern et al. [13]. Mortality (above), clinical failure (middle), and bacteremia (below). Total RR across 1 (left favors short, right favors long). Abbreviations: RCT, randomized control study, CI, confidence interval; RR, risk ratio [12,[19][20][21][22][23][24][25][26][27][28]. Table 2. Summary of risk of bias in all the randomized controlled trials including the results reported by de Jonge et al. [12], Ram et al. [25], and Kumar et al. [22], and the results of the systematic review by Stern et al. [13]. The risk of bias included randomization sequence, concealment, blinding of participant and clinician, incomplete outcome data, selective reporting, and others. The color of risk of bias: green, low risk of bias; yellow, unclear risk of bias; red, high risk of bias.  ⊕⊕ Low a,b * The risk in the intervention group (and its 95% confidence interval) is based on the assumed risk in the comparison group and the relative effect of the intervention (and its 95% CI). CI: confidence interval; RR: risk ratio GRADE Working Group grades of evidence High certainty: we are very confident that the true effect lies close to that of the estimate of the effect. Moderate certainty: we are moderately confident in the effect estimate: the true effect is likely to be close to the estimate of the effect, but there is a possibility that it is substantially different. Low certainty: our confidence in the effect estimate is limited: the true effect may be substantially different from the estimate of the effect. Very low certainty: we have very little confidence in the effect estimate: the true effect is likely to be substantially different from the estimate of the effect.

Randomization
Explanations: a . methods of randomization and allocation concealment were unclear in most studies. Most studies were unblinded; b . effect estimate overlapping no effect with wide confidence interval. c . variable and inconsistent definition of clinical failure across studies.

Sensitivity Analysis
The RR for mortality was 1.07 (95% CI 0.39-2.92) in the studies with a low risk of bias for allocation concealment (five trials), compared with an RR of 1.65 (95% CI 0.82-3.29) in the studies with an unclear risk of bias for allocation concealment (p = 0.51 for subgroup differences; Supplementary Figure S2).

Discussion
In this systematic review, we examined the short-and long-term duration of antibiotic management for neutropenic fever in patients with hematological malignancy. We identified eleven RCTs (comprising 1128 distinct patients with FN). In our systematic review, we found that the efficacy of short-term duration of treatment may not differ statistically from that of long-term.
According to these studies, it may still be difficult for clinicians to implement a short duration of antibiotics for FN.
However, multidrug-resistant GNB, including carbapenem-resistant GNB and those caused by extended spectrum beta-lactamase-producing Enterobacteriaceae, is increasing worldwide in cancer patients [29]. Independent risk factors for CRE bloodstream infection in this study were prior β-lactam/β-lactamase inhibitor or carbapenem use. In another study, antimicrobial resistance was associated with unfavorable outcomes, such as high mortality in patients with cancer [30]. As a result, it is necessary to reduce long-term antibiotic exposure in cancer patients.
We are still debating whether to accept the result that the long-term treatment is preferable, due to the discrepancies among the studies. The use of prophylaxis and the criteria for antibiotic discontinuation were different in each study. Regarding the different characteristics of each study, much longer durations of antibiotic treatment were reported in the study by Aguilar-Guisado et al. [19]. The type of hematological malignancies studied by de Jonge et al. [12] and Aguilar-Guisado et al. [19] were disparate. In the study by Aguilar-Guisado et al. [19], 45% of patients had acute leukemia, and approximately 27% were in induction or re-induction for prolonged neutropenia; in the study by de Jonge et al. [12], 43% of patients had multiple myeloma and about 70% of transplants were autologous, in which the neutropenic duration was shorter.
In the former study, neutrophil recovery did not resume until the patients in the shortterm antibiotic group showed improved clinical symptoms; while in the latter, neutrophil recovery was early; therefore, rendering a comparison with the long-term group was difficult without a cut-off. Additionally, studies by Klaassen et al. [21] and Santolaya et al. [27] mixed the high-risk with low-risk patients. Thus, in the majority of the studies, the shortterm group had very limited time to confirm culture negativity, and the long-term group may have had the desirable outcome, although the two groups were non-inferior. Although the etiology of FN remains unknown with negative culture findings, recent cell-free deoxyribonucleic acid (DNA) technology has shown that viruses and Streptococcus viridans are common in blood culture-negative cases [31]. We believe that if cell-free DNA is incorporated into the studies, patients with infections would be excluded from the short-term treatment group, thus, leading to more favorable outcomes.
In the context of antimicrobial regimens, de Jonge et al. [12] empirically used carbapenems, while Aguilar-Guisado et al. [19] used anti-pseudomonas beta-lactam antibodies. The regimens of other studies in the review were different from the regimen in the current guideline. However, the incidence of resistant Gram-negative-rod (GNR) strains has increased in patients with hematologic malignancies [32], but the studies included in the systematic review did not consider the ration of resistant strains. The efficacy of extended administration of beta-lactam [33] or beta-lactam + aminoglycosides [34] for FN is under investigation. These regimens are also effective against resistant GNR strains. Moreover, the small sample size of these studies also resulted in limited evidence. However, there are several ongoing RCT studies on the discontinuation of antimicrobial therapy for FN in some countries; therefore, the results are still awaited (NCT 04948463, NCT 04270786, and NCT 04637464 in the ClinicalTrials.gov registry of clinical trials).
This study had several limitations. A systematic review examines and synthesizes the information on a subject that is available in the literature; as a result, it may include some bias from the publications. We compiled trials of different designs, including RCTs and prospective non-RCTs. The study by Stern et al. [13] is supplemented by new RCTs in our study, and the authors have already communicated via email with the correspond-ing authors in each of the previous studies in order to learn more specifics about each investigation. We are only able to communicate with corresponding authors in each article since 2018. Although the heterogeneity of studies in our research was low, we believe that more RCTs will further improve the quality of a systematic review. Currently, several retrospective studies have been conducted on the duration of therapy for FN, which were preferable for a short course of antibiotics [35][36][37][38]. Moreover, these studies included highrisk febrile neutropenic patients in hematology. These studies support the findings of our systematic review.

Conclusions
Cancer patients should be exposed to the optimal short exposure duration of antimicrobial therapy, which benefits the implementation of antimicrobial stewardship strategies to improve the use of antimicrobials and limit multidrug resistance, as well as a short hospital stay. The evidence of each RCT is limited, and a short-term duration of beta-lactam antibiotics showed no statistically significant differences in mortality, clinical failure, and bacteremia compared with those for long-term duration antibiotics in our systematic review, possibly owing to the small number of studies, varying clinical among studies, or different study designs.
Supplementary Materials: The following supporting information can be downloaded at: https://www.mdpi.com/article/10.3390/cancers15051611/s1, Figure S1: Summary of findings of short-term duration of antibiotics compared with long-term duration of antibiotics presenting Forest plot for only hematological malignancy patients including stem cell transplantation; Figure S2: Summary of findings of short-term duration of antibiotics compared with long-term duration of antibiotics presenting Forest plot for mortality in allocation concealment bias as sensitivity analysis.