Network meta-analysis of triazole, polyene, and echinocandin antifungal agents in invasive fungal infection prophylaxis in patients with hematological malignancies

Triazole, polyene, and echinocandin antifungal agents are extensively used to treat invasive fungal infections (IFIs); however, the optimal prophylaxis option is not clear. This study aimed to determine the optimal agent against IFIs for patients with hematological malignancies. Randomized controlled trials (RCTs) comparing the effectiveness of triazole, polyene, and echinocandin antifungal agents with each other or placebo for IFIs in patients with hematological malignancies were searched. This Bayesian network meta-analysis was performed for all agents. The network meta-analyses showed that all triazoles, amphotericin B, and caspofungin, but not micafungin, reduced IFIs. Posaconazole was superior to fluconazole [odds ratio (OR), 0.30; 95% credible interval (CrI), 0.12–0.60], itraconazole (OR, 0.40; 95% CrI, 0.15–0.85), and amphotericin B (OR, 4.97; 95% CrI, 1.73–11.35). It also reduced all-cause mortality compared with fluconazole (OR, 0.35; 95% CrI, 0.08–0.96) and itraconazole (OR, 0.33; 95% CrI, 0.07–0.94), and reduced the risk of adverse events compared with fluconazole (OR, 0.02; 95% CrI, 0.00–0.03), itraconazole (OR, 0.01; 95% CrI, 0.00–0.02), posaconazole (OR, 0.02; 95% CrI, 0.00–0.03), voriconazole (OR, 0.005; 95% CrI, 0.00 to 0.01), amphotericin B (OR, 0.004; 95% CrI, 0.00–0.01), and caspofungin (OR, 0.05; 95% CrI, 0.00–0.42) despite no significant difference in the need for empirical treatment and the proportion of successful treatment. Posaconazole might be an optimal prophylaxis agent because it reduced IFIs, all-cause mortality, and adverse events, despite no difference in the need for empirical treatment and the proportion of successful treatment.


Background
Adult patients who were diagnosed with hematological malignancies, such as acute lymphoblastic leukemia [1], acute myeloid leukemia [2], or myelodysplastic syndrome, and then instructed to receive intensive chemotherapy for remission or hematopoietic stem cell transplantation (HSCT) were at high risk of developing invasive fungal infections (IFIs) [3,4], especially Aspergillusand Candida-related IFIs [5,6]. IFIs contribute a lot to the morbidity and mortality in patients with hematological malignancies [4,7] because the symptoms and signs are absent or nonspecific in the early stage [8,9]. Thus, antifungal prophylaxis remains central to the containment of IFIs, making the early identification of IFIs difficult [2,10,11].
Triazole, polyene, and echinocandin antifungal agents have been extensively applied to prevent and treat IFIs [3]. A large number of clinical trials have been performed to investigate the role of antifungal prophylaxis against IFIs [1,2,[12][13][14][15]. Meanwhile, several metaanalyses have been performed to investigate the comparative efficacy and safety of the treatments [3,4,16,17]. However, the previous meta-analyses were limited by some drawbacks such as insufficient number of eligible studies and treatments. Thus, which treatments should be preferably prescribed to patients who were at high risk of IFIs remained unclear.
The present Bayesian network meta-analysis combined direct and indirect evidence comparing the relative efficacy of all antifungal prophylaxis regimes to determine the optimal agents against IFIs among high-risk patients.

Methods
This systematic review and network meta-analysis was performed according to the methodology framework recommended by the Cochrane Collaboration, and all summarized results were reported in accordance with the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) statement [18] (Supplementary file 1) and the International Society for Pharmacoeconomics and Outcomes Research Task Force on Indirect Treatment Comparisons Good Research Practices [19]. A formal protocol was not developed for this study.

Study identification
A systematic search of the databases PubMed, Cochrane Central Register of Controlled Trials, and Embase was conducted to capture all potential studies evaluating the prophylactic use of triazole, polyene, and echinocandin antifungal agents from their inception to April 2020. Each search strategy was modified depending on the specific requirements of the individual database under the assistance of a senior investigator. The reference lists of all eligible studies and topic-related reviews and the clinicaltrials.gov were also searched to include additional studies. The details of all search strategies for the three targeted databases after completing the electronic search are shown in Supplementary file 2. Any disagreement in study identification was resolved by consensus.

Study selection
Two investigators (Jie Wu and Jing Lai) were assigned to finish the study selection in the following three steps: (a) first, all duplicate records were eliminated using the Duplicates Elimination function of EndNote software; (b) the relevance of each record was evaluated by reviewing title and abstract; and (c) the eligibility of the remaining studies was checked by reviewing the full text eventually. Any divergence in study selection was resolved by consensus. Inclusion and exclusion criteria were developed to guide the study selection. The inclusion criteria were as follows: (a) adult patients with hematological malignancies receiving intensive chemotherapy for remission or HSCT; (b) randomized controlled trials (RCTs) comparing triazole, polyene, and echinocandin antifungal agents with placebo or with each other as prophylaxis against IFIs; (c) the overall incidence of proven or probable IFIs defined as the primary outcome, while the incidence of invasive Aspergillus and Candida infection, all-cause and IFI-related mortality, overall incidence of adverse events, withdrawal due to adverse events, need for empirical treatments, and proportion of successful treatment regarded as secondary outcomes; and (d) only studies published in English language.
A study was excluded if at least one of the following criteria was met: (a) studies without sufficient data and additional information not added through contacting the lead author and (b) duplicate study with relatively insufficient data.

Data extraction
Two investigators (Bing Yu and Bo Wang) independently extracted the following information, name of the first author, publication year, study design (multicenter and single center), country of the corresponding author, basic characteristics of participants (sample size, age, and sex ratio), details of treatments, follow-up time, outcomes, and details of the risk of bias. Any divergence in data extraction was resolved by consensus.

Quality assessment
The quality of eligible studies was assessed with the Cochrane risk-of-bias assessment tool [20] based on the random sequence generation; allocation concealment; blinding of participants and personnel; blinding of outcome assessment; incomplete outcome data; selective reporting; and other bias, which were performed by two independent investigators (Jie Wu and Jing Lai). A study was labeled as a low risk of bias if all items of the assessment tool were covered. A study was rated as a high risk of bias if at least one of the seven items was not fulfilled. Beyond that, a study was labeled as an unclear risk of bias. Any divergence in the quality assessment of studies was settled by consensus.

Statistical analysis
The data was statistically analyzed by two independent investigators (Zhuman Wu and Chengnian Wu). In this systematic review and network meta-analysis, all outcomes of interest were dichotomous data. Therefore, the pooled risk ratio (RR) with 95% confidence intervals (CIs) was calculated to express it [21]. In a pairwise meta-analysis, heterogeneity across studies was first qualitatively assessed with the Cochrane Q, and then I 2 statistic was used to quantitatively estimate the level of heterogeneity [22]. Studies were deemed to be homogeneous if P > 0.1 and I 2 < 50. Otherwise, studies were considered as heterogeneous when P < 0.1 and I 2 > 50. All traditional head-to-head meta-analyses were performed with the random-effects model, which simultaneously considered within-and between-study heterogeneity. Publication bias was checked by drawing a funnel plot when the number of eligible studies for individual outcome was more than 10 [23], and an asymmetry suggested publication bias [24]. Traditional pairwise metaanalysis was performed using Review Manager 5.3 (Cochrane Collaboration, Copenhagen, Denmark).
Random-effects network meta-analyses were conducted using Markov Chain Monte-Carlo Methods in OpenBUGS 3.2.3 (MRC Biostatistics Unit, Cambridge, UK) following the methods described by Lu and Ades [25,26]. The initial value automatically generated from the software was used to fit the model [27]. The Markov Chain Monte-Carlo method with 50,000 iterations and 20,000 burn-in was used to gain convergence. The summary treatment effect estimates were presented as odds ratios (ORs), with 95% credible interval (CrI) for treatment comparisons. The comparison-adjusted funnel plot was drawn to assess the small-study effects when the number of studies included in one pair of comparison was more than 10 [28]. The inconsistency factor was calculated using the loop-specific method to assess the inconsistency [29]. The ranking probabilities of being at each possible rank were estimated for all treatments, and the surface under the cumulative ranking curve values was used to provide a hierarchy of treatments [30].
Thirteen comparisons reported the incidence of allcause mortality, and the meta-analysis did not identify significant differences. All pooled results are delineated in Supplementary file 3 (Fig. S4). Moreover, nine comparisons also reported the incidence of IFI-related mortality. No significant difference was observed among all comparisons, which are delineated in Supplementary file 3 (Fig. S5).
Eleven comparisons reported the proportion of successful treatment. The meta-analysis suggested that The hierarchies of all drugs were generated on the basis of SUCRA values for prophylaxis against proven and probable IFIs. The results indicated that   (Fig. S10).
The network meta-analysis showed no significant difference among all comparisons in terms of the need for empirical treatment and the proportion of successful treatment (Table 4).

Publication bias and network coherence
The split-node method was adopted to generate the inconsistency plot so as to check the consistency of results from direct and indirect comparisons. The results of inconsistency plot indicated consistency in terms of proven and probable IFIs (Fig. 4). No evidence of publication bias based on comparison-adjusted funnel plot asymmetry was found (Fig. 5), although the number of studies included in each comparison was very small, thereby making the available methods for evaluating publication bias somewhat unreliable.

Discussion
IFIs remain a leading cause of morbidity and mortality among patients at high risk [4,7] due to elusive identification of IFIs in the early stage [8,9]. Therefore, prophylaxis strategies are crucial in the containment of IFIs [3]. Previous traditional direct meta-analyses and network meta-analyses did not consider all prophylaxis treatments and did not incorporate all potentially eligible studies, thus restricting the reference value of previous findings for making decisions in clinical practice. The present network meta-analysis was performed on 35 studies, including 37 RCTs involving 8513 patients, to generate more comprehensive and reliable results.
The valuable findings of this network meta-analysis were as follows: (a) fluconazole, itraconazole, posaconazole, voriconazole, amphotericin B, and caspofungin, but not micafungin, had the potential of reducing the incidence of proven and probable IFIs; (b) posaconazole was superior to fluconazole, itraconazole, and amphotericin B against proven and probable IFIs; (c) posaconazole was superior to fluconazole, itraconazole, voriconazole, amphotericin B, micafungin, and placebo against IArelated IFIs, and itraconazole had the potential of reducing IC-related IFIs; (d) posaconazole was superior to fluconazole, itraconazole, and amphotericin B in terms of all-cause mortality; (e) fluconazole, itraconazole, posaconazole, voriconazole, amphotericin B, or caspofungin had the potential of reducing the risk of adverse events; (f) fluconazole, itraconazole, posaconazole, voriconazole, and amphotericin B were superior to micafungin in reducing the risk of adverse events, and fluconazole and voriconazole were superior to posaconazole and amphotericin B; (g) all treatments were not different in terms of the need for empirical treatment and the proportion of successful treatment; and (h) posaconazole had the highest probability of being ranked the best against proven and probable IFIs.
To date, four topic-related meta-analyses [3,4,16,17] included two traditional pairwise meta-analyses [16,17] and two network meta-analyses [3,4]. In 2002, Bow and colleagues [16] performed a meta-analysis of randomized controlled clinical trials to investigate the overall clinical efficacy of antifungal prophylaxis, including azole antifungal agents and low-dose intravenous amphotericin B, for severely neutropenic chemotherapy recipients. The aforementioned analysis included 38 eligible studies and showed that antifungal prophylaxis could reduce allcause mortality and IFI-related mortality. However, the efficacy and safety of individual antifungal prophylaxis agents were not investigated, thereby mitigating the reference value of the findings. On the contrary, the    present analysis explored pure efficacy and safety of individual agents against IFIs and suggested that posaconazole was associated with the reduced incidence of allcause mortality. In 2005, Vardakas and colleagues [17] separately investigated the comparative efficacy of fluconazole versus itraconazole for antifungal prophylaxis in neutropenic patients with hematological malignancies. The pooled results based on five RCTs suggested that itraconazole was more effective than fluconazole in preventing IFIs in neutropenic patients with hematological malignancies; however, it was also associated with more adverse effects. The present analysis incorporated 35 studies involving 37 RCTs to estimate the mixed efficacy of antifungal prophylaxis agents and found no significant difference between fluconazole and itraconazole in terms of the incidence of IFIs, mortality, and adverse events; the need for empirical treatment; and the proportion of successful treatment. In 2011, Freemantle et al. [4] compared between a systematic review and mixed treatment to investigate the potential of empirical, pre-emptive, and directed treatment strategies for invasive mold infections. This study suggested that caspofungin was superior to amphotericin B and voriconazole in the outcome of survival, and voriconazole was superior to amphotericin B for overall survival. However, the present study found no difference among caspofungin, amphotericin B, and voriconazole in terms of mortality. In 2016, Zhao and colleagues published a network metaanalysis [3] and found that all triazole antifungals were effective in preventing IFIs, which was consistent with the findings of the present analysis. Better than Zhao's network meta-analysis, the present analysis also suggested that amphotericin B and caspofungin were effective against IFIs. Moreover, Zhao et al. found that posaconazole was more efficacious in reducing IFIs and all-cause death compared with fluconazole and itraconazole, which were also consistent with the findings of the present analysis. The strength of this meta-analysis included the comprehensive and simultaneous assessment of the relative efficacy of all treatments against IFIs among patients at high risk. Given limited comparative effectiveness studies, it was difficult for patients and physicians to make informed decisions regarding which treatments were the most effective against IFIs. However, the meta-analysis had certain limitations related to both network analysis and individual studies, which merits further discussion. First, direct comparative effectiveness studies were scarce. Second, network meta-analyses might be susceptible to misinterpretation. The biggest threat to the validity of a network meta-analysis was conceptual heterogeneity involving considerable differences in participants, interventions, and specified regimes of targeted treatments, thus limiting the comparability of trials. It was assumed that patients enrolled in all included studies were sampled from the same theoretical population [59,60]. However, subtle differences were found in characteristics related to patients (adult patients, pediatric patients, patients receiving intensive chemotherapy for remission, and patients undergoing HCST), treatments (dose or form of individual treatment), and administration of agents (intravenous and oral). Third, ranking probabilities might be challenging to understand and did not always imply a clinically important difference. Hence, clinical decisions based on the findings should be made cautiously.
The individual studies included in the analysis also had some limitations, which also undermined the strength of the meta-analysis. Most of the studies focused on the efficacy against IFIs, with very few studies on mortality and adverse events, which limited the assessment of benefits of treatments, and hence a thorough assessment of risk-benefit profile could not be performed. Studies were also under the risk of detection bias with the suboptimal reporting of blinding of outcome assessors. Various study designs, including multicenter and single center, were used in different eligible studies. However, further sensitivity analysis or subgroup analysis was not designed based on the study design due to an insufficient number of eligible studies for the majority of comparisons. Therefore, it was critical to further investigate the impact of study design on pooled results when a sufficient number of eligible studies were published. Moreover, subgroup or sensitivity analysis was not designed according to the follow-up time due to an insufficient number of eligible studies for individual comparison. However, the time effects of treatments were investigated in individual studies, and no novel findings were reported [13,51].

Conclusions
Despite these limitations, the present network metaanalysis provided a better understanding of the comparative efficacy of all potential treatments against IFIs among patients who were at high risk. Posaconazole might be a promising option against IFIs because it was superior to fluconazole, itraconazole, amphotericin B, voriconazole, or micafungin, although no significant difference was detected compared with caspofungin in terms of proven and probable IFIs and IA-related IFIs. Moreover, posaconazole also reduced all-cause mortality compared with fluconazole and itraconazole, and reduced the risk of adverse events compared with amphotericin B, fluconazole, itraconazole, posaconazole, voriconazole, amphotericin B, and caspofungin, although all treatments showed no significant difference in terms of the need for empirical treatment and the proportion of successful treatment.