Background
Invasive fungal infections (IFI) are a significant cause of morbidity and mortality in allogeneic hematopoietic cell transplant (alloHCT) recipients, with invasive mould infections due to
Aspergillus spp. (invasive aspergillosis, IA) being most prevalent [
1-
4]. Early treatment strategies and antifungal prophylaxis are options for mitigating the impact of IFI in this population [
5-
7]. Although a meta-analysis published in 2007 concluded that antifungal prophylaxis reduced all-cause mortality, IFI-related mortality, and IFI incidence in alloHCT recipients [
8], a more recent systematic review failed to demonstrate consistent treatment effects for these outcomes using direct and indirect comparisons [
9]. For antifungal prophylaxis, particularly in the long-term outpatient setting, oral antifungals have the potential to be convenient and cost-effective [
10,
11]. However, the optimum oral agent for antifungal prophylaxis in alloHCT recipients post-transplant remains uncertain.
For physicians faced with the challenge of selecting a systemically active oral antifungal, the principal choices are fluconazole, which lacks anti-mould activity, and the mould-active agents itraconazole, posaconazole, and voriconazole. To our knowledge no single head-to-head randomized clinical trial (RCT) has directly compared more than 2 of these options in alloHCT recipients. The paucity of such studies impedes the use of traditional pairwise meta-analysis to inform the clinical decision-making process.
Network meta-analysis, can synthesize head-to-head comparisons of interventions not directly compared in clinical trials, as long as these interventions share one or more common comparators in a network of evidence. Furthermore, mixed treatment comparison (MTC) network meta-analyses allow for the combination of both direct and indirect evidence [
12,
13], and have been successfully employed to address similar questions in numerous clinical areas, including cardiovascular disease, osteoporosis, and bacterial infections [
14-
18], as well as for comparisons of different agents and strategies for antifungal treatment [
19,
20]. To extend a previously published traditional meta-analysis of antifungal prophylaxis [
8], we conducted a systematic literature review and MTC of RCTs evaluating fluconazole, itraconazole, posaconazole, and voriconazole as primary antifungal prophylaxis in alloHCT recipients, including recently published trials. Our objective was to compare the efficacy of these agents for the prevention of documented IFI in alloHCT recipients based on several key outcomes, with the purpose of informing medical decision-making.
Discussion
Transplant physicians are frequently faced with the difficult choice of selecting the most appropriate and efficacious option for oral antifungal prophylaxis in alloHCT recipients. In the absence of a large, multi-arm RCT comparing all systemically active oral antifungals, network meta-analyses can provide relevant information to help guide health intervention decision-making; this methodology is increasingly utilized for similar purposes across therapeutic areas [
12,
36].
Bayesian statistical inference differs from classical statistics in that it provides probability distributions for treatment effects, expressed as posterior credible intervals (rather than confidence intervals) [
37]. One advantage of using Bayesian credible intervals is that they can be more intuitively interpreted in terms of the probability that relative efficacy lies within a specific range. Bayesian methods, therefore, allow us to directly report the probabilities of outperforming fluconazole or of being the best agent overall for each mould-active azole, which physicians can then factor into the selection process [
38].
While our base-case MTC analysis allowed for substantial heterogeneity across the studies, our conservative approach came at the cost of a reduced statistical inference and wider credible intervals, thereby reducing our ability to detect actual differences between treatments. Thus, the available data did not allow our base-case MTC to distinguish between itraconazole, posaconazole, and voriconazole when using the 5% threshold for type-I error typically employed in classical hypothesis testing. Our sensitivity analysis using an empirical prior yielded some comparisons that did meet the traditional 5% threshold, ie, itraconazole and voriconazole had lower IFI risk and posaconazole and voriconazole had lower IA risk than fluconazole. Regardless, our objective was not to test causal research hypotheses (when prespecified confidence thresholds are useful), but rather to help decision-makers compare the efficacy of different interventions [
39,
40].
Although regulatory authorities will always prefer a high degree of certainty for these comparisons, the requirement for a specific threshold of “confidence” (typically 95%) may result in sub-optimal outcomes in situations where a treatment decision cannot be deferred [
41,
42]. Such is the case for antifungal prophylaxis in alloHCT recipients: while the most efficacious oral agent is currently unknown, a choice must still be made in those patients who are deemed to benefit from a prophylactic approach. The probabilities of superiority estimated by MTC represent an objective measure of comparative efficacy that can be taken into account when making this choice.
We note that our post-hoc empirical Bayesian sensitivity analysis did achieve statistical significance at the standard 95% level in several of the important outcomes: the posterior probabilities of itraconazole/voriconazole being superior to fluconazole for prevention of IFI overall, posaconazole/voriconazole being better than fluconazole for prevention of IA, and itraconazole being better than fluconazole for prevention of IC. The probability of voriconazole being superior to fluconazole for the reduction of OLAT use was found to be 94% in this sensitivity analysis. Empirical Bayesian methods are sometimes criticized for “using the data twice” [
43], which is the reason this approach was not chosen for the base-case analysis. However, this method can improve statistical inference and has previously been shown to provide accurate inference in random effects meta-analysis [
24].
Based on estimated probabilities of superiority, our analyses suggest that broad-spectrum mould-active azoles are more effective than fluconazole as antifungal prophylaxis in alloHCT recipients post-transplant, a result largely driven by fluconazole’s lack of anti-mould activity; this finding is consistent with a previously published meta-analysis [
8]. Among the mould-active azoles, posaconazole and voriconazole reduced the risk of IA more than itraconazole. In contrast, itraconazole was the most effective in preventing IC, which is also consistent with published meta-analyses [
20,
44]. Compared with fluconazole, voriconazole had the greatest probability of reducing OLAT use, which may have both clinical and pharmacoeconomic implications. Other outcomes of potential interest, such as incidence of possible IFI and fungal-free survival, could not be evaluated, since relevant data were not consistently reported in the eligible RCTs.
Currently, IFI caused by
Aspergillus spp. predominate [
2,
3] and reduced intensity conditioning (RIC) transplants are now commonplace, with 42% of patients in the recent voriconazole versus itraconazole study having undergone RIC/nonmyeloablative conditioning [
10]. In RIC patients, IFIs, particularly invasive mould infections, tend to occur during the late post-engraftment period (ie, after day 100) [
2,
45]; these patients may therefore require longer periods of mould-active antifungal prophylaxis. Of note, the studies included into our evidence network assessed patients for a post-engraftment period of up to 180 days, but the at-risk period for invasive mould infections extends beyond this period [
2,
3,
45].
The included studies were heterogeneous in terms of the study design, patient population and risk of IFI, such that recognition of possible treatment effects may have been obscured. For example, the IFI rates varied across the studies and are likely to reflect a similar variance in IFI risk. The highest IFI rate was observed in the study by Winston and colleagues [
29] where a greater proportion of fluconazole recipients received unrelated donor stem cells, and had a higher incidence of acute and chronic graft-versus-host disease (GvHD), thus amplifying the difference in IFI event rates between study and control groups.
The incidence of grades II–IV acute GvHD as a risk factor for IFI also varied significantly among the studies from 100% [
27], to 64% [
26], 46% [
10], 41% [
28], and 37% [
29]. The IFI risk may have been influenced by the heterogeneity of conditioning regimen intensities among the studies included in the analysis.
All (100%) subjects in the Seattle study [
26] and the study of Blood and Marrow Clinical Trials Network [
28] received myeloablative conditioning regimens compared to 78% in the study by Winston et al. [
29] and 58% in the multicentre European study [
10]. Accordingly, the influence of pre-engraftment myelosuppression and cytotoxic therapy-induced intestinal epithelial damage on IFI risk may have varied among the studies.
There was a significant variance in the prophylaxis start dates among the studies ranging from the beginning of conditioning [
26], to the day of transplant [
10,
28], the day after transplant [
29], and the day of documentation of GvHD (median of day 64 post-transplant) [
27]. This latter study only reported IFI incidence at 112 days post-treatment initiation, and did not address the incidence of IFI and OLAT use between the time of transplant and the start of study prophylaxis [
27]. The decision to include the posaconazole trial was validated by the alignment of the results of the base-case analysis with those of the post-hoc sensitivity analysis in which the study was excluded.
Similarly, there were significant variations in toxicity- or intolerance-driven drug withdrawal rates that likely influenced prophylaxis drug exposure and efficacy. Itraconazole withdrawal rates ranged from a low of 8.5% [
29], to 36% [
26] and 43% [
10]. Fluconazole withdrawal rates ranged from 1.5% [
29], to 16% [
26], 38% [
27], and 44% [
28]. Voriconazole withdrawal rates were similar at 41% [
28] and 37% [
10]. The posaconazole withdrawal rate was 34% [
27]. Our inability to control for all of these different variables, reflected in the heterogeneity of the included studies, reduced the sensitivity of the analysis to detect treatment effects for the outcome of interest, the use of a random effects model notwithstanding.
Application of the 2008 revised definitions for the end-points for the prophylaxis studies [
46] may have provided a more robust basis for prophylaxis efficacy outcomes as has been noted for treatment outcomes [
47,
48]; however, we used the definitions for invasive fungal infection employed in the methods of the included trials for consistency.
Comparative efficacy notwithstanding, considerations that may impact the decision-making process include cost differences, ease of use, availability of an intravenous formulation (in case of mucositis and/or intestinal GvHD), adverse event and drug-drug interaction profile, availability of expertise and diagnostic tools for early diagnosis of invasive mould infection, and local IFI epidemiology. While substantial cost differences exist between generic fluconazole and itraconazole on the one hand and posaconazole and voriconazole on the other, the drug costs associated with OLAT should be considered as well. Oral and gastrointestinal mucositis may limit the role of posaconazole due to the unavailability of an intravenous formulation and the requirement for administration with a full meal [
49]. Of note, all of the mould-active azoles adversely interact with immunomodulatory and antineoplastic drugs. Prophylactic fluconazole may be a worthwhile alternative to mould-active prophylaxis in alloHCT in centres practicing early diagnostics-driven therapy [
8].
It should be noted that patient-level data (from published papers), rather than (raw) data extracted from clinical study reports, were used to drive this MTC meta-analysis.
Competing interests
Some of the authors (RC, MK, DW, HS) are employees of Pfizer and as such were involved in gaining consensus on the systematic review parameters, but were not involved in conducting the actual literature review. EB has served as a consultant for and received honoraria from Pfizer, Astellas, and Merck-Frosst, and has served as a consultant for Teva. CC has received honoraria from Astellas, and has received honoraria from and served in a consultant/advisory role for Gilead Science, Pfizer, and MSD. OC has served in a consultant/advisory role for Sanofi Pasteur, received research funding from Actelion, Bayer, Celgene, Genzyme, Miltenyi, Quintiles, and Viropharma, received research funding from and served in a consultant/advisory role for 3 M, Basilea, Cubist, F2G, GSK, and Optimer, and received research funding and honoraria from and served in a consultant/advisory role for Astellas, Gilead, Merck/MSD, and Pfizer. LC and SS have served with Evidera in a consultant/advisory role for Pfizer. DV has served with Evidera in a consultant/advisory role for Pfizer, and has received research funding from Evidera. AS has served with Evidera in a consultant/advisory role for, and has received research funding from, Pfizer. AP has received honoraria from Pfizer and Merck. HS owns stock in Pfizer. MS has received research funding and honoraria from and served in a consultant/advisory role for Pfizer, Merck, and Gilead. CS has received honoraria from and served in a consultant/advisory role for MSD, Astellas, and Pfizer. DM has nothing to disclose.
Authors’ contributions
Conception: EJB, DW Study design: EJB, MS, CC, OAC, DIM, AP, CS, MK, HS (systematic literature review); DJV, LC, SS, AJS, RC (mixed treatment comparison) Collection and assembly of source data: LC, AJS, SS, EJB Data analysis and interpretation: All authors Manuscript writing: All authors read and approval the final manuscript.