Meta-analysis of the safety of voriconazole in definitive, empirical, and prophylactic therapies for invasive fungal infections

We conducted a literature search of PubMed, Embase, and the Cochrane Library from their inceptions up to December 15, 2016. An additional manual literature search was performed by checking the reference lists in eligible articles. The MeSH terms used for keyword and text-word searches were “antifungal agents” and “voriconazole”. The search was limited to human studies.

Two reviewers independently evaluated each study and identified whether they met the predefined inclusion criteria. We used the PRISMA criteria for searching and selecting studies (Fig. 1). The following inclusion criteria were applied to identify eligible studies: (i) published in the English language before December 15, 2016; (ii) designed as an RCT; (iii) comparing the safety between VRC and other antifungal agents in definitive, empirical, or prophylactic therapy for IFIs; and (iv) providing the incidence of toxicity at least one term among discontinuation, neurotoxicity, visual toxicity, hepatotoxicity, or nephrotoxicity. We excluded studies that enrolled only pediatric patients, studies only focusing on superficial or mucocutaneous fungal infection, and studies of pharmacokinetics, clinical efficacy, and mycology.

The primary outcome was the proportion of patients who discontinued antifungal therapy due to adverse events. The secondary outcome of interest was the cumulative incidence of adverse events related to neurotoxicity, including headache, dizziness, restlessness, hallucination, insomnia and depression and some other nervous system disorders and psychiatric disorders according to version 4.0 of the Common Terminology Criteria for Adverse Events (CTCAE) [16]. Another secondary outcome is visual toxicity which was defined as the occurrence of abnormal vision, photophobia, chromatopsia and some other eye disorders according to CTCAE. We also evaluated the cumulative incidence of laboratory tests of liver dysfunction (serum transaminase, alanine transaminase, alkaline phosphatase, or bilirubin levels) and renal dysfunction (serum creatinine), which represented the safety outcomes of hepatotoxicity and nephrotoxicity. We directly used the laboratory test cutoff values reported for individual studies, since the criteria varied among the different studies.

Two reviewers independently extracted specific information from studies, including year of publication, authors, study design, therapeutic purposes, antifungals administered, sample size, age, underlying diseases, dose and duration of administration, and safety outcomes including tolerability, neurotoxicity, visual toxicity, hepatotoxicity, and nephrotoxicity. Disagreements about the specific data between two reviewers were resolved through consensus. We attempted to contact the authors by email to seek required data that were missing from the original reports.

Two investigators independently appraised the sources of bias in RCTs according to the Cochrane Collaboration guidelines, and classified each RCT into low, unclear, or high risk by evaluating the following domains: random sequence generation, allocation concealment, blinding of participants and outcome assessment, incomplete outcome data, selective outcome reporting, and other issues [17]. Discrepancies were resolved by discussion with a third investigator. RCTs for which two or more domains were judged as having a high or unclear risk were regarded as having a high overall risk of bias.

The electronic and manual searches identified 6547 studies, of which 16 RCTs [18‐33] involving 4122 randomized patients met the criteria and so were entered into the final analysis (Table 1). A flow diagram of the study selection process is shown in Fig. 1 (see more detailed searching strategy in Additional file 1). The safety outcomes for tolerability, neurotoxicity, visual toxicity, hepatotoxicity, and nephrotoxicity were reported for 12, 10, 13, 14, and 9 studies, respectively. The main underlying disease of patients in the included RCTs was hematological malignancy, followed by solid cancers and other diseases with a high risk of fungal infection. The antifungal agents used in the counterpart groups included fluconazole (n = 3), amphotericin B (n = 3), itraconazole (n = 4), micafungin (n = 4), isavuconazole (n = 1), and a combination of fluconazole and AMB (n = 1). There were eight, three, and five studies that had definitive, empirical, and prophylactic treatment purposes, respectively. The duration of drug therapy was at least 2 weeks in 12 RCTs, less than 2 weeks in 3 RCTs, and unknown in the remaining RCTs.

Figure 2 presents the results of the systematic bias analysis. There was a low risk of bias for most items, except for the presence of performance and detection bias due to the lack of a double-blind design and blind outcome assessment in four studies [18, 22, 27, 32].

In a further subgroup analysis by types of antifungal agents, therapeutic purpose, and duration, the risk of adverse events differed between treatment with VRC and other antifungals (Table 2).

The risk of discontinuation did not differ significantly between VRC and each single individual antifungal except for micafungin, which was associated with significantly reduced odds of discontinuation compared with VRC. According to the subgroup analysis of various treatment purposes, we found that VRC showed a significant poor tolerance in the definitive treatment compared with the empirical and prophylaxis treatment. Concerning the duration of treatment, VRC was more likely to be discontinued than other antifungal agents in patients treated for longer than 14 days. VRC was associated with significantly high risk of neurotoxicity compared with AMB and micafungin while other triazoles including fluconazole, itraconazole and isavuconazole failed to provide a significant reduction or increase in the incidence rate of neurotoxicity. Additionally, we also identified statistically significance in the odd of neurotoxicity in VRC versus other antifungal agents in the subgroup of empirical treatment and duration less than 14 days. Of notes, the studies involved in these two groups are the same, including the counterparts of AMB and micafungin, which may lead to this significant diffference owing to the antifungal type. However, there was no statistically significant difference in the group of definitive and prophylactic treatment as well as duration ≥14 days. The high risk of visual toxicity during VRC treatment was consistent across all analyzed subgroups and the heterogeneity was also reduced attribute to the stratification of three subgroups analysis. For the outcome of hepatotoxicity, VRC showed a higher risk of abnormal liver enzyme levels compared with itraconazole, micafungin, and isavuconazole, but not for fluconazole or AMB. Patients who were diagnosed as definitive IFIs had a high risk of the elevation of liver enzyme levels in the VRC group. In addition, a duration of at least 2 weeks seemed to be a risk factor for liver injury when patients were treated with VRC rather than other antifungals. The heterogeneity of hepatotoxicity was also reduced by stratifying the data by all three subgroups. The occurrence of abnormal renal function did not differ significantly between VRC and the other antifungal agents except AMB, which was associated with a higher risk of nephrotoxicity. The probability of abnormal renal function in definitive and empirical fungal therapies was lower for VRC than for the composite of the other five antifungals. However, the risk of nephrotoxicity was significantly higher for patients treated with other antifungals than for VRC for treatment durations shorter than 14 days.

The sensitivity analysis verified the robustness of the findings for each outcome. A leave-one-out sensitivity analysis showed that no individual study excessively influenced the pooled effect in our meta-analysis except for the outcome of neurotoxicity (see Additional file 2: Figs. S1–S5), which was affected by removing 5 single study (including studies of Walsh, Queiroz-Telles, Kohno, Mattiuzzi and Oyake 2 [21, 26, 29, 30, 32]). In addition, the results for each outcome remained stable after excluding RCTs with samples smaller than 50, with a single-center design or with a high risk of bias (see in Additional file 3: Table S1).

The previous meta-analysis regarding the status of voriconazole among all antifungal agents mostly focused on the efficacy and pharmacoeconomics [42‐44]. To our knowledge, this is the first study to have comprehensively compared five common adverse events induced by VRC and other antifungals by systematically reviewing RCTs. The results can be used by clinicians to avoid unnecessary adverse effects and select a better antifungal therapeutic scheme for patients. Furthermore, we also quantified the specific outcomes concerning different antifungals, treatment purposes, and durations by subgroup analysis for four safety outcomes.

Several limitations of our study should be considered. First, as is inevitable for any meta-analysis, our study shares the limitations of the original studies. However, we performed the largest pooled comparison of the toxicity profiles of VRC and a composite of other common antifungal agents based on RCTs. Second, we directly used the criteria applied in each study because the thresholds for abnormal liver and renal enzyme levels varied between them. Finally, the combination of various antifungals in the counterparts group may have been responsible for the moderate heterogeneity and a certain degree of publication bias among the studies (see in Additional file 4: Figure S6). The present findings therefore need to be confirmed in future high-quality studies.