Comparative Risks of High-Grade Adverse Events Among FDA-Approved Systemic Therapies in Advanced Melanoma: Systematic Review and Network Meta-Analysis

Background: Head-to-head evidence is lacking in comparative risks of high-grade adverse events (AEs) among different systemic treatment options for advanced melanoma. Methods: An up-to-date systematic review and network meta-analysis (NMA) was performed. Randomized controlled trials (RCTs) of patients with advanced melanoma were eligible if at least one intervention was the Food and Drug Administration–approved targeted or immune checkpoint inhibitors. Risks of high-grade AEs were estimated by random-effects Bayesian NMAs, based on relative risks. Surface under the cumulative ranking probabilities was used to assess relative ranking of treatments. The summary incidences were calculated. Results: Twenty-five RCTs (12,925 patients) comparing 10 different systemic treatment options were included. BRAF/MEK had the highest risk of overall high-grade AEs (pooled incidence: 32.11%). BRAF had the highest risk of high-grade arthralgia (0.39%), whereas MEK had the highest risk of high-grade hypertension (2.28%) and nausea (0.37%). Cytotoxic T-lymphocyte antigen 4 (CTLA-4)/chemo had the highest risk of high-grade diarrhea (1.31%), alanine aminotransferase (0.60%), and aspartate aminotransferase elevation (0.59%). Programmed cell death 1 (PD-1)/CTLA-4 had the highest risks of high-grade pyrexia (1.14%) and rash (0.94%). Using PD-1 inhibitor alone had the lowest risks of overall high-grade AEs. Conclusions: Different systemic treatment options have varying high-grade AEs in advanced melanoma treatment. Current evidences highlight the important risks of BRAF/MEK, CTLA-4/chemo, and PD-1/CTLA-4.


INTRODUCTION
Systemic therapy is the main treatment modality for patients with advanced melanoma (1). The landscape of systemic treatment options is changing rapidly in recent years from traditional interferon α to novel mitogen-activated protein kinase pathway inhibitors (i.e., BRAF inhibitors and MEK inhibitors) and immune checkpoint inhibitors (ICIs) [i.e., programmed cell death 1 inhibitors (PD-1) and cytotoxic T-lymphocyte antigen 4 inhibitors (CTLA-4)] (2). Results from randomized controlled trials (RCTs) have shown that these new agents have drastically improved progression-free survival (PFS) and overall survival (OS) in patients with advanced melanoma (3,4). However, highgrade adverse events (AEs) related to these targeted inhibitors and ICIs remain a concern in clinical practice (1).
Medical decision-making for patients with advanced melanoma is a major challenge for clinicians. It is important to balance between the clinical benefits and potential high-grade risks of each systemic treatment option during decision making (2,5). Systematic review and network meta-analysis (NMA) have been conducted to provide high-quality evidences to support the medical decision-making. For example, previous studies found BRAF plus MEK combination was the most favorable therapy to improve PFS, whereas PD-1 was associated with improved OS benefit (5)(6)(7)(8)(9). However, these studies were mainly focused on the comparative efficacy. The risks of severe, life-threatening AEs or deaths related to the BRAF-or MEK-targeted inhibitors or ICIs treatments were not adequately summarized for patients with advanced melanoma (6,9).
The decision about systemic therapies to patients with advanced melanoma should be informed not only by the reduction of recurrence risk or OS improvement, but also by careful management of high-grade risks (10). In the absence of a direct comparison among different systemic treatment options to guide the clinical decision-making, it has been unclear which treatment strategy has the highest high-grade AEs to patients with advanced melanoma. A comprehensive understanding of the high-grade AEs of these novel targeted and immunotherapy agents is needed for informed these clinical decisions. We conducted a NMA to compare high-grade AEs of the Food and Drug Administration (FDA)-approved ICIs and targeted inhibitors for patients with advanced melanoma.

Study Design
This NMA was reported based on the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (11,12). A priori established review protocol was followed when the study was conducted. The review protocol was registered in the PROSPERO international prospective register of systematic reviews (CRD42020160453).

Search Strategy and Selection Criteria
The final searches of PubMed, EMBASE, and Cochrane Library were conducted up to December 20, 2019, using the combinations of the following terms: (melanoma OR melanocyte) AND (ipilimumab OR yervoy OR nivolumab OR opdivo OR pembrolizumab OR keytruda OR binimetinib OR mektovi OR cobimetinib OR cotellic OR dabrafenib OR tafinlar OR encorafenib OR braftovi OR trametinib OR mekinist OR vemurafenib OR zelboraf OR "cytotoxic T-lymphocyte antigen 4" OR "programmed cell death 1 receptor" OR "BRAF" OR "MEK") AND (random OR control OR phase II OR phase III OR placebo) without restriction on year of publication or language. The detailed search strategies are listed in Supplementary Table 1.
Trials were eligible if the following inclusion criteria were met: (1) patients with advanced melanoma regardless mutation status; (2) at least one of the interventions compared in the trial was either the FDA-approved ipilimumab, nivolumab, pembrolizumab, binimetinib, cobimetinib, dabrafenib, encorafenib, trametinib, vemurafenib, their combinations, or chemotherapy with their combinations; (3) high-grade AEs were extractable either from published articles or unpublished reports from clinicaltrial.gov; (4) phase II or III RCTs. We excluded (1) commentaries, letters, editorials, protocols or reviews; (2) trials only in conference abstracts/posters form; (3) phase I, dose escalation or single-arm trials; (4) in vitro or animal studies; and (5) studies of cost-effectiveness analyses or quality of life. The titles, abstracts and full texts were evaluated sequentially.

Quality Assessment
The risk of bias was assessed by two authors (HY and FH) independently. The domains assessed included random sequence generation (selection bias), allocation concealment (selection bias), blinding of participants and personnel (performance bias), blinding of outcome assessment (detection bias), incomplete   outcome data (attrition bias), selective reporting (reporting bias), and other bias (13).

Outcome Measures
The primary outcome was the incidence of overall highgrade AEs. The secondary outcome was the incidence of general symptomatic high-grade AEs (fatigue, pyrexia), general laboratory results-related high-grade AEs (ALT/AST elevation, hypertension), musculoskeletal/pain-related high-grade AEs (arthralgia, myalgia), gastrointestinal high-grade AEs (diarrhea, nausea), and cutaneous high-grade AEs (rash). Both the primary and secondary outcomes were defined as grades 3-5 AEs basing on the National Cancer Institute Common Terminology Criteria for Adverse Events (CTCAE) version 4.0.

Data Synthesis and Statistical Analysis
NMA was conducted based on the Bayesian framework using a Markov Chain Monte Carlo (MCMC) simulation technique. Non-informative priors were used to estimate the posterior distribution (14). The MCMC model was updated with 100,000 simulated draws after a burn-in of 20,000 iterations. We used a thinning interval of 10 for each chain. Brooks-Gelman-Rubin statistic was used to assess the adequacy of burn-in and convergence (15). Relative risks (RRs) along with corresponding 95% credible intervals were reported. Random-effects model was used because they generally show better goodness of fit. The posterior mean of the residual deviance was calculated to assess goodness of model fit. The incidence of both primary and secondary outcomes was estimated (incidence = 100 × assumed placebo risk × RR, the assumed placebo risk was generated by using traditional meta-analyses with random-effects model).
Hierarchy of both the primary and secondary outcomes was respectively estimated for all the treatment classes using median ranks and surface under the cumulative ranking curve (SUCRA). SUCRA was the percentage of drug safety on AEs that would be ranked first without uncertainty. When the drug safety was certain to be the best, the SUCRA value would equal one, whereas it would equal zero when the safety was certain to be the worst (16). The presence of inconsistency was evaluated by node splitting analysis in the entire network on particular comparisons (17,18). The P < 0.05 was regarded as significant inconsistency. All the data analyses were conducted using STATA version 14.0 and WinBUGs version 1.4.3.

Selection of Trials
Initially, 2,955 unduplicated records were identified by literature search. After screening of titles and abstracts, 2,895 records were excluded. Sixty articles were assessed for eligibility. Finally, 49 articles involving 25 RCTs were included for qualitative and quantitative synthesis (3,4, (Supplementary Figure 1).

Characteristics of Trials and Patients
The 25 RCTs covered 10 treatment classes and included 12,925 patients with advanced melanoma (   Table 4 shows the details of risk-of-bias assessment based on each trial.

Summary of Key Findings
This study fills a crucial knowledge gap regarding the comparative risks of high-grade AEs among the current FDAapproved systemic therapies in advanced melanoma. First, we found that the risk of overall high-grade AEs was highest for the BRAF/MEK inhibitor. Second, there were differences in the spectra of high-grade AEs among BRAF-targeted inhibitor (musculoskeletal toxicities and fatigue), MEK-targeted inhibitor (hypertension and nausea), CTLA-4 inhibitor (diarrhea and ALT/AST elevation), and PD-1/CTLA-4 inhibitors (pyrexia and rash). Third, using PD-1 inhibitor alone had the lowest risks of high-grade AEs for patients with advanced melanoma. Fourth, using CTLA-4 inhibitor alone at a low dose (i.e., ipilimumab at 3 mg/kg) decreased overall high-grade AEs significantly compared with using CTLA-4 inhibitor at a high dose (i.e., ipilimumab at 10 mg/kg).

Comparison With Other Studies
Our study agreed with the result from Franken et al. (5) that using PD-1 inhibitor alone was associated with the lowest risk of high-grade AEs. Devji et al. (7) showed that BRAF/MEK was associated with lower risk of high-grade AEs compared with using BRAF inhibitor alone despite the result was not significant. On the contrary, we found that BRAF/MEK inhibitors had the highest risk of overall high-grade AEs. The differences between our results and the results from Devji and colleagues' study may contribute to the updated trials included in our study (4,44,52,56,63). In addition, we focused on the FDA-approved targeted inhibitors and ICIs only. We considered that it would be more clinically relevant and would provide more useful evidence into clinical practice.
Previous studies combined high-and low-dose ipilimumab into one arm in analyses (5,7). In this study, we classified the ipilimumab into the high-dose (10 mg/kg) and low-dose (3 mg/kg) arms when calculating the comparative risks of SAEs. We found that using CTLA-4 inhibitor alone at a high dose (i.e., ipilimumab at 10 mg/kg) was associated with increased risk of high-grade AEs compared with PD-1 or chemotherapy.

Strength and Limitations of Study
To our knowledge, this is the first and most comprehensive NMA that investigated high-grade AEs among the FDA-approved ICIs and targeted inhibitors for patients with advanced melanoma. Previous NMAs either focused only on treatment efficacy or provided limited information on high-grade AEs. In addition, we classified the treatments by mechanism of action rather than analyzing the drugs separately. Thus, multiple trials would contribute to the comparison between two treatment categories. The network would be concise. It avoided yielding very sparse networks in analyses because of the current limited number of available trials. Combining different drugs of the same class within a single category may introduce heterogeneity. However, the values of the posterior mean of the residual deviance closely corresponded to the number of data points for the outcomes, indicating satisfactory model's fit.
Four limitations should be noted. First, this study provided evidence only on high-grade AEs with limited types. The tolerability of different treatments was not systematically investigated. In addition, because of limited information provided in each of the included studies and the very low incidence of grade 5 AEs (treatment-related deaths), we combined grades 3-4 AEs and grade 5 AEs. We used grades 3-5 AEs as the main outcome of this study. More clinically meaningful outcomes such as all-grade AEs, treatment-related deaths, or treatment discontinuation due to toxicities should be studied to compare the tolerability of different treatments in the future when more trials provide the detailed information. Further researches could also focus on other common AEs such as loss of weight, altered neurobehavioral responses, or other general laboratory results such as changes in blood or lipid profile. Second, the overall high-grade AEs investigated in this study included both non-immune-related reactions and immunerelated reactions. The latter was usually late onset. Current clinical trials of ICIs may not have follow-up interval that is long enough to identify the potential risks. Therefore, the incidence of high-grade AEs of ICIs may be underestimated. Standardized method that specifies the clinical criteria for immune-related AEs would be suggested to be published in the future. Third, commentaries, letters, or trials only in conference abstracts were excluded in this study because of the limited information they provided. Publication bias would be a threat if only fulltext articles with published data were extracted. Nevertheless, unpublished data from clinicaltrial.gov were obtained in this study to avoid publication bias. Last but not least, individual patient data (IPD) was not accessed in this study. Despite similar inclusion criteria across the included trials have added our confidence in the ability to estimate comparisons across the network of evidence, we still encourage IPD meta-analysis to be conducted in the future because it would provide more detailed patients' characteristics to identify the potential effect modifiers between the treatment options and the high-grade AEs.

Clinical and Research Implications
This study has obtained some unique clinical findings. First, we found a very similar overall high-grade AE risk between PD-1/CTLA-4 (i.e., nivolumab 1 mg/kg plus ipilimumab 3 mg/kg) and using ipilimumab at 10 mg/kg alone (RR = 1.00, 95% CrI = 0.86-1.24). However, the spectra of toxicity between them were different. Using ipilimumab at 10 mg/kg had a higher risk of diarrhea and ALT/AST elevation, whereas the combination of nivolumab 1 mg/kg and ipilimumab 3 mg/kg had a higher risk of high-grade pyrexia and rash. Second, the combination of ipilimumab and chemotherapy (CTLA-4/chemo) had the highest risk of high-grade fatigue, ALT/AST elevation, and diarrhea. CTLA-4/chemo was likely to be associated with increased overall high-grade AEs compared with using ipilimumab at 3 mg/kg alone, despite statistical significance was not detected. Third, focusing specifically on ipilimumab, we found that using ipilimumab at 3 mg/kg alone had a higher safety ranking compared with ipilimumab at 10 mg/kg. It is to be noted that ipilimumab at 3 mg/kg decreased high-grade diarrhea risk significantly, compared with using ipilimumab at 10 mg/kg (RR = 0.85, 95% CrI = 0.61-0.98). These findings presented above indicate that it is necessary for clinicians to be fully aware of these high-grade AEs and manage them appropriately according to the diagnosis criteria and treatment guidelines used across related trials. These comparative evidences of high-grade AEs could be used as important references when the clinicians balance against the improvements in clinical efficacy among different FDA-approved systemic therapeutic options and perform shared decision making with patients in advanced melanoma during clinical practice.
Two research implications could be noted. First, current RCTs conducted by pharmaceutical companies were mainly used as evidences to support the new drug application. Most RCTs would have chemotherapy as the control group. Direct evidence compared among targeted inhibitors and ICIs is still lacking. For example, BRAF inhibitors and PD-1 inhibitors have never been directly compared. We encourage more RCTs of real-world study be conducted in the future to focus on head-to-head comparisons among targeted inhibitors and ICIs. Second, the outcomes in this study were defined basing on CTCAE, because previous study showed that data from the analysis of AEs by severity to define serious AEs (SAEs) would be more informative (66). However, the safety data provided to FDA for new drug applications usually include only SAEs, which may not adequately reflect the safety signal. We encourage more data from the analysis of AEs by severity be reported in the future.

CONCLUSIONS
This current systematic review and NMA provides the most comprehensive comparison of high-grade AEs between targeted inhibitors and ICIs for the treatment of advanced melanoma. Our results show that different systemic treatment options have varying high-grade AEs and highlight the important risks of BRAF/MEK, CTLA-4/chemo and PD-1/CTLA-4 in advanced melanoma treatment.

DATA AVAILABILITY STATEMENT
All datasets generated for this study are included in the article/supplementary material.

AUTHOR CONTRIBUTIONS
Y-fH, W-jX, and JD: study conception and paper writing. Y-fH and JD: study design and discussion of the findings. Y-fH and H-yF: data extraction and elaboration. Y-fH, W-jX, and H-yF: data analysis and interpretation. Y-fH, W-jX, H-yF, and JD: all coauthors have read and approved the manuscript in its present form, agreed to be personally accountable for the author's own contributions and to ensure that questions related to the accuracy or integrity of any part of the work, even ones in which the author was not personally involved, are appropriately investigated, resolved, and the resolution documented in the literature. All authors contributed to the article and approved the submitted version.