A matching-adjusted indirect comparison of combination nivolumab plus ipilimumab with BRAF plus MEK inhibitors for the treatment of BRAF-mutant advanced melanoma☆

Background Approved first-line treatments for patients with BRAF V600–mutant advanced melanoma include nivolumab (a programmed cell death protein 1 inhibitor) plus ipilimumab (a cytotoxic T lymphocyte antigen-4 inhibitor; NIVO+IPI) and the BRAF/MEK inhibitors dabrafenib plus trametinib (DAB+TRAM), encorafenib plus binimetinib (ENCO+BINI), and vemurafenib plus cobimetinib (VEM+COBI). Results from prospective randomized clinical trials (RCTs) comparing these treatments have not yet been reported. This analysis evaluated the relative efficacy and safety of NIVO+IPI versus DAB+TRAM, ENCO+BINI, and VEM+COBI in patients with BRAF-mutant advanced melanoma using a matching-adjusted indirect comparison (MAIC). Patients and methods A systematic literature review identified RCTs for DAB+TRAM, ENCO+BINI, and VEM+COBI in patients with BRAF-mutant advanced melanoma. Individual patient-level data for NIVO+IPI were derived from the phase III CheckMate 067 trial (BRAF-mutant cohort) and restricted to match the inclusion/exclusion criteria of the comparator trials. Treatment effects for overall survival (OS) and progression-free survival (PFS) were estimated using Cox proportional hazards and time-varying hazard ratio (HR) models. Safety outcomes (grade 3 or 4 treatment-related adverse events) with NIVO+IPI and the comparators were compared. Results In the Cox proportional hazards analysis, NIVO+IPI showed improved OS compared with DAB+TRAM (HR = 0.53; 95% confidence interval [CI], 0.39-0.73), ENCO+BINI (HR = 0.60; CI, 0.42-0.85), and VEM+COBI (HR = 0.50; CI, 0.36-0.70) for the overall study period. In the time-varying analysis, NIVO+IPI was associated with significant improvements in OS and PFS compared with the BRAF/MEK inhibitors 12 months after treatment initiation. There were no significant differences between NIVO+IPI and BRAF/MEK inhibitor treatment from 0 to 12 months. Safety outcomes favored DAB+TRAM over NIVO+IPI, whereas NIVO+IPI was comparable to VEM+COBI. Conclusion Results of this MAIC demonstrated durable OS and PFS benefits for patients with BRAF-mutant advanced melanoma treated with NIVO+IPI compared with BRAF/MEK inhibitors, with the greatest benefits noted after 12 months.


INTRODUCTION
Patients with unresectable advanced (stage III/IV) melanoma have historically had limited treatment options and faced a poor prognosis [median overall survival (OS) with metastatic disease of <12 months]. 1,2 However, several treatments have been introduced over the past decade that have dramatically improved outcomes in this setting, especially for patients with BRAF V600emutant disease, who constitute approximately 50% of the metastatic cutaneous melanoma population. 3 First-line treatments approved by the US Food and Drug Administration (FDA) for patients with BRAF-mutant melanoma includedamong othersdthe immunotherapy combination of nivolumab [a programmed death (PD)-1 inhibitor] plus ipilimumab [a cytotoxic T-lymphocyte antigen-4 (CTLA-4) inhibitor; NIVOþIPI] and targeted therapy combinations with the BRAF/MEK inhibitors dabrafenib plus trametinib (DABþ TRAM), encorafenib plus binimetinib (ENCOþBINI), and vemurafenib plus cobimetinib (VEMþCOBI). 4 In the phase III CheckMate 067 study (NCT01844505), NIVOþIPI or NIVO alone was associated with significant and durable OS benefits compared with IPI alone in previously untreated patients with advanced melanoma at 5-year follow-up, regardless of BRAF-mutation status. 5,6 In the subgroup of patients treated with NIVOþIPI having tumors harboring BRAF mutations, the 5-year OS rate was 60%. 6 BRAF/MEK inhibitor combinations have also significantly improved long-term OS in previously untreated patients with BRAFmutant advanced melanoma, with recent results demonstrating 5-year OS rates of 34% for DABþTRAM 7 and 31% for VEMþCOBI. 8 Whether disease control can be maintained indefinitely, especially without continued treatment, remains to be determined.
Given that a substantial proportion of patients with metastatic melanoma have BRAF-mutant disease, there is tremendous value in exploring the relative efficacy of therapeutic options for this patient population. However, results from prospective randomized clinical trials (RCTs) comparing immunotherapy and targeted therapy have not yet been reported. Therefore, clinicians must rely on indirect comparisons to assess long-term outcomes with these treatments. Although indirect comparison of therapies across RCTs can be inherently biased because of differences in patient populations and study designs, statistical methods can reduce this bias and increase the precision of the results. A widely accepted statistical approach for estimating the relative efficacy of treatments from different trials is the matching-adjusted indirect comparison (MAIC). 9,10 In this approach, individual patient-level data (IPD) for a treatment of interest (index therapy) are compared with aggregate-level (summary) data for comparator therapies by adjusting for differences in patient populations between similarly designed trials. The MAIC methodology has been applied extensively in oncology to assess the effects of treatments for various tumor types. [11][12][13][14][15][16][17][18] The objective of this analysis was to evaluate the relative efficacy and safety of NIVOþIPI (index therapy) versus DABþTRAM, ENCOþBINI, and VEMþCOBI (comparators) in patients with BRAF-mutant advanced melanoma using MAICs. This analysis expands on a previously reported MAIC analysis that compared NIVOþIPI with DABþTRAM and VEMþCOBI in this patient population. 19

Evidence base
A systematic literature review (SLR) of published articles and medical congress abstracts was conducted to identify RCTs that enrolled patients with BRAF-mutant advanced melanoma treated with DABþTRAM, ENCOþBINI, or VEMþCOBI. Studies involving patients who were exposed to prior systemic therapy were not excluded. Study selection methods are described in the Appendix, and study selection criteria are listed in Supplementary Table S1 6 and were restricted to match the inclusion/exclusion criteria of the comparator trials. Baseline characteristics included in the MAIC were age (>55, >56, or >57 years, based on medians reported in the comparator trials), sex, Eastern Cooperative Oncology Group (ECOG) performance status (PS; 0 or !1), lactate dehydrogenase (LDH) level [above or below the upper limit of normal (ULN)], and metastatic stage (M1b or M1c using the American Joint Committee on Cancer staging system seventh edition 2 ). IPD for NIVOþIPI were weighted by adjusting for covariates known to impact treatment outcomes (effect modifiers). The impact of weighting was measured using Kish's effective sample size (ESS). 20 Treatment effects were estimated for the overall study period using a Cox proportional hazards model, which determined hazard ratios (HRs) and corresponding confidence intervals (CIs) for the OS and PFS KaplaneMeier curves. In the event that the proportional hazards assumption was violated, treatment effects were estimated for specific monthly time points and 12-month averages using a time-varying model, which determined HRs and corresponding 95% credible intervals (CrIs) using fractional polynomials. Models of varying complexity were tested and ranked according to deviance information criteria. Curve fit and extrapolation beyond the trial data were subsequently evaluated, and less complex models were used if they had a goodness of fit comparable to the more complex models. Relative safety was assessed by comparing the incidence of grade 3 or 4 treatment-related adverse events (TRAEs) between NIVOþIPI and the comparator therapies. The safety analysis was restricted to the weighted population with BRAF-mutant melanoma, and statistical significance was determined by calculating odds ratios (ORs) and corresponding 95% CrIs. All analyses were conducted using JAGS (v4.3.0) and programmed in R (v3.6.1).

Evidence base
The analysis set for the MAIC comprised IPD from Check-Mate 067 6 for NIVOþIPI and data from seven publications reporting the results of three comparator RCTs: COMBI-d/v (NCT01584648 and NCT01597908, respectively) for DABþ TRAM, 7,21 COLUMBUS (NCT01909453) for ENCOþBINI, 22 Check-Mate 067 and the comparator trials had similar study designs, and OS and PFS results were available from all the comparator trials. Grade 3 or 4 TRAEs were not evaluated in the COLUMBUS trial 22-24 and therefore were not available for ENCOþBINI. Although safety outcomes were evaluated in the COMBI-v trial, only all-cause and select adverse event (AE) rates were reported 8 ; therefore, safety results (grade 3 or 4 TRAEs) for DABþTRAM were obtained only from the COMBI-d trial. 21 Baseline characteristics of patients treated with NIVOþIPI or the BRAF/MEK inhibitors in the RCTs are shown in Table 1. CheckMate 067 enrolled patients regardless of BRAF-mutation status. 6 Of the 314 patients treated with NIVOþIPI in CheckMate 067, 103 had known BRAF-mutant disease and were included in this analysis. The DABþTRAM, ENCOþBINI, and VEMþCOBI groups from the comparator trials included 563, 192, and 247 patients, respectively. After weighting the data, ESSs for the comparisons of NIVOþIPI with DABþTRAM, ENCOþBINI, and VEMþCOBI were 91.68, 93.81, and 91.69, respectively. Baseline characteristics of the subgroup of patients with BRAF-mutant disease who received NIVOþIPI in CheckMate 067 were similar to those of patients receiving DABþTRAM, ENCOþBINI, or VEMþCOBI in the comparator trials. However, compared with the comparator trials, CheckMate 067 had larger percentages of patients aged over 55 years and with an ECOG PS of 0, as well as a smaller percentage with stage M1b or M1c disease.  (Figures 1 and 2).

Analysis of OS
Cox proportional hazards model analysis of OS demonstrated statistically significant reductions in the risk of death of patients treated with NIVOþIPI compared with those treated with DABþTRAM (HR ¼ 0.53; 95% CI, 0.39-0.73), ENCOþBINI (HR ¼ 0.60; 95% CI, 0.42-0.85), or VEMþCOBI (HR ¼ 0.50; 95% CI, 0.36-0.70) ( Table 2). KaplaneMeier OS curves for NIVOþIPI (weighted) and for each of the three comparators are shown in Figure 1. However, the proportional hazards assumption was violated in all of the comparisons in the OS analysis, and a time-varying analysis was therefore performed.
In the time-varying analysis, the risk for death was similar between NIVOþIPI and BRAF/MEK inhibitor treatment from 0 to 12 months (Table 2). However, NIVOþIPI treatment was associated with a significant reduction in the risk of death after 12 months compared with all three comparators. Similar trends were observed with a time-varying model that estimated HRs at discreet 3-month or 6-month landmarks, in which HRs favored NIVOþIPI over all three comparators beginning at 12 months and at every time point thereafter (Supplementary Table S3 the COLUMBUS trial [22][23][24] at the time of this analysis, OS outcomes for ENCOþBINI were extrapolated to 60 months.

Analysis of PFS
Cox proportional hazards model analysis of PFS demonstrated statistically significant reductions in the risk of progression or death for patients treated with NIVOþIPI compared with those treated with DABþTRAM (HR ¼ 0.73; 95% CI, 0.55-0.96) and VEMþCOBI (HR ¼ 0.74; 95% CI, 0.55-0.99) ( Table 3). However, the difference in the risk of progression or death between NIVOþIPI and ENCOþBINI was not statistically significant (HR ¼ 0.88; 95% CI, 0.63-1.21). KaplaneMeier PFS curves for NIVOþIPI (weighted) and for each of the three comparators are shown in Figure 2. A time-varying analysis was performed because the proportional hazards assumption was violated for all three comparisons in the PFS analysis. The time-varying analysis showed no significant difference in the reduction in risk of progression or death based on HRs averaged over the time period between 0 and 12 months (Table 3). However, NIVOþIPI was associated with significant reductions in the risk of progression or death after 12 months compared with all three comparators. The time-varying analysis estimating monthly HRs showed similar trends, with a significant difference favoring NIVOþIPI over each of the three comparators beginning at 12 months and at every time point thereafter (Supplementary Table S4, available at https://doi.org/10. 1016/j.esmoop.2021.100050). As in the OS analysis, PFS outcomes for ENCOþBINI were extrapolated to 60 months.

Analysis of safety
Among patients with BRAF-mutant melanoma treated with NIVOþIPI in CheckMate 067, 66% (68/103) experienced a grade 3 or 4 TRAE. After weighting the data, ESSs were 83.3 and 91.7 for the safety comparisons between NIVOþIPI and DABþTRAM (from the COMBI-d trial) 21 or VEMþCOBI (from the coBRIM trial), 8

DISCUSSION
The introduction of immunotherapy and targeted therapy has transformed the treatment of patients with BRAFmutant advanced melanoma, but there is a lack of comparative data from RCTs, making treatment decisions difficult. Consequently, a MAIC analysis was conducted to estimate the relative efficacy and safety of the immunotherapy combination of NIVOþIPI compared with the targeted therapy combinations of DABþTRAM, ENCOþBINI, and VEMþCOBI. This MAIC used the longest follow-up data available for NIVOþIPI treatment (5-year minimum followup results from CheckMate 067 6 ) and the most current data available from the comparator trials at the time of the analysis. 7,8,[21][22][23][24][25] This MAIC analysis demonstrated OS and PFS benefits with NIVOþIPI compared with BRAF/MEK inhibitors among patients with BRAF-mutant advanced melanoma. NIVOþIPI was associated with significant OS benefits compared with DABþTRAM, ENCOþBINI, and VEMþCOBI, and significant PFS benefits compared with DABþTRAM and VEMþCOBI, while the overall treatment effect for PFS in comparison with ENCOþBINI was not statistically significant. OS and PFS benefits were durable, with the magnitude of the treatment effects increasing over time. In the first 12 months after treatment initiation, outcomes for NIVOþIPI and the BRAF/ MEK inhibitors were similar. However, beginning at 12 months, statistically significant improvement in OS and PFS were evident with NIVOþIPI compared with all of the BRAF/MEK inhibitor combinations. In particular, PFS curves for NIVOþIPI were essentially flat after 18 months, suggesting durability of response, while continuing to decrease for the BRAF/MEK inhibitors.
In the safety analysis involving grade 3 or 4 TRAEs, DABþTRAM appeared to be better tolerated than  NIVOþIPI, whereas NIVOþIPI and VEMþCOBI demonstrated comparable safety outcomes. Using TRAEs to assess comparative safety, however, may be challenging given that commonly reported TRAEs can vary across therapeutic classes with regard to AE type, onset, duration, and response to management strategies. For example, BRAF/ MEK inhibitors are usually associated with moderate TRAEs that resolve or diminish rapidly after dose reduction or treatment interruption. 26 In contrast, immunotherapies have a propensity to elicit immune-mediated adverse events, which may have a prolonged duration in some instances and often require treatment with corticosteroids. [27][28][29] It is also important to note that different adverse event categories may differentially impact quality of life, impeding safety comparisons across therapeutic classes. Therefore, the overall safety profile of a treatment, along with efficacy benefits, should be considered when selecting therapeutic options for patients with BRAF-mutant advanced melanoma. This analysis expands on a previously reported MAIC analysis comparing NIVOþIPI with targeted therapy, specifically DABþTRAM and VEMþCOBI, in patients with BRAF-mutant advanced melanoma. 19 In that earlier analysis, data for NIVOþIPI were pooled from CheckMate 067 (4-year follow-up) and the phase II CheckMate 069 study, and data for DABþTRAM and VEMþCOBI were derived from the COMBI-d/v and coBRIM trials, respectively. Results from that analysis were similar to the results presented here, with NIVOþIPI being associated with significantly improved OS compared with DABþTRAM (HR ¼ 0.64; 95% CI, 0.46-0.89) or VEMþCOBI (HR ¼ 0.56; 95% CI, 0.36-0.89) in Cox model analyses. As in the current study, OS outcomes were similar between NIVOþIPI and the comparators early in treatment (0-12 months), with benefits associated with NIVOþIPI emerging after 12 months of treatment. Similar results were noted with PFS. As in the current study, grade 3 or 4 TRAEs were reported significantly more often with NIVOþIPI than with DABþTRAM (54% versus 32%; OR ¼ 2.6; 95% CI, 1.8-3.6) but similarly with NIVOþIPI and VEMþCOBI (54% versus 59%; OR ¼ 0.8; 95% CI, 0.6-1.1). Findings from the current MAIC analysis augment the results of the prior analysis by involving extended follow-up times, a newer BRAF/MEK inhibitor combination (ENCOþBINI), NIVOþIPI data only from CheckMate 067 (which reported results for a minimum follow-up of 5 years), 6 and restriction of the safety analysis to patients with BRAF-mutant disease (the previous analysis did not involve restriction, only weighting).

Strengths and limitations
The strengths of this analysis rest primarily in the application of IPD from a large, robust, RCT (CheckMate 067), with comparator evidence identified through an extensive SLR process. This analysis was based on good practices in the field of indirect comparison research, with application of extensive restrictions and weighting of IPD to establish a comparable patient cohort. 9 As with any indirect comparison approach, the MAIC analysis reported here was associated with limitations. Despite population restrictions and weighting, this analysis accounted only for known, measured, and reported effect modifiers, and unobserved differences between the trials may have confounded the results. However, the similarity of the study designs provided some confidence that the observed outcomes were not simply the result of differences between studies. In addition, the statistical power of MAIC analyses can be diminished if the ESS becomes small when too few patients are assigned a high weight. Nonetheless, the impact of weighting by ESS in this analysis suggested that the comparisons were sufficiently powered. This analysis may also have been limited by the lack of consideration of other efficacy outcomes, such as duration of response and survival among patients reaching complete and/or partial tumor response. The current analysis also did not evaluate comparisons by patient subgroups of interest (e.g. those with elevated baseline LDH levels) due to limited patient numbers, which precluded a robust subgroup analysis. Moreover, results of the comparison of NIVOþIPI and ENCOþBINI should be interpreted with caution as only 4-year follow-up data were available from the COLUMBUS trial at the time of this analysis, necessitating extrapolation of OS and PFS outcomes to 5 years. However, because the 4-year OS and PFS data from COLUMBUS were mature, results of the comparison are unlikely to change with the use of longer follow-up data. Furthermore, the weighted NIVOþIPI OS curves in all three comparisons fell slightly below the observed OS curve, suggesting that patients enrolled in the targeted therapy trials may have had a worse prognosis than those enrolled in CheckMate 067. This finding underscores the bias involved in making naive (unadjusted) comparisons across trials. It is also unclear how the results of this MAIC should be interpreted relative to outcomes with the combination of atezolizumab (a programmed deathligand 1 inhibitor) and VEMþCOBI, which was recently approved as a first-line treatment of BRAF-mutant advanced melanoma. 30 Data for the combination of atezolizumab and VEMþCOBI were not available at the time of this analysis and did not appear to be mature; future MAICs should include this new therapeutic option.
An important limitation of indirect comparisons of cancer treatments is the inability to control for differences in the use of subsequent therapy between trials, which may influence OS outcomes unequally. Selection of subsequent therapy in oncology trials may be based on several factors, including the availability of specific treatment options at the time of the study and at the geographic location of the study site. Geographic differences existed between the trials included in this analysis. Patients enrolled in Europe comprised 74% and 81% of the coBRIM 25 and COLUMBUS 22,23 trial populations, respectively, compared with 56% of the NIVOþIPI treatment arm in CheckMate 067 (geographic distribution information was not available for COMBI-v/d). 29 Perhaps because of these geographic differences, subsequent therapy differed between the comparator trials and CheckMate 067. In the COMBI-v/d, 7 COLUMBUS, 31 and coBRIM 32 trials, immunotherapy was the most prevalent subsequent therapy. Although the start of enrollment in the COMBI-v/d trials preceded that of the coBRIM and COLUMBUS trials by approximately 1 year, use of subsequent treatment was similar in these trials. In contrast, the most frequently used subsequent therapy among patients with BRAF-mutant melanoma in CheckMate 067 was BRAF inhibitors (data on file). Therefore, this MAIC analysis should not be considered a true comparison of treatment sequencing but rather a comparison of first-line treatment choices in which OS may have been affected by the use of subsequent therapy. However, the inability to control for differences in the use of subsequent therapy may have been offset by the use of PFS outcomes in this analysis. PFS is considered to be less affected by subsequent therapy than OS, and the PFS results in this analysis demonstrated statistically significant differences favoring NIVOþIPI over the comparators after 12 months.

CONCLUSION
In the absence of head-to-head evidence from RCTs, this MAIC provides insights into the comparative efficacy and safety of therapies approved by the US FDA for patients with BRAF-mutant advanced melanoma. This analysis demonstrated durable OS and PFS benefits among patients with BRAF-mutant melanoma treated with NIVOþIPI compared with those treated with DABþTRAM, ENCOþBINI, or VEMþCOBI. These benefits increased over time, with significant treatment effects with NIVOþIPI emerging after 12 months, suggesting that NIVOþIPI had a more durable benefit than the targeted therapies. These findings supplement the long-term efficacy and safety results reported in RCTs and may provide clinicians with additional information relevant to the selection of treatments for BRAF-mutant advanced melanoma. Results of the ongoing phase III DREAMseq trial (NCT02224781) and the phase II SECOMBIT trial (NCT02631447), which are comparing sequential regimens of NIVOþIPI plus DABþTRAM or ENCOþBINI, respectively, may confirm the optimal sequence of initial therapy for patients with BRAFmutant advanced melanoma.