First-line BRAF/MEK inhibitors versus anti-PD-1 monotherapy in BRAFV600-mutant advanced melanoma patients: a propensity-matched survival analysis

Background Anti-PD-1 antibodies and BRAF/MEK inhibitors are the two main groups of systemic therapy in the treatment of BRAFV600-mutant advanced melanoma. Until now, data are inconclusive on which therapy to use as first-line treatment. The aim of this study was to use propensity score matching to compare first-line anti-PD-1 monotherapy vs. BRAF/MEK inhibitors in advanced BRAFV600-mutant melanoma patients. Methods We selected patients diagnosed between 2014 and 2017 with advanced melanoma and a known BRAFV600-mutation treated with first-line BRAF/MEK inhibitors or anti-PD-1 antibodies, registered in the Dutch Melanoma Treatment Registry. Patients were matched based on their propensity scores using the nearest neighbour and the optimal matching method. Results Between 2014 and 2017, a total of 330 and 254 advanced melanoma patients received BRAF/MEK inhibitors and anti-PD-1 monotherapy as first-line systemic therapy. In the matched cohort, patients receiving anti-PD-1 antibodies as a first-line treatment had a higher median and 2-year overall survival compared to patients treated with first-line BRAF/MEK inhibitors, 42.3 months (95% CI: 37.3-NE) vs. 19.8 months (95% CI: 16.7–24.3) and 65.4% (95% CI: 58.1–73.6) vs. 41.7% (95% CI: 34.2–51.0). Conclusions Our data suggest that in the matched BRAFV600-mutant advanced melanoma patients, anti-PD-1 monotherapy is the preferred first-line treatment in patients with relatively favourable patient and tumour characteristics.

As BRAF/MEK inhibitors have a rapid onset and a high overall response rate (ORR), they are often considered the preferred primary treatment for patients with extensive (symptomatic) disease and a BRAF V600 -mutation. However, in such patients, the response duration is relatively short due to acquired resistance. 7 As time to response for immunotherapy is generally longer and ORR lower, checkpoint inhibition is usually not the preferred treatment for BRAF-mutant patients with poor prognostic factors such as a poor performance score or elevated LDH. 8 Therefore, the perception that checkpoint inhibition leads to more durable responses could at least partially be explained by the fact that patients with better www.nature.com/bjc prognostic factors are being treated with immunotherapy. Although interstudy comparison suggests that immunotherapy results in more durable responses as compared to targeted therapy, 9 these results need to be interpreted with caution due to the potential biases associated with this type of analysis.
Due to the confounding by indication bias, a head-to-head comparison between both groups, based on real-world data is difficult. Propensity score matching, a statistical method developed by Rosenbaum and Rubins, 10 accounts for the unequal distribution of confounders. A propensity score is the chance of receiving the treatment, based on measured confounders such as age and tumour stage. 11,12 This method matches patients based on their propensity score, creating two groups with similar distribution of confounders.
In this study, we (1) compare OS of BRAF/MEK inhibitors and anti-PD-1 monotherapy in BRAF-mutant advanced melanoma patients, and (2) use propensity score matching to determine the overall survival in the BRAF/MEK inhibitors vs. anti-PD-1 monotherapy groups after matching.

Study design and population
This longitudinal observational study used data from the Dutch Melanoma Treatment Registry (DMTR), a population-based, prospective registry capturing all patients with unresectable stage IIIc and IV melanoma in the Netherlands. Since the 1 July 2013, all patients with unresectable stage IIIC or stage IV melanoma seen in one of the 14 melanoma centres have been registered in the DMTR. Data are prospectively registered from diagnosis until death. Patients are followed every three months for changes in disease status, patients' characteristics or treatment characteristics by an independent trained data manager. A detailed description of the DMTR setup has been published by Jochems et al. 13 All patients of 18 years and older, diagnosed between 2014 and 2017 (dataset cut-off date was 01-08-2019) with unresectable stage IIIC and IV melanoma, with a known BRAF V600 -mutation who used anti-PD-1 monotherapy (pembrolizumab or nivolumab) or BRAF/MEK inhibitors (dabrafenib + trametinib or vemurafenib + cobimetinib) as a first-line systemic therapy, were selected for analysis. Combination therapies of encorafenib + binimetinib and ipilimumab + nivolumab were not yet available during this time period. Line of therapy was defined as the start of a new drug class. Patients with uveal or mucosal melanoma were excluded from this analysis.

Statistical analysis
Baseline patient-and disease characteristics of BRAF V600 -mutant patients were analysed using descriptive statistics. Categorical variables were compared using the Chi-square test. Overall survival (OS) was defined as time from start systemic treatment until death from any cause with corresponding two-sided 95% confidence intervals and was estimated using the Kaplan-Meier method. TTNT was defined as the time until next treatment of a new drug class or death from any cause. OS and TTNT between subgroups was compared using log-rank tests for categorical variables. Patients alive or lost to follow-up were right-censored at the time of last registered contact.
In order to reduce confounding by indication, propensity score matching was used. As a first step, a multivariable logistic regression analysis was used to estimate propensity scores. Covariates used for the multivariable logistic regression were age at diagnosis (<75, ≥75), baseline ECOG-performance status (0-1 and ≥2), baseline lactate dehydrogenase (LDH) (normal, 250-500 U/L and >500 U/L), stage at diagnosis (unresectable IIIc, IV-M1a, M1b and IV-M1c), distant metastasis (<3 organ sites and ≥3 organ sites involved), brain metastases (none, asymptomatic, and symptomatic) and liver metastases (yes/no) and use of immunomodulating agents (corticosteroids, azathioprine and interferon). Covariates were chosen based on clinical practice and previous research identifying these prognostic risk factors. 14,15 A Monte Carlo simulation study has demonstrated that his method of selecting covariates for propensity score matching works very well. 16 A multivariable Cox proportional hazards model was used to assess the association of prognostic factors with OS. Prognostic factors estimated were age at diagnosis (<75, ≥75), baseline ECOG-performance status (0-1 and ≥2), baseline lactate dehydrogenase (LDH; normal, 250-500 U/L and >500 U/L), distant metastases (<3 organ sites and ≥3 organ sites involved), brain metastases (none, asymptomatic and symptomatic) and liver metastases (yes/no). The proportional hazards assumption was tested with the scaled Schoenfeld residuals. None of the covariates violated the proportional hazards assumption.
For matching purposes, 1:1 nearest neighbour matching with a calliper set of 0.001 was used for propensity score matching. To assess the quality of matching, standardised mean differences for all covariates were calculated. Standardised mean differences of <0.1 were assumed to be negligible. After matching, a univariable Cox proportional hazards model with a robust variance estimator to account for the matched nature of the sample was used to estimate the relative change in hazard of survival. 17,18 To address potential biases due to incomplete matching, we compared our results to a second matched sample set using optimal matching. Optimal matching forms matched pairs to minimise the average within-pair difference in propensity scores. 19 We used both matching techniques in order to observe the influence of wellbalanced prognostic factors across both groups in the nearest neighbour-matched group on OS and TTNT compared to the optimal matching method. Due to the matched nature, the use of a log-rank test comparing the survival curves will likely result in type I error rates that are artificially low, 20 stratified log-rank tests were used to compare the survival curves of the matched samples. 18 Statistical software used was R (version 3.5.2; packages car, tidyverse, survival and matchit).
In   Table 4). Kaplan-Meier estimates of the optimal matched cohort are shown in Fig. 4. The estimated HR with a robust variance estimator was 0.43 (95% CI: 0.34-0.56). Thus, treatment with first-line anti-PD-1 reduced the hazard of death by 57% in the optimal matched sample.

DISCUSSION
This report, based on real-world population-based data, describes the overall survival of BRAF V600 -mutant advanced melanoma patients after using first-line treatment with BRAF-MEK inhibitors or anti-PD-1 monotherapy. In total 310 patients were matched, resulting in 155 matched pairs. In the matched cohort, patients treated with anti-PD-1 monotherapy as a first-line treatment showed a higher 2-year survival compared to patients treated with first-line BRAF-MEK. Median OS in the anti-PD-1 monotherapy cohort was 42.3 months while the median OS was 19.8 months in patients receiving BRAF/MEK inhibitors as first-line treatment.
In the nearest neighbour-matched cohort, patients treated with anti-PD-1 as a first-line treatment have higher OS compared to patients treated with BRAF/MEK inhibitors as a first-line treatment. There were also patients who could not be matched. Overall, we matched patients with relatively good prognosis factors who   In the optimal matching method, we matched all patients receiving first-line anti-PD-1, resulting in lower OS for BRAF/MEK inhibitors as sicker patients were matched, which could, in turn, lead to residual confounding. 18 Consequently, we cannot compare the efficacy of BRAF/MEK inhibitors or anti-PD-1 in the unmatched group. In BRAF-mutant patients with a very poor prognosis, BRAF-MEK inhibitors might be the preferred choice of first-line treatment as this treatment results in quick antitumour response. 21 This study suggests that in the matched cohort, consisting of patients with relative favourable patient and tumour characteristics, anti-PD-1 monotherapy is the first-line treatment option of choice. However, the results need to be confirmed in randomised clinical trials to assess optimal front-line-and sequence therapy. Currently, several clinical trials are investigating the optimal frontline treatment of immunotherapy and targeted therapy. The COWBOY study evaluates a planned sequence vemurafenib and cobimetinib, followed by ipilimumab and nivolumab vs. ipilimumab and nivolumab (NCT02968303). 22 The DREAMseq study (NCT02224781) evaluates a sequence of dabrafenib and trametinib until progression, followed by ipilimumab and nivolumab or ipilimumab and nivolumab until progression followed by dabrafenib and trametinib. The SECOMBIT (NCT02631447) evaluates the sequence of encorafenib plus binimetinib followed by ipilimumab plus nivolumab or the opposite sequence or encorafenib plus binimetinib for 8 weeks followed by ipilimumab plus nivolumab until progression followed by encorafenib plus binimetinib. 23,24 These studies are expected to be completed by the end of 2021 and 2022.  TTNT and OS estimates shown as survival rates at 6,12 and 24 months and median overall survival stratified for first-line therapy (BRAF/MEK inhibitors vs. anti-PD-1 monotherapy). NE non estimable.
In Phase 3 RCT's, 1-year survival was 74.1% for pembrolizumab and 72.9% for nivolumab. 4,25 Comparing the results of this study with the randomised clinical trials of both anti-PD-1 ligands, we find a higher 1-year survival (81.7%, 95% CI: 77.0-86.6). This is remarkable because we would expect 1-year survival of patients treated in clinical practice to be similar or worse due to unfavourable patient and tumour characteristics. However, in the CHECKMATE-066 study, only BRAF wild-type patients were included and the KEYNOTE-006 included only 35% of BRAF-mutant patients. These patients lack treatment options with BRAF/MEK inhibitors,     Another real-world study investigating the outcomes of first-line immunotherapy and targeted therapy was performed by Luke et al. 27 Opposed to our current study, they find a lack of survival benefit for patients receiving first-line immunotherapy, despite a favourable imbalance of prognostic factors. These differences could be explained by the percentage of M1c patients (77.4%) in our study vs. 59.7% and 51.6% in the study by Luke et al. A direct comparison between both studies is hard due to differences in patient and tumour characteristics, study design and geographic location.
There are limitations to this study. First, this study did not include patients who received ipilimumab monotherapy or ipilimumab combined with nivolumab. Ipilimumab monotherapy is no longer given as first-line treatment since anti-PD-1 has proven superior efficacy. 28 However, combination therapy of ipilimumab and nivolumab has proven to be effective in metastatic melanoma patients in the CHECKMATE-067 trial. 29 We did not have enough patients who received combination therapy of ipilimumab and nivolumab to make a reliable analysis of this group. Second, the observational nature of the DMTR may have introduced bias. However, since the start of this registry, independent data managers have been trained annually and data are checked and confirmed by treating physicians. The online registration platform warns data managers of inconsistent or missing values in the registry. Previous studies have shown the quality and validity of the Dutch population-based registries. 30 We therefore argue that the data used in this study is of high quality. The third limitation is that we could not assess how subsequent therapies impact the outcome. By studying patients receiving second-line treatment, we would introduce a selection bias. Only patients fit enough would be able to receive second-line treatment. Studying the impact of subsequent therapies on outcomes would require a randomised controlled trial with several predefined arms. In the real-world, choice of type of subsequent therapy depends on patient and tumour characteristics as well as preference of the oncologist.
Based on the matched cohort, our data suggest that in BRAF V600 -mutant cutaneous metastatic melanoma patients, anti-PD-1 monotherapy should be the preferred treatment if a quick antitumour response, for example for symptom relief, is not the primary aim of treatment. Additional results from RCTs are necessary to confirm these results to determine optimal frontline and sequential therapy for BRAF-mutant cutaneous metastatic melanoma patients.

AUTHOR CONTRIBUTIONS
J.B. performed the analyses and wrote the manuscript. M.W. was responsible for the study conception and design, analysis and interpretation of data and drafting of the manuscript. D.H. was responsible for the study conception and design, analysis and interpretation of data and drafting of the manuscript. W.B. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. J.H. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. C.B. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. M.A. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. F.B. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript, performed critical revision and final approval of the manuscript. J.G. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. G.H. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. E.K. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. D.P. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. R.R. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. K.S. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. A.T. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. A.V. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. G.V. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. M.B. performed data acquisition, study conception, helped with the analysis and interpretation of the data, drafted the manuscript and performed critical revision and final approval of the manuscript. A.E. was responsible for the study conception and design, analysis and interpretation of data and drafting of the manuscript.

ADDITIONAL INFORMATION
Ethics approval and consent to participate In compliance with Dutch regulations, the DMTR was approved by a medical ethical committee (METC Leiden University Medical Center, 2013) and is not considered subject to the Medical Research Involving Human Subjects Act.
Data availability The datasets generated during and/or analysed during the current study are not publicly available due to privacy regulations in the Netherlands but are available from the corresponding author on reasonable request. Supplementary information is available for this paper at https://doi.org/10.1038/ s41416-020-01229-1.
Note This work is published under the standard license to publish agreement. After 12 months the work will become freely available and the license terms will switch to a Creative Commons Attribution 4.0 International (CC BY 4.0).
Publisher's note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.