An observational study of drug utilization and associated outcomes among adult patients diagnosed with BRAF‐mutant advanced melanoma treated with first‐line anti‐PD‐1 monotherapies or BRAF/MEK inhibitors in a community‐based oncology setting

Abstract Introduction Anti‐PD‐1 monotherapies (aPD‐1) and BRAF/MEK inhibitors (BRAF/MEKi) changed the BRAF‐mutant advanced melanoma treatment landscape. This study aimed to improve the understanding of real‐world treatment patterns and optimal treatment sequence. Methods This was a retrospective study of BRAF‐mutant advanced melanoma patients who initiated 1L aPD‐1 or BRAF/MEKi in the US Oncology Network between 1 January 2014 and 31 December 2017, followed through 31 December 2018. Patient and treatment characteristics were assessed descriptively, with Kaplan‐Meier methods used for time‐to‐event endpoints. As the primary analysis, overall survival (OS) and physician‐assessed progression‐free survival (rwPFS) were evaluated with Cox proportional hazard regression models and propensity score matching (n = 49). Results A total of 224 patients were included (median age 61 years, 62.9% male, 89.7% white): 36.2% received aPD‐1 and 63.8% BRAF/MEKi. Median OS and rwPFS were longer among aPD‐1 vs BRAF/MEKi patients (OS: not reached vs 13.9 months, log‐rank P = .0169; rwPFS: 7.6 vs 6.5 months, log‐rank P = .0144). Receipt of aPD‐1 was associated with improved OS (HR = 0.602 vs BRAF/MEKi [95%CI 0.382‐0.949]; P = .0287). Among patients without an event within 6 months of 1L initiation, receipt of aPD‐1 was associated with a decreased risk of progression or death from 6 months onwards (HR = 0.228 [95%CI 0.106‐0.493]; P = .0002). This association was not observed among patients within 6 months of 1L initiation (HR = 1.146; 95% CI 0.755‐1.738). Results from the propensity score‐matched pairs were consistent with these trends. Conclusion These results suggest a clinical benefit of 1L aPD‐1 compared to BRAF/MEKi after 6 months of treatment for BRAF‐mutant advanced melanoma. Future research should explore factors associated with early progression and their relationship with clinical outcomes.

Moser et al (2019) retrospectively assessed electronic health record (EHR) data of patients with BRAF-mutant melanoma who received first-line (1L) treatment with aPD-1, BRAF/MEKi, or nivolumab/ipilimumab. 18  specifically with 1L therapies as opposed to all therapies combined. While studies of treatment sequencing are in progress, these data are not yet available to support clinical decision-making. The aim of this study was therefore to assess treatment patterns and outcomes among patients with BRAF-mutant advanced melanoma who initiated 1L treatment with aPD-1 or BRAF/MEKi in a large network of community oncology clinics. Patients were followed as they advanced through their treatment sequences.

| Study design and data sources
This was a retrospective cohort study of the US adult patients with BRAF-mutant advanced melanoma who initiated 1L treatment with aPD-1 or BRAF/MEKi combination therapy between 01 January 2014 and 31 December 2017 from practices in the US Oncology Network (USON). Patients were followed until their last record, death date, or end of the study period (31 December 2018), whichever occurred first. This was an extension of previous studies of advanced melanoma patients who initiated 1L treatment within the USON. [19][20][21] The combination of nivolumab/ipilimumab was granted accelerated approval for advanced melanoma patients with BRAF mutations in September 2015 and, regardless of BRAF status, in January 2016. 22 Given the period of observation, the decision was made to focus this study on patients who received 1L aPD-1 or BRAF/MEKi due to the limited number of evaluable patients who received nivolumab/ipilimumab during the time period of analysis. 19 The USON is affiliated with approximately 1400 physicians in more than 470 sites of care across 25 states in the US, representing approximately 12% of the US patients newly diagnosed with cancer. 23 Data were obtained via programmatic database extraction from the EHR of the USON, iKnowMed. iKnowMed is an oncology-specific EHR system that captures outpatient practice encounter history for patients receiving treatment within the USON, including laboratory tests, diagnosis, administration of infused therapies, prescription of oral therapies, staging, comorbidities, and performance status. This product has been implemented across the USON. Data were first extracted from structured EHR fields, with chart review performed to capture and/or verify other information of interest, including physician-assessed response, patients' treatment histories, and dates of diagnoses. Supplemental vital status information was provided by the Social Security Administration's Limited Access Death Master File.
Eligible patients were at least 18 years of age at advanced melanoma diagnosis and had at least two visits within the USON or a record of death during the study observation period. Patients were excluded if they were enrolled in clinical trials or had another primary cancer diagnosis during the study period (with the exception of basal cell carcinoma, squamous cell carcinoma, bladder carcinoma in situ, and cervical carcinoma in situ).
The study was reviewed and granted an exception and waiver of consent by the US Oncology, Inc Institutional Review Board.

| Statistical analysis
Patient demographic, clinical, and treatment characteristics were assessed descriptively, with the Kaplan-Meier method used to assess time-to-event endpoints, including OS, physician-assessed PFS (rwPFS), time to treatment discontinuation (TTD), time to next treatment (TTNT), and duration of physician-assessed response (DOR).
OS was defined as the duration from 1L treatment initiation until death from any cause, with censoring of patients who did not die by the end of the study period. TTD was defined as the duration from 1L treatment initiation to discontinuation for any reason, with censoring of patients who did not discontinue 1L treatment by the end of the study period. Date of discontinuation was captured through chart review as the stop date explicitly documented by a provider (for oral therapies), last administration date (for infused therapies), date of hospice admission, or death date. TTNT was defined as the duration from 1L treatment initiation to the start of a new treatment, with censoring of patients who did not initiate a new treatment by the end of the study period. rwPFS was defined as the duration from 1L treatment initiation until either physician-documented disease progression or death, with censoring of patients who did not die or experience physician-documented disease progression by the end of the study period. DOR was defined as the duration from first documented response to 1L treatment initiation until date of first documented disease progression, with censoring of patients who did not experience response followed by disease progression by the end of the study period. For all time-to-event analyses, patients who did not experience the event were censored at the end of the study period, last visit date recorded, or death, whichever came first.
The primary objective of this study was to use multivariable Cox proportional hazard regression models and propensity score matching to adjust for underlying differences between the aPD-1 and BRAF/MEKi cohorts that may have influenced OS and rwPFS. First, univariate and multivariable Cox proportional hazard regression models were used to identify independent risk factors for OS and rwPFS. To construct these models, all baseline and 1L treatment covariates were included in univariate models, and a stepwise selection approach was used to identify covariates for the multivariable models, with P ≤ .25 for entry and ≤.15 for retention.
Next, a propensity score matching technique was used to reduce confounding of variables that may have influenced outcomes. To balance the cohorts for this analysis, patients initiating BRAF/MEKi prior to 01 September 2014 were removed to align with the first FDA approval date of pembrolizumab for patients with melanoma, which was an accelerated approval for patients no longer responding to other therapies. 24 Potential confounders included in the model were the following: age at 1L initiation, practice region, presence of other metastases, albumin result at 1L initiation, presence of bone metastases, radiation prior to 1L initiation, Deyo-adapted Charlson score, presence of brain metastases, smoking history, Eastern Cooperative Oncology Group (ECOG) at 1L initiation, presence of liver metastases, stage at initial diagnosis, gender, presence of lung metastases, and surgical resection prior to 1L initiation.
A combination of univariate logistic regression results, consideration of clinically relevant variables, and stepwise selection was used to determine variables for inclusion into the multivariable logistic regression model. The multivariable logistic regression model using the chosen covariates was used to generate the propensity score. Once the score was generated, a greedy matching algorithm was used to pair up patients from both treatment cohorts. Attrition was expected and the remaining cohorts were verified to be well balanced between clinically relevant variables that may have otherwise confounded outcomes.
An alpha level of .05 was the primary criterion for statistical significance. Analyses were conducted using SAS ® 9.4 (SAS Institute Inc).
Median age at diagnosis was 61 years, 62.9% were male, 89.7% were white, 48.2% had Stage IV disease, and 63.4% had an ECOG performance status 0-1 (Table 1). A higher proportion of aPD-1 patients had ECOG 0-1 performance status compared with the BRAF/MEKi group. Most patients in both treatment groups had M1c metastatic status, with higher proportions of the 1L BRAF/MEKi group having documented lung, brain, and liver metastases. Lactate dehydrogenase (LDH) status as 1L treatment initiation was recorded as being "elevated" in the EHR among 18.5% of aPD-1 patients and 26.6% of BRAF/MEKi patients.
KIT, NRAS, and PD-L1 status were not documented for the majority of patients in both treatment groups (Table 1). A higher proportion of the 1L BRAF/MEKi group lacked documentation of PD-L1 status (93.0% vs 81.5%).
Across the study population, 87.5% discontinued their 1L treatment; a higher proportion of patients who received 1L BRAF/MEKi discontinued compared with those who received 1L aPD-1 (93.7% vs 76.5%, respectively, P = .0002; Table 1). The top two discontinuation reasons in the aPD-1 group were disease progression (45.2%) and treatment-related toxicities (11.3%). Similarly, the two most common discontinuation reasons in the BRAF/MEKi group were disease progression (38.1%) and treatment-related toxicities (14.2%). Discontinuation of 1L due to death occurred among 3.2% of the aPD-1 group and 10.4% of the BRAF/MEKi group.
In total, 6.2% of the aPD-1 group and 24.5% of the BRAF/ MEKi group advanced to 3L (Table 1).
Median TTNT from 1L initiation was significantly longer among the aPD-1 group compared with the BRAF/ MEKi group ( ; P = .0003) was significantly associated with an increased risk of progression or death (Table 4). Among patients without an event within the first 6 months of 1L treatment initiation, receipt of aPD-1 was associated with a decreased risk of progression or death from 6 months onwards (HR = 0.228 [95%CI 0.106-0.493]; P = .0002). No statistically significant difference between treatment groups was observed in the risk of disease progression or death within 6 months of 1L treatment initiation (HR = 1.146 [95% CI 0.755-1.738] P = .522).

| Propensity score analysis
After propensity score matching, 49 matched pairs were selected for analysis. Standardized differences of baseline covariates before and after matching are presented in Table 5 and baseline characteristics of the matched pairs are presented in Table 6. The median OS for those receiving aPD-1 vs BRAF/MEKi was 27.3 months (95%CI 11.4-NR) and 13.9 months (95%CI 11.8-36.5), respectively. The median rwPFS was 5.1 months (95%CI 2.6-8.1) and 6.8 months (95%CI 5.2-10.6), respectively. A multivariable Cox regression analysis was performed: older age at 1L initiation (HR = 1.038 per year increase; 95%CI 1.011-1.067; P = .0065) and low albumin (HR = 2.724; 95%CI 1.358-5.46; P = .0048) were statistically associated with an increased risk of death. Among patients who had not died or experienced progression within 10 months of 1L treatment initiation, receipt of aPD-1 was associated with a statistically lower risk of   26 These trends were reflected in our results, which found that among patients with BRAF-mutant disease who initiated 1L treatment, over half (62.5%) were younger than 65 years, and the majority (89.7%) were white and/or male (62.9%). Additionally, 63.4% had a 0-1 ECOG status.
In the present study, most patients lacked documentation for PD-L1 status. PD-L1 testing is not a requirement for aPD-1 or BRAF/MEKi. 27,28 As such, inconsistent use of PD-L1 testing for treatment selection has been described across National Comprehensive Cancer Network (NCCN) institutions. 3 The proportions of patients advancing from 1L to 2L and 3L are consistent with similar studies of community-based care of melanoma. In the real-world study of patients with BRAF-mutant advanced melanoma initiating 1L treatment, Luke et al (2019) reported that only 49.8% received 1L treatment, 43.5% received both 1L and 2L, and only 6.7% received 3L. 29 In another real-world study of treatment-naïve patients with BRAF-mutant advanced melanoma, Whitman et al (2019) found that 57.0% of the BRAF-mutant cohort received only 1L, 28.0% 1L and 2L, and 15.0% 3L+. 26 In the present study, 53.1% of the study population received 2L and 17.9%, 3L.  Current NCCN guidelines list nivolumab and pembrolizumab as preferred (Category 1) regimens, with BRAF/MEKi as preferred regimens among patients with BRAF V600activating mutations. 3 Nivolumab/ipilimumab is also recommended for patients willing and able to tolerate the increased toxicity associated with this regimen. The most common 1L treatments observed in this study are in accordance with these guidelines. Like Whitman et al (2019), approximately one-third of this study population received 1L aPD-1. Due to the lack of comparative Phase III trials, the NCCN guidelines recommend choosing 1L treatment based on the speed of progression, cancer-related symptoms, autoimmune disease, or risk status. 3 The most common 2L treatments in this study, pembrolizumab monotherapy, nivolumab/ipilimumab, dabrafenib/ trametinib, and nivolumab monotherapy, are recommended by NCCN guidelines as second-and subsequent-line treatments. 3 Patients who received 1L aPD-1 monotherapy most commonly received 2L treatment with a BRAF/MEKi combination therapy; conversely, immunotherapies were the most common 2L treatments among patients who received 1L BRAF/MEKi.
In this study, aPD-1 was associated with improved outcomes, including OS, rwPFS, DOR, and TTNT, following 1L initiation among patients with BRAF-mutant advanced melanoma compared to BRAF/MEKi. The lack of statistical significance observed in the propensity score matching results may be due to the relatively small sample size, and these results may nonetheless have clinical relevance. This favorable profile is consistent with published clinical trials, including KEYNOTE-006 and CheckMate 067. 5,12 For this study, the median rwPFS was 7.6 months among the aPD-1 group, which was shorter than the 11.6 months reported in KEYNOTE-006 for the subset of treatment-naïve patients who received pembrolizumab. 5 This difference may be due to underlying differences in this real-world population compared to the clinical trial, or approach to assessment of progression. In contrast, KEYNOTE-006 reported a similar 24-month OS rate, 58.0%, among the subset of patients who received 1L as observed in this study (57.7%). Median OS was not reached among aPD-1 patients in this study or the subset of 1L BRAF-mutant patients in KEYNOTE-006.
Nearly 90% of patients in this study discontinued their 1L treatment, with a higher number of BRAF/MEKi patients discontinuing than those who received aPD-1. While BRAF/ MEKi patients could have had greater data maturity given the FDA approval history, the median follow-up durations were similar between groups (11.5 months for BRAF/MEKi and 11.3 months for aPD-1).
The top two reasons for treatment discontinuation were the same for both treatment groups: disease progression (45.2% and 38.1% of aPD-1 and BRAF/MEKi patients, respectively) and treatment-related toxicities (11.3% and 14.2% of aPD-1 and BRAF/MEKi patients, respectively). Disease progression was also the most common treatment discontinuation reason in pooled data from four clinical trials; specifically, 82% of BRAF-mutant patients discontinued due to progression, with 5% discontinuing due to treatment-related toxicity. 17 In KEYNOTE-001, 42% discontinued due to progressive disease, 25% due to toxicity, 12% due to physician decision, and 5% due to patient withdrawal, with less than 1% lost to follow-up. 30 In CheckMate 067, 7.7% discontinued due to treatment-related adverse events, respectively. 12 Limited real-world data have compared aPD-1 and BRAF/MEKi. Luke et al (2019) retrospectively assessed BRAF-mutant metastatic melanoma patients treated with immunotherapies (nivolumab/ipilimumab or aPD-1) or BRAF/ MEKi in the community oncology setting. The au thors report similar durations of therapy between patients who received dabrafenib/trametinib or aPD-1 (11.4 vs 12.0 months, respectively). 29 Moreover, Whitman et al (2019) observed a median OS of 20.7 months for those receiving aPD-1 and 12.0 months for BRAF/MEKi. 26 They also observed 1-year estimated survival rates of 67.5% and 48.9% for the respective treatments. Several factors may interfere with comparisons across studies. Differences may exist between patient characteristics and eligibility criteria of this study vs other studies, both retrospective observational studies and clinical trials. While the Whitman et al (2019) study included both BRAF-positive and BRAF wild-type disease, it did not measure OS for the separate groups. 26 Also, in Whitman et al (2019), 12.1% had a history of brain metastases compared with 31.3% in the present study. In Luke et al (2019), 11.3% of those receiving aPD-1 had a history of brain metastases compared with 21.0% in our study, and 9.9% of those treated with targeted therapy had a history of brain metastases compared with 37.1% in the BRAF/MEKi 1L group in our study. 29 Also, Luke et al (2019) reported the proportions of patients with brain metastases in the BRAF/MEKi and aPD-1 groups, but did not report a significant difference. More patients in Luke et al (2019) had liver metastases compared with our study (32.4% vs 8.6% in the respective aPD-1 groups and 46.3% vs 30.8% in the BRAF/MEKi groups). Also, treatment outcomes were measured in different ways across studies and/or reported differently across publications, making comparison difficult. For example, in clinical trials, RECIST 1.1 or immune-related response is generally used to measure response (eg, CheckMate 067, KEYNOTE-001, and KEYNOTE-006), whereas in this study, provider-assessed response assessments were used. 12,30,31 Secondly, patients in our study initiated 1L treatment over a 4-year period, whereas in clinical trials, patients initiate treatment within a much shorter time period.
As a retrospective observational study, underlying patient characteristics may have confounded treatment selection, which would have contributed to the observed clinical outcome differences between patients who received BRAF/ MEK treatment and those who received aPD-1. Attempts were made to control for these differences using multivariable Cox regression analyses and propensity score matching. Based on these models, there was an increased risk of death of about 4% with each yearly increase in age and patients diagnosed with Stage III-IV disease, brain metastases, and/or elevated LDH having approximately twice the mortality rate. Conversely, receipt of 1L aPD-1 was associated with a 60% lower risk of death. A similar clinical benefit of aPD-1 was observed among the propensity scorematched pairs.
Prior radiation was associated with an 80% increase in risk of progression. Approximately 20% of patients across both cohorts had both prior radiation and brain metastases at initiation of 1L treatment. Given the known association