Healthcare resource utilization in patients with metastatic melanoma receiving first-line therapy with dabrafenib + trametinib versus nivolumab or pembrolizumab monotherapy

Abstract

Objective: To compare healthcare resource utilization (HRU) between patients with metastatic melanoma (MM) initiated on first-line (1L) combination therapy with the BRAF inhibitor dabrafenib and the MEK inhibitor trametinib (D + T; oral) and those initiated on 1 L monotherapy with the anti-PD1 monoclonal antibodies nivolumab or pembrolizumab (N/P; intravenous).

Methods: Patients with melanoma initiated on D + T or N/P from Q1/2014 to Q2/2016 (defined as 1 L treatment for MM) were identified in the Truven MarketScan database. Entropy balancing was used to reweight the N/P cohort in order to make it comparable to the D + T cohort with respect to the mean and variance of baseline covariates. HRU outcomes during 1 L therapy, reported per patient-year (PPY), were described and compared between the two cohorts post-weighting (i.e. independently of baseline covariates).

Results: Of the 445 patients included, 202 and 243 were initiated on D + T and N/P, respectively. After weighting, patients initiated on N/P had more outpatient visits for drug administration during 1 L therapy than those initiated on D + T (difference = 18.6 visits PPY [95% CI = 16.0–21.1]). Patients initiated on N/P also had more outpatient office visits for reasons other than drug administration (difference = 8.1 visits PPY [95% CI = 1.9–13.7]). No significant differences were observed for other HRU parameters (i.e. inpatient admissions, inpatient days, and emergency department visits during 1 L therapy).

Conclusions: HRU during 1 L therapy was generally similar between patients initiated on D + T and N/P. Nonetheless, patients initiated on N/P had more outpatient visits, including more outpatient visits for reasons unrelated to drug administration.

Keywords:

• Melanoma
• Health resources
• Dabrafenib
• Trametinib
• Nivolumab
• Pembrolizumab
• First-line

Introduction

Over the past decade, the median overall survival of patients with metastatic melanoma has increased from 7.5 months¹ to at least 25 months^2–4 due to the advent of new therapies that are more effective in slowing disease progression. For patients with metastatic melanoma whose tumors harbor BRAF V600E/K activating mutations, representing ∼40–50% of melanoma tumors^5,6, the v1.2018 NCCN guidelines recommend first-line (1L) treatment with either targeted therapies or immune checkpoint inhibitors. Combination therapy with the BRAF inhibitor dabrafenib and the MEK inhibitor trametinib (D + T) and monotherapy with the anti-PD1 monoclonal antibodies nivolumab and pembrolizumab (N/P) are among the most commonly used treatments in clinical practice in their respective treatment classes^7–9. However, while both D + T^2,10,11 and N/P^3,4,11–13 demonstrated a significantly better efficacy compared to older treatments in their respective classes, head-to-head comparisons between D + T and N/P with respect to either clinical or economic outcomes are lacking. Given the treatment and care of patients with metastatic melanoma still require substantial healthcare resources, comparative data on these treatments are needed to inform treatment decisions for physicians, payer, and patients. To the best of our knowledge, such updated data, focusing specifically on healthcare resource utilization (HRU), do not exist. Using a large claims database in the US, the aim of this study was to compare HRU between commercially insured patients initiated on D + T and N/P as 1 L treatment for metastatic melanoma.

Methods

Data source

Data from the Truven Health Analytics MarketScan Commercial database were used (January 1, 2008–June 30, 2016). This database contains health services claims for more than 230 million individuals from ∼100 employers and several health plans. All census regions are represented, but the South and North Central (Midwest) regions are predominant. Available data include information on history of health plan enrollment, demographics, diagnoses, claims for medical care received across all settings, and claims for pharmacy services for the covered employees and their dependents. Data are de-identified and comply with the confidentiality requirements of the Health Insurance Portability and Accountability Act.

Study design

A retrospective cohort study design was used. The index date was defined as the date of initiation of D + T or N/P (for the D + T and N/P study cohorts, respectively). Patient baseline characteristics were measured in the 6 months prior to the index date (baseline period). HRU outcomes (please see Study outcome measures section) were measured per patient-year (PPY) during 1 L therapy with D + T or N/P. Patients still on 1 L therapy at the end of data availability or disenrollment/death were censored.

Study sample

The study sample included patients with melanoma initiated on D + T or N/P on or after January 1, 2014, without previous use of any targeted therapy or immune checkpoint inhibitor for metastatic melanoma (per NCCN guidelines, D + T and N/P initiated in this setting were assumed to be 1 L treatments for metastatic melanoma). Sample selection is described in more details in Figure 1.

Figure 1. Flowchart for the selection of patients in the study sample. 1L = first-line. [1] Immune-checkpoint inhibitors: Ipilimumab, Pembrolizumab, Nivolumab (used alone, in combination with each other, or in combination with other antineoplastic drugs). [2] Targeted therapies: Vemurafenib, Dabrafenib, Trametinib (used alone, in combination with each other, or in combination with other other antineoplastic drugs). [3] Melanoma was identified based on ICD-9 codes 172.xx or ICD-10 codes C43.xx. [4] Identified based on ICD-9 diagnostic code V70.7 (examination of participant in clinical trial) or ICD-10 code Z00.6. [5] End of line of therapy was identified as having a gap of >45 days after end of line therapy before data cut-off or disenrollment, or the start of a new line of therapy.

Identification of study cohorts

Identification of 1 L therapy was adapted from previously published algorithms^14–16. 1 L therapy initiation corresponded to the first targeted therapy or immune checkpoint inhibitor for metastatic melanoma received between January 1, 2008 and June 30, 2016 (i.e. the study period; by design, only patients who initiated 1 L therapy on or after January 1, 2014 were included). If dabrafenib and trametinib were the only anti-neoplastic agents prescribed or administered in the 28 days following 1 L therapy initiation, patients were considered to be treated with 1 L D + T combination therapy. Similarly, if nivolumab or pembrolizumab was the only anti-neoplastic agent prescribed or administered in the 28 days following 1 L start, patients were considered to be treated with 1 L N/P monotherapy. For N/P, which are administered intravenously, 1 L discontinuation was defined as the earliest of (a) 21 days after the last intravenous administration, followed by a gap of >45 consecutive days without administration of agents in the treatment regimen and without use of corticosteroid during the treatment gap (to account for short treatment gaps due to immune adverse events), or (b) the day before initiation of a new antineoplastic agent that was not part of the N/P regimen. For D + T, which are administered orally, 1 L discontinuation was defined as the earliest date of (a) end of supply of both dabrafenib and trametinib (i.e. all agents in the 1 L regimen), before a gap of >45 days without dispensing of either dabrafenib or trametinib, or (b) the day before administration of a new antineoplastic agent that was not part of the D + T regimen.

Study outcome measures

HRU outcomes during 1 L therapy included the number of inpatient admissions, number of inpatient days, number of emergency department (ED) visits, number of outpatient visits excluding drug administration (stratified by home care, office, and other visits [including ambulatory surgical care, laboratory tests, imaging, etc.]), number of outpatient visits for drug administration (any drug), and number of visits for dental and vision care. For patients with multiple outpatient visits occurring on the same day, visits were counted separately provided they fell into different categories (i.e. three categories for outpatient visits excluding drug administration: home care, office visits, other visits; and one category for outpatient visits for drug administration [any drug]). However, multiple daily visits that fell into the same category (among those previously listed) were counted as single visits. HRU was reported per patient-year of 1 L therapy.

The following potential confounders measured in the baseline period were evaluated and controlled for in the analyses (using a cohort weighting approach as described in the Statistical analysis section): age, use of prior pharmacological (using NDC and HCPCS codes) or radiological cancer-directed therapies (using CPT, HCPCS, and ICD-9/10 diagnosis and procedure codes), presence of brain or bone metastases (using ICD-9/10 diagnosis codes), number of metastatic sites (using ICD-9/10 diagnosis codes), Charlson comorbidity index (CCI; using ICD-9/10 diagnosis codes)¹⁷, specific comorbidities at baseline (using ICD-9/10 diagnosis codes), inpatient admissions during the baseline period, and all-cause costs during the baseline period. Other characteristics that were assessed included gender, tumor-related surgeries (i.e., skin biopsy, excision of skin tumor, and lymph node dissection; using CPT, and ICD-9/10 diagnosis and procedure codes), comorbidities unrelated to cancer (using ICD-9/10 diagnosis codes), and ED visits in the baseline period.

Statistical analysis

Descriptive statistics for patient baseline characteristics and HRU outcomes relied on frequencies and proportions for categorical variables, and means, medians, standard deviations, and interquartile ranges for continuous variables. Baseline characteristics were compared between study cohorts using Chi-square tests for categorical variables and t-tests for continuous variables.

To control for potential confounding in HRU comparisons between the study cohorts, a weighting approach was used to make the study cohorts comparable at baseline. While this can be accomplished using several different approaches, the current study relied on the entropy-balance (e-balance) method, which achieves better covariate balance than other commonly used weighting methods such as inverse probability of treatment weighting (IPTW), with the added advantage of having no reliance on correct propensity score model specification. While e-balance and IPTW use different methods to derive weights that make cohorts comparable, both use weights in a similar way to control for confounding: IPTW re-weights both study cohorts to make them similar in covariate distribution to the study sample, while e-balance re-weights only one cohort to make it similar to the other cohort in covariate distribution. The e-balance method derives weights so that pre-specified moments (e.g. mean only, mean and standard deviation) for the covariate distribution will be the same in the weighted cohort (here N/P) as in the reference cohort (here D + T). Although recently developed, e-balance is increasingly used in observational studies^18–21. To ensure that the weighting approach achieved its purpose, balancing of the two cohorts before and after e-balance was assessed using standardized differences. While there is no consensus as to what threshold of standardized difference should be used to highlight the absence of meaningful confounding, rule of thumb guidelines indicate that standardized differences below |10|–|20|% represent reasonable cut-offs^22–26.

In weighted analyses (i.e. analyses that control for observed potential confounders), confidence intervals (CIs) for differences between the weighted study cohorts in HRU outcomes were computed using bootstrap with 499 resamples²⁷.

Results

Description of study cohorts prior to e-balance weighting (i.e. as observed)

After applying all eligibility criteria, the study population consisted of 445 patients with metastatic melanoma, including 202 and 243 in the D + T and N/P cohorts, respectively.

Before e-balance weighting (Table 1, columns 1–4), patient mean age was 56.7 and 65.3 years in the D + T and N/P cohorts, respectively (p < .001) and male patients represented 63.9% and 63.0% of D + T- and N/P-initiated patients, respectively (p = .845). Although patients in the D + T cohort were younger than those in the N/P cohort, they appeared to have a higher disease burden at baseline with respect to the site and number of metastases (e.g. brain metastases: 40.1% vs 21.4%, p < .001), mean number of metastatic sites (i.e. 2.5 vs 2.1, p < .001), and use of emergency care (patients with ≥1 inpatient admission: 50.5% vs 36.2%, p = .002; patients with ≥1 ED visit: 36.6% vs 30.9%, p = .199; Table 1). Comorbidities unrelated to cancer were generally similar between the study cohorts before weighting, with the exception of diabetes, which appeared more common in the N/P cohort compared to the D + T cohort (20.6% vs 8.9%, p < .001; Table 1).

Table 1. Characteristics of patients treated with dabrafenib + trametinib or nivolumab/pembrolizumab in 1 L post index, before and after e-balance weighting.

Patient characteristics	1L Therapy
	Dabrafenib + Trametinib	Nivolumab/Pembrolizumab
	Reference^a E-balance weighting	Before E-balance weighting	After E-balance weighting^b
	(n = 202)	(n = 243)	(n = 243)^c
Demographics (at index date)
*Age
Mean ± SD	56.7 ± 13.9	65.3 ± 13.3	56.7 ± 13.9
Median [IQR]	57.5 [48.0–64.0]	63.0 [56.0–76.0]	56.0 [46.0–64.0]
Males, n (%)	129 (63.9%)	153 (63.0%)	147 (60.6%)
Underlying tumor burden (during baseline period)
Prior cancer-directed therapies used, n (%)
*Pharmacological	16 (7.9%)	40 (16.5%)	19 (7.9%)
*Radiation therapy	88 (43.6%)	63 (25.9%)	106 (43.6%)
Surgery (one patient can have surgery in more than 1 category), n (%)
Skin biopsy	48 (23.8%)	56 (23.0%)	43 (17.6%)
Excision of skin tumor	55 (27.2%)	78 (32.1%)	72 (29.8%)
Lymph node dissection	46 (22.8%)	49 (20.2%)	48 (19.8%)
*Brain metastases, n (%)	81 (40.1%)	52 (21.4%)	97 (40.1%)
*Bone metastases, n (%)	52 (25.7%)	52 (21.4%)	63 (25.7%)
*Number of metastatic sites (max =6)
Mean ± SD	2.5 ± 1.4	2.1 ± 1.3	2.5 ± 1.4
Median [IQR]	2.0 [1.0–3.0]	2.0 [1.0–3.0]	2.0 [1.0–4.0]
Comorbidities (during baseline period)
*Charlson comorbidity score [CCI; Deyo et al.^17)
Mean ± SD	6.9 ± 1.2	7.0 ± 1.2	6.9 ± 1.2
Median [IQR]	6.0 [6.0–8.0]	7.0 [6.0–8.0]	6.0 [6.0–8.0]
Specific comorbidities, n (%)
*Anemia	42 (20.8%)	61 (25.1%)	51 (20.8%)
Cardiovascular disease^d	88 (43.6%)	109 (44.9%)	91 (37.4%)
*Diabetes (type I or II)	18 (8.9%)	50 (20.6%)	22 (8.9%)
Hemiplegia	7 (3.5%)	9 (3.7%)	12 (4.8%)
Immune diseases^e	19 (9.4%)	22 (9.1%)	16 (6.6%)
Liver disease	49 (24.3%)	57 (23.5%)	62 (25.4%)
Renal disease	22 (10.9%)	30 (12.3%)	15 (6.3%)
All-cause HRU and healthcare cost (during baseline period)
HRU in the baseline period, n (%)
*Patients with inpatient admission(s)	102 (50.5%)	88 (36.2%)	123 (50.5%)
Patients with emergency department visit(s)	74 (36.6%)	75 (30.9%)	63 (25.8%)
*Total healthcare cost in the baseline period
Mean ± SD	65,873 ± 85,746	77,260 ± 121,369	65,873 ± 72,476
Median [IQR]	38,145 [15,478–83,348]	38,207 [11,700–87,233]	43,071 [21,232–87,233]

*Covariate was used in the balancing algorithm (i.e. potential confounder).

^aUnlike other weighting approaches like inverse probability of treatment weighting (IPTW), e-balance adjusts results in the adjustment of covariates for only one cohort, leaving those of the other cohort (the “reference” cohort) unchanged.

^bWeights in the N/P group were calculated by e-balance weighting, so that the first and second moments were balanced for the following covariates measured in the baseline period: age, prior cancer therapies used (pharmacological, radiation therapy), CCI, number of metastases, presence of brain or bone metastases, anemia, diabetes (type I or II), any IP admission and any ER visit in the baseline period. In addition, all-cause cost in the baseline period was balanced on the first moment. By design, weights were set to 1 for all patients in the D + T group (reference group).

^cWeights have been normalized to preserve the original size of the cohorts.

^dAny of the following: cerebrovascular disease, congestive heart failure, myocardial infarction, peripheral vascular disease.

^eAny of the following: ankylosing spondylitis; thyroiditis; Graves’ disease; inflammatory bowel disease [including Chron’s disease and ulcerative colitis], multiple sclerosis, psoriasis, psoriatic arthritis, rheumatoid arthritis, systemic lupus erythematosus, or vitiligo.

Abbreviations. 1L, first-line; SD, standard deviation; IQR, interquartile range; HRU, healthcare resource utilization.

Covariate balance after e-balance weighting

E-balance weighting resulted in well-balanced cohorts (i.e. standardized differences <|10|%), with a few exceptions for ED visits during the baseline period (standardized difference = 0.23) and skin biopsy, cardiovascular, and renal disease (standardized differences = 10–20%; Figure 2).

Figure 2. Balance of characteristics of patients treated with dabrafenib + trametinib or nivolumab/pembrolizumab in 1L post-index, before and after e-balance weighting. Abbreviations. IP, Inpatient; ED, Emergency Department; CCI, Charlson comorbidity index (Deyo et al.¹⁷). [1] While there is no consensus as to what threshold of standardized difference should be used to highlight the absence of meaningful confounding, rule of thumb guidelines indicate that a standardized difference below |10|–|20|% represents reasonable cut-offs (Cohen²², Yang and Dalton²³, Austin²⁴, Harder et al.²⁵, Stuart et al.²⁶). Of note, a standardized difference of <|0.2| indicates <15% non-overlap in the distribution of covariate in the two cohorts (Yang and Dalton²³).

HRU comparison between D + T and N/P cohorts after e-balance weighting

Over a mean duration of 1 L therapy of 24 weeks for the D + T cohort and 16 weeks for the N/P cohort (Figure 1), patients in the D + T cohort had significantly fewer outpatient office visits (difference = –8.1 visits PPY of 1 L therapy, 95% CI = –13.7 to –1.9) and outpatient visits for drug administration (difference = –18.6 visits PPY, 95% CI = –21.1 to –16.0; Figure 3) relative to patients in the e-balance weighted N/P cohort. All other HRU outcomes were similar between the weighted cohorts (Figure 3 shows all HRU components where both cohorts had >1 visit PPY on average; see the Appendix for all HRU components).

Figure 3. All-cause healthcare resource utilization during 1L post-index, after e-balance weighting. Abbreviations. 1L, first-line; IP, Inpatient; ED, Emergency Department; OP, Outpatient; D + T, dabrafenib + trametinib combination therapy cohort; N/P, nivolumab or pembrolizumab monotherapy cohort. [1] Weights in the Nivolumab/Pembrolizumab group were calculated by e-balance weighting, so that the first and second moments were balanced for the following covariates measured in the baseline period: age, prior cancer therapies used (pharmacological, radiation therapy), CCI, number of metastases, presence of brain or bone metastases, anemia, diabetes (type I or II), any IP admission and any ER visit in the baseline period. In addition, all-cause cost in the baseline period was balanced on the first moment. By design, weights were set to 1 for all patients in the Dabrafenib + Trametinib group (reference group).

Discussion

In this retrospective claims-based study, HRU was compared between two cohorts of real-world patients with metastatic melanoma initiated on either D + T or N/P in 1 L. Before e-balance weighting, patients initiated on D + T appeared to be younger and sicker than those initiated on N/P. Notably, the proportion of patients with brain metastases in the D + T cohort was nearly twice as high as that of patients in the N/P cohort at baseline. After controlling for confounding related to cohort differences at baseline in brain metastases and other factors, HRU was generally similar between the two treatments, particularly for inpatient admissions and ED visits. Nonetheless, patients initiated on N/P had more outpatient visits than those initiated on D + T, including outpatient visits for reasons unrelated to drug administration.

In the current study, brain metastases were more common among patients initiated on D + T in 1 L as compared to those initiated on N/P in 1 L prior to e-balance weighting. Given patients with brain metastases are often excluded¹² or significantly under-represented^3,13,28 in clinical trials investigating targeted therapies and immune checkpoint inhibitors, real-world data are needed on this difficult-to-treat population. Indeed, a recent study estimated that 55% of real-world patients with metastatic melanoma were not eligible to recent immunotherapy trials due to the presence of brain metastases or Eastern Cooperative Oncology Group performance status ≥2²⁹. Given patients with brain metastases typically have a worse prognosis than other patients with metastatic melanoma, this may prompt physicians to opt for the therapy option that elicits the most rapid response to prolong survival (i.e. targeted therapies) as recommended in the most recent NCCN guidelines³⁰. The results of the current study corroborate this hypothesis. In fact, in addition to including a higher proportion of patients with brain metastases, those treated with D + T in the current study also had a significantly higher number of metastatic sites. While there is evidence for the clinical activity of immune checkpoint inhibitors in treating brain metastases³¹, physicians may be aware of data supporting the higher efficacy of targeted therapies in the treatment of brain metastases and favor these treatments^31,32.

After controlling for confounding due to the presence of brain metastases and other factors via e-balance weighting, HRU was generally similar between patients initiated on N/P and D + T in 1 L, notably for inpatient admissions and ED visits during 1 L. Nonetheless, patients initiated on N/P had more outpatient visits during 1 L, including more outpatient visits for reasons other than drug administration, and also more visits for drug administration. Some of the differences in outpatient visits between the N/P and D + T cohorts are likely explained by the modes of administration of these two drugs: small molecule inhibitors like D + T can be administered orally, whereas immune checkpoint inhibitors need to be administered either subcutaneously or intravenously. To date, the only approved mode of administration of nivolumab and pembrolizumab by the US Food and Drug Administration is intravenously, which involves several drawbacks from a healthcare and patient convenience perspective. First, because intravenous administration requires that drug administration is supervised by medical staff, this entails additional healthcare resources and most likely explains the observed difference in the number of outpatient visits related to drug administration between the two cohorts. Furthermore, infusion reactions may occur with intravenous drug administration, also leading to increased healthcare resource utilization³³. Indeed, infusion reactions, mostly of mild-to-moderate severity, have been reported at rates of 6.4% and 2.5% for nivolumab in the CHECK MATE 037 and CHECK MATE 067, respectively. However, the majority of infusion reactions occur during the first administration of the drug^33,34, which means that the impact of infusion reactions on HRU may be more diluted over longer follow-up periods. Other possible reasons for the observed differences between the study cohorts with respect to outpatient visits for reasons other than drug administration remain speculative, but monitoring and testing likely contribute. In fact, an unweighted exploratory analysis conducted in the current study showed that, in both treatment cohorts, the most common procedure codes during outpatient visits for reasons other than drug administration were related to the evaluation and management of established patients, blood count tests, and tests for other serum biomarkers (data not shown). This observation also held true when analyzing the procedure codes associated with the sub-set of outpatient office visits that occurred on the same day as drug administration. However, a plethora of low-frequency procedures was also observed among the study patients, complicating the interpretation of results, and no one procedure code stood out as a single reason for the difference in outpatient visits. Further studies are needed to assess whether differences in outpatient care during 1 L for reasons other than drug administration are driven by differences in the safety and efficacy profiles of N/P and D + T or by physician decisions to monitor more closely patients with specific characteristics.

The current study is subject to a few limitations. First, D + T was approved and started to be used before N/P. Consequently, treatment end was observed for fewer patients in the N/P cohort, and the observed duration of 1 L was shorter. However, this should not affect results if HRU does not vary over the duration of 1 L. Second, although treatment discontinuation before the end of 1 L therapy is common among N/P-treated patients³⁰, the present study could not assess HRU between treatment discontinuation and progression because (a) the follow-up period for N/P-initiated patients is too short (see first limitation) and (b) progression cannot be assessed with claims data. Third, although e-balance weighting was used to control for confounding in comparative analyses, residual confounding may still exist due to unobserved factors (e.g. performance status at 1 L initiation) or to observed factors that remained unbalanced after e-balance weighting (i.e. ED visits remained more frequent in the D + T cohort after e-balance weighting, suggesting the magnitude of the differences for outpatient visits, which favored the D + T cohort, is likely under-estimated). Fourth, claims databases may contain omissions and inaccuracies, although this should equally impact both cohorts. For example, BRAF V600+ tumors are likely more common in D + T- than N/P-treated patients. Since BRAF-activating mutations are associated with a worse prognosis³⁵, the HRU estimates provided in the current study may be under-estimated for the N/P cohort. Fifth, due to the relatively small sample size, estimates for uncommon types of HRU are less precise, limiting their interpretability. Finally, given ipilimumab + nivolumab was approved by the FDA in late 2015, only 76 patients received this treatment in Truven data over the study period. Thus, N/P and D + T were selected for the analysis as they represented the largest treatment groups in Truven data.

Conclusions

In this retrospective claims-based study, HRU was compared during 1 L treatment between real-world populations with metastatic melanoma initiated on D + T (i.e. targeted therapies) and N/P (i.e. immune checkpoint inhibitors). The results suggest that patients presenting with brain metastases tend to be treated with D + T. In addition, patients initiated on D + T incurred a lower number of office visits related to drug administration, and, to a lesser extent, office visits unrelated to drug administration, suggesting that intravenous administration of nivolumab and pembrolizumab increases the healthcare burden of these treatment options. To inform treatment decisions for physicians, payers, and patients, the results from the current study should be corroborated with results from studies that compare other clinical, economic, and quality-of-life outcomes between patients with metastatic melanoma initiated on D + T or N/P in the 1L.

Transparency

Declaration of funding

This work was supported by Novartis Pharmaceuticals Corporation.

Declaration of financial/other relationships

Sameer R. Ghate, Briana Ndife, and Antonio Nakasato are employees of Novartis Pharmaceuticals Corporation and may own stock/stock options. Raluca Ionescu-Ittu, François Laliberté, Rebecca Burne, and Mei Sheng Duh are employees of Analysis Group Inc., a consultancy company that received financial support from Novartis Pharmaceuticals Corporation in connection with this study. A CMRO peer reviewer on this manuscript declares performing clinical trials with the agents mentioned in this publication. Another CMRO peer reviewer declares receiving speakers’ honoraria from Roche, Novartis, and Bristol-Myers Squibb. Other CMRO peer reviewers have no financial/other relationships to disclose.

Acknowledgments

Medical writing assistance was provided by Samuel Rochette, an employee of Analysis Group, Inc.

Appendix: All-cause healthcare resource utilization during 1 L post-index, after e-balance weighting

Table

All-cause healthcare resource utilization component per patient-year of 1L therapy, mean ± SD [median]	All-cause healthcare resource utilization during 1L
	Before/After Weighting	Before Weighting	After Weighting
	D + T in 1L	N/P in 1L	N/P in 1L	Difference (95% CI)
	(n = 202)	(n = 243)	(n = 243)
Number of IP admissions	2.3 ± 4.2 [0.0]	1.3 ± 4.0 [0.0]	2.2 ± 5.3 [0.0]	0.1 (–1.3; 1.4)
Number of IP days	15.6 ± 40.6 [0.0]	8.1 ± 28.7 [0.0]	15.2 ± 43.9 [0.0]	0.4 (–12.3; 12.7)
Number of IP days with at least one  hospice/skilled nursing claim	1.2 ± 7.0 [0.0]	0.2 ± 2.1 [0.0]	1.0 ± 5.4 [0.0]	0.2 (–1.7; 2.0)
Number of ED visits	1.6 ± 3.5 [0.0]	1.7 ± 4.9 [0.0]	2.1 ± 5.3 [0.0]	–0.5 (–2.2; 0.8)
Number of days with hospice/skilled nursing claims outside IP	1.0 ± 6.0 [0.0]	0.9 ± 7.4 [0.0]	1.2 ± 9.6 [0.0]	–0.2 (–3.3; 1.4)
Number of OP visits excluding drug administration	46.8 ± 33.3 [42.9]	49.9 ± 29.6 [43.9]	53.7 ± 29.0 [52.0]	–6.9 (–13.0; 0.2)
Home care	3.5 ± 13.7 [0.0]	2.8 ± 13.9 [0.0]	3.1 ± 13.6 [0.0]	0.4 (–3.0; 3.2)
Office	41.6 ± 29.7 [37.8]	45.9 ± 27.0 [41.2]	49.7 ± 25.9 [49.2]	–8.1 (–13.7; –1.9)
Other	1.6 ± 5.3 [0.0]	1.1 ± 3.9 [0.0]	0.9 ± 3.6 [0.0]	0.8 (–0.3; 1.8)
Number of OP visits for drug administration (any drug)	3.4 ± 6.1 [0.0]	20.9 ± 9.6 [19.2]	22.0 ± 10.8 [18.8]	–18.6 (–21.1; –16.0)
Number of visits for dental and vision care	0.1 ± 0.7 [0.0]	0.4 ± 2.5 [0.0]	0.3 ± 1.4 [0.0]	–0.2 (–0.6; 0.1)

Abbreviations. 1 L, first-line; IP, Inpatient; ED, Emergency Department; OP, Outpatient; D + T, dabrafenib + trametinib combination therapy cohort; N/P, nivolumab or pembrolizumab monotherapy cohort.