Venous thromboembolism and bleeding in advanced melanoma: a propensity-matched cohort analysis comparing first-line immunotherapy and targeted therapy

Abstract

Background

Venous thromboembolism (VTE) is a leading cause of morbidity in cancer. Immune checkpoint inhibitors and BRAF/MEK-targeted therapy (BRAF/MEKi) are competing first-line options for advanced melanoma, but their comparative thrombotic and bleeding risks remain poorly characterized.

Objectives

To compare VTE and bleeding outcomes between first-line ipilimumab/nivolumab (Ipi/Nivo) and BRAF/MEKi in advanced melanoma.

Methods

We conducted a retrospective cohort study using the TriNetX federated electronic health record network. Adults with advanced melanoma initiating Ipi/Nivo or BRAF/MEKi were matched 1:1 by propensity score matching. The primary outcome was composite VTE (pulmonary embolism or deep vein thrombosis) at 6 months; secondary outcomes included bleeding events and all-cause mortality. Hazard ratios (HRs) with 95% CIs were estimated using Cox proportional hazards models.

Results

After 1:1 matching, 1043 patients per group were included, with well-balanced covariates. Ipi/Nivo was associated with significantly higher 6-month composite VTE (7.0% vs 4.1%; HR, 1.76; 95% CI, 1.19-2.61; P = .004), pulmonary embolism (5.0% vs 2.8%; HR, 1.89; 95% CI, 1.17-3.04; P = .008), and deep vein thrombosis (3.2% vs 1.7%; HR, 1.91; 95% CI, 1.04-3.50; P = .03). No significant differences were observed in composite bleeding (HR, 1.07; 95% CI, 0.79-1.45; P = .65), intracranial hemorrhage, or gastrointestinal bleeding. Six-month all-cause mortality was numerically higher with Ipi/Nivo but did not reach statistical significance (18.9% vs 16.1%; HR, 1.22; 95% CI, 0.99-1.51; P = .06).

Conclusion

First-line Ipi/Nivo appears to be associated with a higher VTE risk compared with BRAF/MEKi in advanced melanoma, without a concomitant increase in bleeding. These hypothesis-generating findings support prospective evaluation of thromboprophylaxis strategies in Ipi/Nivo-treated patients.

Essentials

• •VTE and bleeding risks with first-line immune checkpoint inhibitors versus BRAF/MEK-targeted therapy in advanced melanoma remain poorly characterized.
• •We conducted a propensity-matched retrospective cohort study of 1,043 patients per group using the TriNetX federated electronic health record network (2014-2025).
• •First-line ipilimumab/nivolumab was associated with significantly higher 6-month VTE risk than BRAF/MEK inhibitors (7.0% vs 4.1%; HR 1.76, 95% CI, 1.19–2.61), without a corresponding increase in bleeding.
• •These hypothesis-generating findings support prospective evaluation of thromboprophylaxis strategies in patients receiving first-line ipilimumab/nivolumab for advanced melanoma.

1. Introduction

Venous thromboembolism (VTE) is a leading cause of morbidity and mortality in cancer [1]. VTE burden increases substantially in advanced melanoma, approaching rates seen in high-risk malignancies [2]. In a recent population-based analysis of cancer-associated VTE in the United States, the overall 12-month cumulative incidence of VTE was 1.7% (95% CI, 1.5%-1.8%), rising to 4.0% (95% CI, 3.6%-4.5%) among patients receiving systemic anticancer therapy [3]. In patients with stage IV melanoma specifically, the prevalence of thromboembolic events has been reported at approximately 25.2% [2]. First-line treatment includes immune checkpoint inhibitors (ICIs) and BRAF/MEK-targeted therapy (BRAF/MEKi). BRAF V600 mutations occur in approximately 40% to 50% of advanced melanomas, making BRAF/MEKi a relevant first-line option for a substantial proportion of patients [4]. ICIs have been linked to an elevated thrombotic risk, with a reported VTE incidence ranging from 7.6% at 6 months [5] to 12.9% over a median of 8.5 months of follow-up [6], and significantly higher VTE rates with combined PD-1/CTLA-4 inhibition compared with ICI monotherapy [7]. BRAF/MEKi carry cardiovascular toxicity, including a higher pulmonary embolism (PE) risk with combination regimens [8]. However, the comparative thrombotic and bleeding risks of first-line ipilimumab/nivolumab (Ipi/Nivo) vs BRAF/MEKi remain poorly characterized. Beyond its prognostic impact, VTE in cancer disrupts systemic therapy through treatment delays, hospitalization, and the added burden of anticoagulation in a population already at competing risk of bleeding [9]. Existing data on ICI-associated thrombosis are predominantly from single-cohort studies without an active comparator, precluding direct risk quantification. Current VTE risk stratification tools, including the Khorana score, were not developed for ICI-treated populations and demonstrate limited discriminatory capacity in this setting [10]. Comparative bleeding risk with Ipi/Nivo—a clinically important consideration when evaluating thromboprophylaxis—remains similarly uncharacterized.

2. Methods

We performed a retrospective cohort study of adults with advanced melanoma using the TriNetX research network, a federated platform providing access to deidentified electronic health record data from more than 100 participating healthcare organizations, with prior evaluations demonstrating broadly representative population coverage compared with U.S. Census benchmarks, supporting its use for real-world evidence generation [11]. TriNetX provides deidentified data compliant with Health Insurance Portability and Accountability Act Privacy Rule standards; therefore, this study was exempt from institutional review board review.

Patient data were extracted from the TriNetX platform for the period 2014 to 2025. Melanoma was defined as ≥2 International Classification of Diseases, 10th Revision (ICD-10) C43 codes ≥7 days apart. Patients were classified by first-line therapy (Ipi/Nivo or BRAF/MEKi); anti–PD-1 monotherapy was excluded. Patients with prior VTE, gastrointestinal bleeding (GIB), or intracranial hemorrhage (ICH) were excluded. To minimize treatment switching bias, patients in the Ipi/Nivo cohort who received pembrolizumab or a BRAF/MEKi agent within 3 years before or up to 3 months after the Ipi/Nivo index date were excluded, and patients in the BRAF/MEKi cohort who received pembrolizumab or Ipi/Nivo within 3 years before or up to 3 months after the BRAF/MEKi index date were excluded. We performed 1:1 greedy nearest-neighbor propensity score matching on demographics, metastasis sites, comorbidities, baseline medications, and healthcare utilization (outpatient services and inpatient/observation care encounters) using a caliper of 0.1 pooled SDs. PE (I26) and GIB (K92.0-K92.2) were identified using ICD-10 codes with published positive predictive values of 79% to 92% and 94% to 100%, respectively, in anticoagulated patient populations [12,13]. Deep vein thrombosis (DVT; I82.40) and ICH (I60-I62) codes were chosen from widely used codes in similar pharmacoepidemiology research [14]. The DVT code (I82.40) captures acute embolism and thrombosis of unspecified deep veins of the lower extremity, without distinguishing proximal from distal segments, and does not include upper-extremity DVT. The primary outcome was composite VTE (DVT or PE) within 6 months; secondary outcomes included individual VTE components, bleeding events (including GIB and ICH), and all-cause mortality. Hazard ratios (HRs) with 95% CIs were estimated using Cox proportional hazards models.

3. Results and Discussion

Before matching, 1784 Ipi/Nivo and 1162 BRAF/MEKi patients were identified; after 1:1 propensity score matching, 1043 patients remained in each cohort with well-balanced baseline covariates (all standardized mean differences ≤ 0.05; Table 1 and Supplementary Figure S1). The median follow-up after matching was 180 days in both cohorts (Ipi/Nivo: mean [SD], 159 [47] days; BRAF/MEKi: mean [SD], 161 [45] days). At 6 months, composite VTE incidence was 7.0% with Ipi/Nivo and 4.1% with BRAF/MEKi (HR, 1.76; 95% CI, 1.19-2.61; P = .004). PE rates were higher with Ipi/Nivo (5.0% vs 2.8%; HR, 1.89; 95% CI, 1.17-3.04; P = .008). DVT rates were also elevated in the Ipi/Nivo group (3.2% vs 1.7%; HR, 1.91; 95% CI, 1.04-3.50; P = .03; Figures 1 and 2A).

Table 1.

Baseline characteristics before and after 1:1 propensity score matching.

Characteristic	Before match			After match
	Ipi/Nivo (n = 1784)	BRAF/MEKi (n = 1162)	SMD	Ipi/Nivo (n = 1043)	BRAF/MEKi (n = 1043)	SMD
Demographics
Age at index (y)	61.1 ± 13.7	57.9 ± 14.8	0.22	58.6 ± 14.2	58.8 ± 14.4	0.02
Female	667 (37.4)	493 (42.4)	0.10	417 (40.0)	433 (41.5)	0.03
White	1589 (89.1)	1027 (88.4)	0.02	931 (89.3)	931 (89.3)	0
Black or African American	19 (1.1)	10 (0.9)	0.02	10 (1.0)	10 (1.0)	0
Asian	23 (1.3)	10 (0.9)	0.04	10 (1.0)	10 (1.0)	0
Not Hispanic or Latino	1382 (77.5)	807 (69.4)	0.18	741 (71.0)	755 (72.4)	0.03
BMI	28.6 ± 6.4	28.8 ± 6.5	0.03	28.8 ± 6.7	28.7 ± 6.4	0.02
Medications
Aspirin	284 (15.9)	196 (16.9)	0.03	170 (16.3)	170 (16.3)	0
Clopidogrel	44 (2.5)	37 (3.2)	0.04	34 (3.3)	30 (2.9)	0.02
Apixaban	66 (3.7)	35 (3.0)	0.04	37 (3.5)	35 (3.4)	0.01
Rivaroxaban	26 (1.5)	20 (1.7)	0.02	16 (1.5)	17 (1.6)	0.01
Warfarin	27 (1.5)	14 (1.2)	0.03	15 (1.4)	13 (1.2)	0.02
Systemic hormonal contraceptives	31 (1.7)	40 (3.4)	0.11	26 (2.5)	23 (2.2)	0.02
Comorbidities
Hypertension	694 (38.9)	449 (38.6)	0.01	386 (37.0)	399 (38.3)	0.03
Hyperlipidemia	524 (29.4)	280 (24.1)	0.12	257 (24.6)	267 (25.6)	0.02
Diabetes mellitus	243 (13.6)	186 (16.0)	0.07	158 (15.1)	155 (14.9)	0.01
Atrial fibrillation and flutter	118 (6.6)	86 (7.4)	0.03	75 (7.2)	76 (7.3)	0
Chronic kidney disease	82 (4.6)	71 (6.1)	0.07	61 (5.8)	60 (5.8)	0
COPD	76 (4.3)	64 (5.5)	0.06	51 (4.9)	56 (5.4)	0.02
Heart failure	57 (3.2)	57 (4.9)	0.09	44 (4.2)	47 (4.5)	0.01
Chronic venous insufficiency	29 (1.6)	15 (1.3)	0.03	12 (1.2)	15 (1.4)	0.03
Healthcare utilization
Outpatient services	1383 (77.5)	782 (67.3)	0.23	760 (72.9)	737 (70.7)	0.05
Inpatient/observation care	387 (21.7)	264 (22.7)	0.03	251 (24.1)	236 (22.6)	0.03
Disease burden
Lymph node metastasis	842 (47.2)	736 (63.3)	0.33	637 (61.1)	629 (60.3)	0.02
Respiratory and digestive organ metastasis	904 (50.7)	493 (42.4)	0.17	465 (44.6)	459 (44.0)	0.01
Brain metastasis	482 (27.0)	288 (24.8)	0.05	276 (26.5)	265 (25.4)	0.02
Other sites metastasis	1254 (70.3)	791 (68.1)	0.05	708 (67.9)	711 (68.2)	0.01

Values are presented as mean ± SD for continuous variables and n (%) for categorical variables. SMDs are shown to assess covariate balance between treatment groups before and after matching. Propensity score matching incorporated demographics, baseline medications, comorbidities, metastatic disease burden, and healthcare utilization (outpatient services and inpatient/observation care encounters) variables listed in the table.

BMI, body mass index; BRAF/MEKi, BRAF/MEK-targeted therapy; COPD, chronic obstructive pulmonary disease; Ipi/Nivo, ipilimumab/nivolumab; SMD, standardized mean difference.

Figure 1.

Six-month thrombotic, bleeding, and mortality outcomes comparing ipilimumab/nivolumab (Ipi/Nivo) with BRAF/MEK-targeted therapy (BRAF/MEKi) in advanced melanoma after propensity score matching. The forest plot displays hazard ratios (HRs) with 95% CIs for 6-month outcomes in the matched cohort (n = 1043 per group). The vertical reference line indicates HR = 1.0. Values to the right of 1.0 indicate higher risk with Ipi/Nivo; values to the left indicate higher risk with BRAF/MEKi. HRs were estimated using Cox proportional hazards models. DVT, deep vein thrombosis; GI, gastrointestinal; PE, pulmonary embolism; VTE, venous thromboembolism.

Figure 2.

Six-month cumulative incidence of composite venous thromboembolism (VTE) and composite bleeding in matched cohorts. Kaplan–Meier curves showing 6-month cumulative incidence of (A) composite VTE (deep vein thrombosis or pulmonary embolism) and (B) composite bleeding (including intracranial hemorrhage and gastrointestinal bleeding) in matched cohorts (n = 1043 per group) treated with ipilimumab/nivolumab (Ipi/Nivo) vs BRAF/MEK-targeted therapy (BRAF/MEKi). Shaded areas represent 95% CIs. HR, hazard ratio.

There was no observed difference in composite bleeding outcomes (9.2% vs 8.8%; HR, 1.07; 95% CI, 0.79-1.45; P = .65), ICH (5.0% vs 4.3%; HR, 1.20; 95% CI, 0.79-1.83; P = .40), or GIB (4.0% vs 4.1%; HR, 0.99; 95% CI, 0.63-1.54; P = .95) at 6 months for Ipi/Nivo compared with BRAF/MEKi. Six-month mortality was numerically higher with Ipi/Nivo but did not reach statistical significance (18.9% vs 16.1%; HR, 1.22; 95% CI, 0.99-1.51; P = .06; Figures 1 and 2B).

In a sensitivity analysis excluding patients with atrial fibrillation/flutter or baseline anticoagulation, reestimation and 1:1 rematching yielded 686 patients per group (Table 2). PE rates remained significantly higher with Ipi/Nivo (27 vs 14; HR, 1.95; 95% CI, 1.02-3.72; P = .04), with a trend toward higher VTE rates (35 vs 21; HR, 1.69; 95% CI, 0.98-2.90; P = .06) at 6 months. Six-month all-cause mortality was significantly higher with Ipi/Nivo in this subgroup (115 vs 84; HR, 1.39; 95% CI, 1.05-1.84; P = .02), while bleeding outcomes remained comparable between groups.

Table 2.

Sensitivity analysis excluding patients with baseline atrial fibrillation/flutter or anticoagulation.

Outcome	Ipi/Nivo events	Ipi/Nivo, n at risk	BRAF/MEKi events	BRAF/MEKi, n at risk	HR (95% CI)	P value
DVT	NR	686	NR	686	NR	NR
PE	27 (3.9)	686	14 (2.0)	686	1.95 (1.02-3.72)	.04
VTE (DVT + PE)	35 (5.1)	686	21 (3.1)	686	1.69 (0.98-2.90)	.06
ICH	26 (3.8)	686	24 (3.5)	686	1.09 (0.63-1.90)	.76
GIB	20 (2.9)	686	18 (2.6)	686	1.12 (0.59-2.11)	.74
Composite bleeding	46 (6.7)	686	43 (6.3)	686	1.08 (0.71-1.64)	.71
Mortality	115 (16.8)	686	84 (12.2)	686	1.39 (1.05-1.84)	.02

Matched cohorts excluding patients with baseline atrial fibrillation/flutter or long-term anticoagulation. Values are presented as event count (%) per group, with the denominator as the number of patients at risk (n = 686 per group). DVT was not reported due to limited case numbers below TriNetX privacy protection thresholds. HRs with 95% CIs were estimated using Cox proportional hazards models.

BRAF/MEKi, BRAF/MEK-targeted therapy; DVT, deep vein thrombosis; GIB, gastrointestinal bleeding; HR, hazard ratio; ICH, intracranial hemorrhage; Ipi/Nivo, ipilimumab/nivolumab; NR, not reported; PE, pulmonary embolism; VTE, venous thromboembolism.

To our knowledge, this is the first propensity-matched study to directly compare VTE and bleeding risks between first-line Ipi/Nivo and BRAF/MEKi in advanced melanoma. Ipi/Nivo was associated with a 76% higher risk of VTE compared with BRAF/MEKi in patients with advanced melanoma, driven by elevated rates of both PE and DVT, without a concomitant increase in bleeding events.

These findings are consistent with prior single-cohort studies. Moik et al. [6] reported a cumulative VTE incidence of 12.9% over a median follow-up of 8.5 months in 672 ICI-treated patients, with VTE conferring a 3-fold increase in mortality. Similarly, in a cohort of more than 12,000 patients, Sussman et al. [5] found a 6-month VTE incidence of 7.6% during ICI therapy. Our VTE rate of 7.0% in the Ipi/Nivo cohort aligns with these estimates and extends the evidence by providing a direct, propensity-matched comparison against an active comparator regimen. The stronger mortality signal in the sensitivity analysis may partly reflect unmitigated VTE-related mortality in patients not on baseline anticoagulation [5,15]. The numerically higher all-cause mortality observed in the Ipi/Nivo group in the sensitivity analysis warrants careful interpretation; it may reflect unmeasured confounding by disease aggressiveness rather than a direct causal effect of VTE. Nevertheless, early identification of high-risk patients and timely VTE management may attenuate this mortality differential, consistent with evidence that VTE treatment with low-molecular-weight heparin or direct oral anticoagulants is associated with improved outcomes in cancer patients [16].

The mechanisms underlying ICI-associated thrombosis likely involve ICI-induced T-cell activation and proinflammatory cytokine release, which upregulate tissue factor expression and activate the coagulation cascade [6]. Combined PD-1/CTLA-4 blockade may amplify this prothrombotic milieu through more robust immune activation than monotherapy, consistent with prior observations of higher VTE rates with combination ICI regimens [7]. Conversely, BRAF/MEKi may exert a relatively lower prothrombotic effect through distinct, non–immune-mediated cardiovascular pathways [8]. The absence of excess bleeding in the Ipi/Nivo cohort is clinically noteworthy, as it suggests that the elevated thrombotic risk is not counterbalanced by a concomitant hemorrhagic diathesis, thereby potentially widening the therapeutic window for thromboprophylaxis in this population. The Apixaban for the Prevention of Venous Thromboembolism in High-Risk Ambulatory Cancer Patients [17] and CASSINI [18] trials demonstrated that direct oral anticoagulants reduce VTE risk in ambulatory cancer patients receiving systemic therapy, with acceptable bleeding profiles. These findings are hypothesis-generating, but they identify Ipi/Nivo-treated patients with advanced melanoma as a potentially high-risk group for VTE. While this study does not establish an optimal thromboprophylaxis strategy, the observed increase in VTE without a corresponding increase in bleeding supports prospective evaluation of thromboprophylaxis in this population.

This study has several limitations. Death is an important competing event for VTE in this population with advanced malignancy. Because formal competing risk modeling was not available in TriNetX, death was treated as a censoring event in Cox models. All-cause mortality was numerically higher in the Ipi/Nivo cohort; therefore, competing risk bias would be expected to underestimate VTE incidence in this group. The higher observed VTE incidence with Ipi/Nivo thus occurred despite bias likely working against this association. Kaplan–Meier estimates should be interpreted cautiously, as they may overestimate absolute incidence in the presence of competing mortality. In addition, BRAF V600 mutation status, Eastern Cooperative Oncology Group performance status, corticosteroid use, and prior radiation therapy were not available for matching. Because BRAF V600 mutations status strongly influences treatment allocation, residual confounding from this and other unmeasured factors may have contributed to the higher all-cause mortality observed in the Ipi/Nivo cohort. These findings represent associations and cannot establish causality. Reliance on ICD-10 codes carries a misclassification risk, and granularity regarding anticoagulant dosing and cancer staging is limited. Specifically, the TriNetX platform records direct oral anticoagulant exposure (including apixaban and rivaroxaban) as a binary indicator of any prescription or administration, without dose, frequency, or indication-level detail, so we were unable to distinguish therapeutic from prophylactic dosing for these agents. Low-molecular-weight heparin use could not be reliably characterized because the platform does not distinguish prophylactic from therapeutic administration, and recorded exposure may reflect routine inpatient thromboprophylaxis. Although cohort identification relied on melanoma diagnosis codes (C43), all patients also received melanoma-directed systemic treatment, which provides additional reassurance that misclassification of the underlying cancer population was minimal. Inherent to the TriNetX federated electronic health record structure, granular missing data counts per variable cannot be extracted from the platform; data completeness is not quantifiable at the individual variable level. Missing covariate values are handled within TriNetX through available-case analysis, whereby matching proceeds on recorded values; any residual missingness is expected to be nondifferential between treatment groups given the parallel electronic health record capture environment, biasing results toward the null and making our estimates conservative [11]. The TriNetX network includes more than 100 healthcare organizations predominantly in the United States, with a patient mix spanning academic medical centers and community hospitals, supporting moderate generalizability to real-world oncology practice. However, the dataset overrepresents insured and health system-engaged patients and may not fully reflect outcomes in underserved or uninsured populations. Additionally, the 6-month observation window may not capture late thrombotic or bleeding events, which are particularly relevant for patients with durable responses to Ipi/Nivo. Prospective studies with rigorous thrombosis phenotyping and longer follow-up are needed to confirm these observations and define optimal prevention strategies.

Current guidelines recommend individualized VTE risk assessment in ambulatory cancer patients using validated tools such as the Khorana score [16]. Prophylactic anticoagulation with apixaban or rivaroxaban reduces VTE incidence in high-risk ambulatory patients, with a modest increase in bleeding [17,18]. Notably, the Khorana score was not originally developed with ICI-treated populations in mind, and its predictive performance in this setting remains uncertain. Whether patients initiating Ipi/Nivo for advanced melanoma would benefit from targeted thromboprophylaxis warrants prospective evaluation, given the absence of excess bleeding in our cohort. Future studies should also explore whether treatment-specific VTE risk models could improve risk stratification beyond existing tools.