Bleeding in patients with lung adenocarcinoma receiving concurrent administration of anticoagulation and VEGF or EGFR inhibitors

Abstract

Background:

Vascular endothelial growth factor inhibitors (VEGFi) and epidermal growth factor receptor inhibitors (EGFRi) are used to treat various cancers including lung adenocarcinoma, but their associated bleeding risks when combined with anticoagulation are largely unknown.

Objectives:

We sought to determine if coadministration of VEGFi and/or EGFRi with anticoagulation in patients with lung adenocarcinoma is associated with higher clinically relevant bleeding risks than in patients receiving anticoagulation with systemic therapy alone.

Patients/Methods:

We conducted a single-center retrospective cohort study of patients treated for lung adenocarcinoma with therapeutic anticoagulation and concurrent systemic cancer therapy alone, VEGFi +/− systemic therapy, or EGFRi +/− systemic therapy. Index was the first day of concomitant anticoagulation and cancer treatment. Patients were followed for 12 months and major bleeding and clinically relevant non-major bleeding (CRNMB) outcomes were collected using an internally validated natural language processing tool.

Results:

A total of 464 patients were included comprising 156 (34%) systemic therapy alone, 208 (45%) EGFRi, and 100 (22%) VEGFi. The 12-month cumulative incidences of major bleeding were statistically similar in the 3 cohorts: VEGFi (10%, 95% CI 3.4–22%), EGFRi (11% 95% CI 6.5–17%), and systemic therapy cohort (9.3%, 95% CI 4.1–17%). We observed statistically similar risks of CRNMB with VEGFi (HR 1.51, 95% CI 0.71–3.20) and EGFRi (HR 1.35 95% CI 0.65–2.80) versus systemic therapy alone. Epistaxis was the most common site, accounting for 25% of clinically relevant events.

Conclusions:

Concomitant administration of VEGFi and anticoagulation was not associated with statistically significant increased risks of major bleeding or CRNMB.

Introduction

Vascular endothelial growth factor (VEGF) inhibitors and epidermal growth factor receptor (EGFR) inhibitors are integral components of systemic therapy across multiple malignancies, including non–small-cell lung cancer (NSCLC). The incorporation of VEGF inhibitors into standard platinum-based chemotherapy regimens for advanced-stage NSCLC confers a significant improvement in overall survival [1]. Similarly, in patients with EGFR-mutated advanced NSCLC, targeted inhibition of EGFR has demonstrated significant improvement in survival over conventional chemotherapy [2, 3].

VEGF inhibitors are associated with an increased risk of bleeding as referenced in black box warnings [4, 5]. In lung cancers specifically, VEGF inhibitors increase the risk of pulmonary hemorrhage while EGFR inhibitors do not appear to substantively influence bleeding outcomes [3, 6]. Prior studies suggest a high rate of bleeding in patients receiving concurrent anticoagulation and VEGF inhibitors but whether this risk is substantively higher than in patients receiving concurrent systemic therapy and anticoagulation is not established [7–9].

The aim of this study is to compare rates of major bleeding and clinically relevant non-major bleeding (CRNMB) among patients with lung adenocarcinoma treated with systemic therapy and/or VEGF or EGFR inhibitors while concurrently receiving therapeutic anticoagulation.

Methods

This study was approved by the Memorial Sloan Kettering Cancer Center Institutional Review Board with a waiver of informed consent. The study was performed in accordance with the Declaration of Helsinki and the manuscript is in line with the International Committee of Medical Journal Editors Recommendations for the Conduct, Reporting, Editing, and Publication of Scholarly Work in Medical Journals.

Study design and patient population

We performed a retrospective cohort study with patients treated at Memorial Sloan Kettering Center between 2012 and 2022 for lung adenocarcinoma. Patients were identified using diagnostic codes for lung cancers and were treated with therapeutic anticoagulation with low molecular weight heparin (LMWH), direct oral anticoagulants (DOACs), or vitamin K antagonists (VKAs) for any indication concurrent with cancer treatment (systemic cancer therapy, VEGF inhibitor, and/or EGFR inhibitor). Patients were included regardless of time from cancer diagnosis and time on anticoagulation. Prophylactic dosing of LMWH or heparin only were excluded (e.g. enoxaparin 40 mg daily for prophylaxis of deep vein thrombosis in hospitalized patients). Cancer treatments and anticoagulation were captured via prescription data, including length of treatment, from inpatient records as well as outpatient appointments. Brain metastases were reported based on the diagnosis code “secondary malignant neoplasm of brain” prior to or after indexing. Patients were indexed on the first day of concurrent cancer treatment and anticoagulation and grouped into one of three cohorts based upon the type of cancer therapy at index: systemic therapy alone (systemic therapy), EGFR inhibitor +/− systemic therapy (EGFRi), or VEGF inhibitor +/− systemic therapy (VEGFi). Systemic therapy included chemotherapy, targeted therapy (not including VEGF or EGFR inhibitors), and immunotherapy. Changes in cancer treatment during the follow-up period were recorded and patients who changed cancer treatments remained in the cohort they were classified at study index. Patients were not censored when changing treatments.

Outcomes

Measured outcomes included major bleeding and clinically relevant non-major bleeding based on the International Society of Thrombosis and Haemostasis (ISTH) definitions [10, 11]. Patients were followed for 12 months after index and censored for major bleeding, discontinuation of either anticoagulation or cancer therapy (60 days after the last prescription), and death. Since prescriptions are generally filled for at least 30 days, we chose to censor patients 60 days after the last prescription for anticoagulation or cancer therapy to account for prolonged impact on hemostasis, delays in prescription fills, and missed doses. Patients were not censored for CRNMB events and thus a major bleeding event could occur following a CRNMB event, but CRNMB events could not occur after a major bleeding event. Additionally, in the time-to-event analysis, if patients had multiple CRNMB events only the first was recorded. Intracranial hemorrhage (ICH) was considered a secondary outcome.

Outcomes were identified using the Clinical Event Detection and Recording System (CEDARS) [12]. CEDARS is a machine learning and natural language processing (NLP) platform used to identify potential bleeding events within the electronic health record of the target cohort, including inpatient notes, outpatient notes, and imaging reports. CEDARS does this by screening each record for pre-determined bleeding keywords, as listed in Supplement 1 [12, 13]. Each medical record instance that includes a keyword is reviewed by a physician for possible endpoint inclusion based on the definition of CRNMB or major bleeding. The type of bleed and additional details were noted. All bleeding events labeled as major bleeding were further adjudicated by a second hematologist. Any disagreements in labeling were resolved by discussion among the authors.

After screening a prespecified proportion of potential subjects (resulting in 293 patients), to validate the accuracy of the CEDARS pipeline for bleeding events, we compared the accuracy of NLP, manual chart review, and ICD codes to a “gold standard” that combined the manual, NLP, and a third physician adjudicator. All event adjudicators were blinded to exposure group. Manual chart review was performed independently by two physicians and demonstrated a 63% (95% CI 50–93%) sensitivity and 96% (95% CI 93–98%) specificity for detecting all bleeding events compared to 83% sensitivity (95% CI 71–91%) and 97% specificity (95% CI 93–99%) using the NLP-platform and 47% sensitivity (95% CI 34–60) and 90% specificity (95% CI 85–94) using ICD codes (Supplemental Table 1) [14]. However, ICD codes are unable to reliably differentiate between major bleeding and clinically relevant non-major bleeding. Given the improved accuracy, NLP (via the CEDARS platform) was utilized to identify outcomes in the entire cohort for this study.

Statistical analysis

Patient characteristics were summarized using frequencies for factor variables and medians with interquartile ranges for continuous variables. Major bleeding and CRNMB were analyzed using a time-to-event analysis, with death as a competing risk and with end of follow-up at 12 months or set to 60 days after the last prescription of anticoagulation or anti-cancer therapy or last clinical visit (whichever is the first event). Cumulative incidences of the events are estimated at 3, 6, and 12 months for each exposure group with 95% confidence intervals (CI). Hazard ratios (HR) with corresponding 95% CIs were then calculated to quantify relative risk of the event under the cause-specific Cox proportional hazard model. Then, patient characteristics that showed evidence of association with the outcomes of interest (i.e. p<0.05), as well as clinically relevant variables, were selected for multivariable model for further evaluation and adjusted hazard ratios (aHR) were reported. Using the one in ten rule, we included a total of 3 variables in the multivariable analysis for major bleeding (age, platelet count, and renal disease) and a total of 6 variables in the multivariable analysis for CRNMB (age, timing of anticoagulation start compared to cancer treatment, platelet count, renal disease, type of anticoagulation, and cancer stage) [15].

Since prior published data has associated VEGFi with higher bleeding risks, we compared our VEGFi cohort to a combined EGFRi and systemic therapy cohort [7–9, 16]. We also explored the various combinations of anti-cancer therapy, including with and without systemic therapy. These secondary analyses included all patients. Recognizing that bleeding risk is generally highest immediately after starting anticoagulation, we calculated hazard ratios for major bleeding and CRNMB among patients who started anticoagulation before versus after starting anti-cancer therapy and performed an exploratory analysis on the subgroup of patients we found to have a numerically higher risk of major bleeding (patients who started anticoagulation before anti-cancer therapy). Understanding that bleeding may be distributed unequally among sites, we performed a subanalysis on major bleeding, CRNMB, and clinically relevant bleeding (major bleeding and CRNMB) among the three most common bleeding sites among the cohorts. Lastly, we performed a subgroup analysis of intracranial hemorrhage in patients with prior brain metastases. All statistical tests were two-sided, with P < 0.05 considered significant. All analyses were conducted using R software (version 4.3.2) [17].

Results

Patient characteristics

The study included 464 patients with a median age of 65 years (IQR 58–73 years) and a median follow-up of 6.1 months (95% CI 5.3–7.1). Baseline characteristics are detailed in Table 1. A total of 288 (62%) patients had stage IV lung cancer and 61 (13%) had brain metastases at the time of index. The median time from cancer diagnosis to study index was 9.6 months (IQR 2.3–33.0 months). Cancer treatment groups included 156 (34%) patients in the systemic therapy cohort, 208 (45%) in the EGFRi cohort, and 100 (22%) in the VEGFi cohort. Among the EGFRi cohort, 186 (89%) received EGFR inhibitors alone and among the VEGFi cohort, 28 (28%) received VEGF inhibitors alone. Chemotherapy was the most common type of systemic therapy used in each cohort. The types of systemic therapies used are shown in Table 1 and Supplemental Table 2. Among the VEGFi cohort, 14/100 (14%) patients switched to an EGFRi during study follow-up and among the EGFRi cohort, 15/208 (7%) patients switched to a VEGFi during study follow-up. The 12-month overall survival across cohorts was 72%, comprising 72% in the systemic therapy cohort, 74% in the EGFRi cohort, and 65% in the VEGFi cohort.

Table 1:

Baseline characteristics of cancer treatment group at index^†

Characteristic		Systemic Therapy n (%)	EGFRi ± Systemic Therapy n (%)	VEGFi ± Systemic Therapy n (%)
Patients		156 (34)	208 (45)	100 (21)
Age (IQR)		68 (61–76)	65 (58–73)	61 (53–69)
Female sex		88 (56)	136 (65)	59 (59)
Body mass index, median (IQR)		26.4 (23.8–29.0)	25.6 (23.2–30.0)	27.3 (24.0–30.4)
Race	Asian/Far-East/Indian	12 (8)	45 (22)	6 (6)
	Black	11 (7)	19 (9)	12 (12)
	White	125 (80)	129 (62)	76 (76)
	Other/unknown	8 (5)	15 (7)	6 (6)
Time from cancer diagnosis to index† in months, median (IQR)		8.9 (2.2–33.6)	12.8 (2.4–34.3)	5.9 (2.1–21.9)
Renal disease		10 (6)	15 (7)	9 (9)
Cancer stage at index†	1	25 (16)	20 (10)	6 (6)
	2	12 (8)	12 (6)	4 (4)
	3	27 (17)	13 (6)	13 (13)
	4	78 (50)	137 (66)	73 (73)
	Unknown	14 (9)	26 (12)	4 (4)
Brain metastases at index†		15 (10)	28 (13)	18 (18)
Type of anticoagulation at index†	DOAC	62 (40)	102 (49)	52 (52)
	LMWH	90 (58)	101 (49)	48 (48)
	VKA	4 (2)	5 (2)	-
Systemic therapy agents at index†‡§	Targeted therapy	37 (24)	9 (4)	2 (2)
	Immunotherapy	34 (22)	4 (2)	1 (1)
	Chemotherapy	83 (54)	15 (7)	89 (89)
EGFR/VEGF inhibitors at index†	Erlotinib	-	62 (30)	-
	Cetuximab	-	1 (1)	-
	Osimertinib	-	145 (69)	-
	Bevacizumab	-	-	73 (73)
	Ramucirumab	-	-	27 (27)

^†Index date is defined as the first day of concurrent therapeutic anticoagulation and cancer treatment

^‡The most common types of systemic therapies are included in table 1. The complete list of systemic agents is in the supplement

^§Some patients were on multiple agents, thus sums within groups may not equal the group n

EGFRi, epidermal growth factor receptor inhibitor; VEGFi, vascular endothelial growth factor inhibitor; IQR, interquartile range; NOS, not otherwise specified; DOAC, direct oral anticoagulant; LMWH, low-molecular-weight heparin; VKA, vitamin K antagonist; EGFR, epidermal growth factor receptor; VEGF, vascular endothelial growth factor

There were 62 (40%) and 90 (58%) patients in the systemic therapy cohort who received DOACs and LMWH, respectively; 102 (49%) and 101 (49%) in the EGFRi cohort; and 52 (52%) and 48 (48%) in the VEGFi cohort. Nine patients across all cohorts received VKAs. The median time on anticoagulation during study follow-up was 3.5 months (IQR 1.4–8.4). Anticoagulation was initiated before cancer treatment (i.e. before index) in 99 (63%) patients in the systemic therapy cohort, 139 (67%) in the EGFRi cohort, and 61 (61%) in the VEGFi cohort. The remainder started anticoagulation at index date. Baseline characteristics were numerically similar across the cancer treatment groups except for race and cancer stage at index. More patients in the VEGFi and EGFRi cohorts had stage 4 cancer, compared to the systemic therapy cohort. In the EGFRi cohort, there were more patients who identified as Asian (n=45, 22% of the cohort) compared with 12 (8%) of the systemic therapy cohort and 6 (6%) of the VEGFi cohort (Table 1).

Major bleeding

There were 32 (7%) major bleeding events: 9 (6%) in the systemic therapy cohort, 17 (8%) in the EGFRi cohort, and 6 (6%) in the VEGFi cohort. The cumulative incidence of major bleeding overall was 6.7% (95% CI 4.4–9.5%) at 6 months and 10% (95% CI 6.8–14%) at 12 months (Figure 1A). The 6- and 12-month cumulative incidence of major bleeding was 5.5% (95% CI 2.3–11%) and 9.3% (95% CI 4.1–17%) in the systemic therapy cohort, 8.8% (95% CI 5.0–14%) and 11% (95% CI 6.517%) for EGFRi, and 4.4% (95% CI 1.4–10%) and 10% (95% CI 3.4–22%) for VEGFi (Figure 1B).

Figure 1: Cumulative incidences of major bleeding events.

This figure shows the cumulative incidence of major bleeding events across all cohorts (A) and by treatment cohort (B) with death as a competing event.

EGFRi, epidermal growth factor receptor inhibitor; VEGFi, vascular endothelial growth factor inhibitor

Univariable Cox-regression revealed no statistically significant difference in major bleeding across the groups (p=0.6). Compared to systemic therapy alone, the unadjusted hazard ratio for major bleeding at 12 months was 1.42 (95% CI 0.63–3.19) in the EGFRi cohort and 0.94 (95% CI 0.332.65) in the VEGFi cohort. We did not observe statistically significant associations between other variables and major bleeding in univariable analysis (Table 2). In multivariable analysis including treatment cohort, age, presence of renal disease, and platelet count, there was no statistically significant difference in major bleeding between the EGFRi cohort and the systemic therapy cohort (aHR 1.29; 95% CI 0.54–3.05) nor between the VEGFi cohort and the systemic therapy cohort (aHR 0.71; 95% CI 024–2.08). There was no statistically significant association between major bleeding and any of the other clinical variables included in the multivariable model (Supplemental Table 4).

Table 2:

Univariate analysis of the predictors of major and clinically relevant non-major bleeding

Variable	Major Bleeding		CRNMB
	HR (95% CI)	p-value	HR (95% CI)	p-value
Treatment cohort		0.6		0.12
Systemic therapy	1.00		1.00
EGFRi	1.42 (0.63–3.19)		1.40 (0.71–2.75)
VEGFi	0.94 (0.33–2.65)		2.09 (1.03–4.23)
Treatment cohort		0.5		0.069
Systemic therapy + EGFRi	1.00		1.00
VEGFi	0.76 (0.31–1.85)		1.70 (0.98–2.96)
Treatment cohort		0.6		0.10
Systemic therapy	1.00		1.00
EGFRi + VEGFi	1.26 (0.58–2.71)		1.64 (0.88–3.05)
Treatment cohort		>0.9		0.15
EGFRi alone	1.00		1.00
Systemic therapy + EGFRi	1.09 (0.25–4.78)		2.20 (0.82–5.89)
Treatment cohort		0.7		0.092
VEGFi alone	1.00		1.00
Systemic therapy + VEGFi	0.68 (0.12–3.70)		0.44 (0.18–1.11)
Start of anticoagulation		0.2		0.7
At index†	1.00		1.00
Before index†	1.65 (0.76–3.58)		0.89 (0.52–1.52)
Type of anticoagulation		0.5		0.021
DOAC	1.00		1.00
LMWH	0.98 (0.49–1.95)		0.53 (0.31–0.91)
VKA	NE		NE
Age		0.6		0.019
<55 years	1.00		1.00
55–64 years	1.76 (0.57–5.40)		0.38 (0.19–0.78)
>65 years	1.31 (0.44–3.96)		0.45 (0.24–0.83)
Cancer stage		0.6		0.036
1	1.00		1.00
2	0.89 (0.16–4.88)		NE
3	0.56 (0.10–3.09)		1.09 (0.32–3.78)
4	1.10 (0.38–3.18)		1.59 (0.63–6.12)
Unknown	0.34 (0.04–3.04)		1.86 (0.57–6.12)
Brain metastases at index †		0.12		0.6
Not present	1.00		1.00
Present	1.98 (0.89–4.42)		0.83 (0.38–1.84)
Renal disease		0.2		0.7
Not present	1.00		1.00
Present	2.23 (0.78–6.35)		0.80 (0.25–2.57)
Platelet count		0.10		0.030
≥450,000 /uL	1.00		1.00
150–449,000 /uL	0.79 (0.23–2.64)		1.26 (0.45–3.52)
<150,000 /uL	3.03 (0.68–13.6)		4.72 (1.32–16.8)

^†Index date is defined as the first day of concurrent therapeutic anticoagulation and cancer treatment

CRNMB, clinically relevant non-major bleeding; HR, hazard ratio; EGFRi, epidermal growth factor receptor inhibitor; VEGFi vascular endothelial growth factor inhibitor; DOAC, direct oral anticoagulant; LMWH, low molecular weight heparin; VKA, vitamin K antagonist; NE, not estimable; uL, microliter

In a cohort combining the EGFRi and VEGFi groups, the incidence of major bleeding at 12 months was 10% (95% CI 6.6–15%). Compared to the systemic therapy cohort, in the combined EGFRi and VEGFi cohort the HR was 1.26 (95% CI 0.58–2.71) for major bleeding. Compared to patients who received EGFRi without systemic therapy the HR for major bleeding was 1.09 (95% CI 0.25–4.78) among patients who received EGFRi with systemic therapy. Compared to patients who received VEGFi without systemic therapy the HR for major bleeding was 0.68 (95% CI 0.12–3.70) among patients who received VEGFi with systemic therapy.

In the 99 patients who initiated anticoagulation before cancer treatment, the 12-month cumulative incidence of major bleeding was 13% (95% CI 7.7–19%) with a HR of 1.65 (95% CI 0.76–3.58) when compared to patients who started anticoagulation following cancer treatment. In the subgroup of patients who started anticoagulation first, the risk of major bleeding was similar between EGFRi (HR 1.30, 95% CI 0.51–3.30) and VEGFi (HR 0.83, 95% CI 0.24–2.86) versus systemic therapy alone. (Supplemental Figure 1, Supplemental Table 3).

Clinically relevant non-major bleeding

There were 56 (12%) CRNMB events across the cohort, comprising 13 (8%) in the systemic therapy cohort, 24 (12%) in the EGFRi cohort, and 19 (19%) in the VEGFi cohort. The cumulative incidence of CRNMB among all patients was 12% (95% CI 9.0–16%) at 6 months and 17% (95% CI 13–22%) at 12 months (Figure 2A). The respective 6 and 12-month cumulative incidence of CRNMB was 9.0% (95% CI 4.4–16%) and 13% (95% CI 7.2–22%) for systemic therapy alone, 12% (95% CI 7.3–17%) and 15% (95% CI 9.7–22%) for EGFRi, and 18% (95% CI 10–26%) and 26% (95% CI 15–39%) for VEGFi (Figure 2B).

Figure 2: Cumulative incidences of clinically relevant non-major bleeding events.

This figure shows the cumulative incidence of clinically relevant non-major bleeding events across all cohorts (A) and by treatment cohort (B) with death as a competing event.

EGFRi, epidermal growth factor receptor inhibitor; VEGFi, vascular endothelial growth factor inhibitor

Compared to systemic therapy alone, in univariable analysis there was more CRNMB among those receiving VEGFi (HR 2.09, 95% CI 1.03–4.23) but not among those in the EGFRi cohort (HR 1.40, 95% CI 0.71–2.75). DOAC treatment, younger age (<55 years), higher cancer stage, and thrombocytopenia (platelet count <150,000 /uL) were also associated with CRNMB in univariable analysis (Table 2). Adjusting for these variables, as well as renal disease and timing of starting anticoagulation, in multivariable analysis there was no statistically significant difference in CRNMB between VEGFi and systemic therapy cohorts (aHR 1.51, 95% CI 0.71–3.20) or between EGFRi and systemic therapy cohorts (aHR 1.35, 95% CI 0.65–2.80). Platelet count <150,000 /uL was significantly associated with CRNMB, which had an aHR of 4.65 (95% CI 1.26–17.1) when compared to platelet count ≥450,000 /uL. When comparing platelet count of 150–450,000 /uL to ≥450,000 /uL there was no statistically significant association (aHR 1.57; 95% CI 0.55–4.48). Additionally, age <55 years remained significantly associated with CRNMB compared to ages 55–64 years, but not when compared to >65 years (Supplemental Table 4).

In a combined cohort of the systemic therapy and EGFRi groups the 6 and 12-month cumulative incidence of CRNMB was 10% (95% CI 7.2–14%) and 14% (95% CI 10–19%), respectively. In multivariable analysis, there was no statistically significant difference in the 12-month incidence of CRNMB between the VEGFi cohort and the combined systemic therapy and EGFRi cohort (aHR 1.42, 95% CI 0.80–2.54). In a cohort combining the EGFRi and VEGFi cohorts, the 12-month incidence of CRNMB was 19% (95% CI 14–25%). Compared to the systemic therapy cohort, the combined EGFRi and VEGFi cohort had a HR of 1.64 (95% CI 0.88–3.05). Compared to patients who received EGFRi without systemic therapy the HR was 2.20 (95% CI 0.82–5.89) for CRNMB among patients who received EGFRi with systemic therapy. Compared to patients who received VEGFi without systemic therapy the HR was 0.44 (95% CI 0.18–1.11) among patients who received VEGFi with systemic therapy.

Patients who initiated anticoagulation before cancer treatment experienced a 12-month cumulative CRNMB incidence of 17% (95% CI 11–23%) with a HR of 0.89 (95% CI 0.52–1.52) when compared to patients who started anticoagulation following cancer treatment. In univariable analysis, among patient who started anticoagulation first, the risk of CRNMB was statistically similar across the VEGFi (HR 2.40, 95% CI 0.99–5.80) and EGFRi (HR 1.09, 95% CI 0.44–2.66) cohorts compared to systemic therapy (Supplemental Figure 2, Supplemental Table 5).

Site of bleeding

There were 88 combined major bleeding and CRNMB events across all treatment groups with the most common sites being epistaxis in 22 patients (25%), gastrointestinal in 18 (21%), and central nervous system in 16 (18%). Most of the epistaxis events were CRNMB (19 of 22; 86%) and 13 of the 22 (59%) epistaxis events occurred in the VEGFi cohort (Table 3). The 12-month cumulative incidence of clinically relevant bleeding (combined major bleeding and CRNMB) due to epistaxis across all cohorts was 6.8% (95% CI 4.2–10%) and 19% (95% CI 8.6–32%) in the VEGFi cohort with a statistically higher rate (HR 9.92, 95% CI 2.23–44.0) on univariable analysis compared to the systemic therapy cohort. No additional variables that we analyzed were predictive of epistaxis in the univariable analysis (Supplemental Table 6). Among all patients with clinically relevant epistaxis, the majority of events occurred within 3 months post-index. Among VEGFi patients with clinically relevant epistaxis the majority of events occurred within 2 months post-index (Supplemental Figure 3). The 12-month cumulative incidence of clinically relevant gastrointestinal bleeding across cohorts was 4.8% (95% CI 2.7–7.6%) with no significant differences between groups, and hazard ratios of 0.88 (95% CI 0.29–2.61) among the EGFRi cohort and 0.93 (95% CI 0.26–3.29) among the VEGFi cohort compared to the systemic therapy cohort. No additional variables that we analyzed were predictive of gastrointestinal bleeding in the univariable analysis (Supplemental Table 6).

Table 3:

Distribution of bleeding sites according to cancer treatment cohort^†‡

Site	Systemic Therapy			EGFRi +/− Systemic Therapy			VEGFi +/− Systemic Therapy
	MB	CRNMB	Combined‡	MB	CRNMB	Combined‡	MB	CRNMB	Combined‡
Total [n (%)]	9 (100)	13 (100)	22 (100)	17 (100)	24 (100)	41 (100)	6 (100)	19 (100)	25 (100)
Gastrointestinal	2 (22)	4 (31)	6 (27)	3 (18)	5 (21)	8 (20)	1 (17)	3 (16)	4 (16)
Hemoptysis	-	2 (15)	2 (9)	1 (6)	4 (17)	5 (12)	-	1 (5)	1 (4)
Central Nervous System	4 (45)	2 (15)	6 (27)	5 (29)	2 (8)	7 (17)	3 (50)	-	3 (12)
Epistaxis	1 (11)	1 (8)	2 (9)	1 (6)	6 (25)	7 (17)	1 (17)	12 (63)	13 (52)
Musculoskeletal	-	-	-	1 (6)	1 (4)	2 (5)	-	-	-
Thoracic	-	3 (23)	3 (14)	3 (18)	1 (4)	4 (10)	-	1 (5)	1 (4)
Surgical	1 (11)	-	1 (5)	1 (6)	-	1 (2)	-	-	-
Other/Unknown	1 (11)	1 (8)	2 (9)	2 (11)	5 (21)	7 (17)	1 (17)	2 (11)	3 (12)

^†Percentages represent the percentage of bleeds within treatment group and specific type of bleed (MB/CRNMB/combined)

^‡Represents the sum of MB and CRNMB

MB, major bleeding; CRNMB, clinically relevant non-major bleeding

Intracranial hemorrhage

There were 12 (3%) intracranial hemorrhages across the cohort including 4/156 (3%) in the systemic therapy group, 5/208 (2%) in the EGFRi cohort, and 3/100 (3%) in the VEGFi cohort. The 12-month cumulative incidence of intracranial hemorrhage across the groups was 3.9% (95% CI 2.0–6.8%). Univariable Cox regression revealed no statistically significant differences in the cumulative incidence of intracranial hemorrhage across the groups (p>0.9). The hazard ratio for intracranial hemorrhage compared to the systemic therapy group was 0.91 (95% CI 0.24–3.39) for EGFRi and 0.99 (95% CI 0.22–4.41) for VEGFi.

Among the 12 intracranial hemorrhages, 2 occurred in the 61 patients with brain metastases at index, both occurring within 3 months post-index. Intracranial hemorrhage occurred in 1 of 15 patients (7%) with brain metastases in the systemic therapy cohort (25% of all intracranial hemorrhages in this cohort), in 1/28 (4%) in the EGFRi cohort (20% of intracranial hemorrhages in this cohort), and in 0/18 patients in the VEGFi cohort.

Among the 61 patients with brain metastases at index, the 6-month cumulative incidence of intracranial hemorrhage across all cohorts was 3.4% (95% CI 0.62–11%). Specifically, among the 18 patients receiving VEGFi and anticoagulation, the 6-month cumulative incidence of ICH was 0%, compared with 3.7% (95% CI 0.25–16%) in the EGFRi cohort (n=28), and 6.7% (95% CI 0.38–27%) in the systemic therapy alone cohort (n=15).

Bleeding events by type of anticoagulant

Among the 32 major bleeding events overall, 15 patients were receiving a DOAC and 17 LMWH. There were no major bleeding events among 9 patients on a VKA. The 6- and 12-month cumulative incidences of major bleeding events were 5.5% (95% CI 2.9–9.4%) and 10% (95% CI 5.6–17%) for the DOAC cohort and 7.8% (95% CI 4.5–12%) and 10% (95% CI 5.9–16%) for the LMWH cohort. The HR for major bleeding at 12 months of LMWH compared to DOAC was 0.98 (95% CI 0.49–1.95).

Among the 56 CRNMB events overall, 34 patients were receiving a DOAC, 22 LMWH, and 0 VKA. The 6- and 12-month cumulative incidences of CRNMB events were 16% (95% CI 11–21%) and 22% (95% CI 15–29%) for the DOAC cohort and 9.4% (95% CI 5.7–14%) and 13% (95% CI 8.3–19%) for the LMWH cohort. Accordingly, there were numerically fewer CRNMB at 12 months with LMWH compared to DOAC (aHR 0.59, 95% CI 0.34–1.03) (Supplemental Table 7).

Discussion

Anticoagulation is commonly administered in patients with lung cancer either for the management of acute venous thromboembolism or atrial fibrillation [18, 19]. While it is known that all anticoagulants increase the risk of hemorrhage, the relative safety is often influenced by concurrent medications [20, 21]. VEGF inhibitors are known to increase the risk of hemorrhage with higher rates reported with concurrent anticoagulation. We aimed to address whether this increased risk of hemorrhage with VEGF inhibitors with anticoagulation was comparable to the risk of hemorrhage observed with other antineoplastic agents administered with anticoagulation. We observed that treatment with VEGF or EGFR inhibitors with or without systemic therapy did not confer an increased risk of major bleeding compared to systemic therapy alone in patients with lung adenocarcinoma also receiving therapeutic anticoagulation.

Meta-analyses have shown VEGF inhibitors to be associated with an increased bleeding risk in patients not concurrently receiving anticoagulation[19, 22, 23]. Some studies have reported higher rates of bleeding when VEGF inhibitors are administered concurrently with anticoagulation compared to anticoagulation alone [7, 9]. For instance, a small retrospective cohort study compared outcomes for 86 patients with various types of cancer who received VEGF inhibitors with a factor Xa inhibitor versus Xa inhibitor alone. The hazard ratio for clinically relevant bleeding (major bleeding plus CRNMB) was 2.45 (95% CI 1.28–4.69). However, in this study when the patients received a Xa inhibitor without a VEGF inhibitor they did not receive any additional cancer treatment including systemic therapy [7]. Similarly, in a cohort study of 184 patients, those receiving VEGF inhibitors without anticoagulation had a major bleeding rate of 3.3% compared with 8.1% in patients receiving VEGF inhibitors with anticoagulation (p=0.026) [8]. In the current study we observed no significant differences in major hemorrhages among patients receiving anticoagulation and either VEGFi, EGFRi, or systemic therapy alone.

In a post-hoc analysis of the Caravaggio study, which compared outcomes for apixaban and dalteparin in patients with cancer and acute venous thromboembolism, there was not a significant difference in the risk of major bleeding among those receiving VEGF inhibitors compared to other anticancer agents (relative risk 1.58, 95% CI 0.69–3.68) [9]. This analysis included patients across a spectrum of tumor types; considering that VEGF inhibitors are administered only in certain types of tumors (e.g., lung, melanoma, renal cell carcinoma), it is difficult to discern whether bleeding rates were confounded by tumor types rather than type of antitumor therapy administered. Reassuringly, our current analyses suggest that the combination of anticoagulation with VEGF inhibitors does not increase the risk of major bleeding when compared to a similar cohort of patients with lung adenocarcinoma who received systemic therapy alone or EGFR-inhibitors. The CRNMB incidence, primarily epistaxis, was numerically higher with VEGF inhibitors.

The safety of anticoagulation is a particular concern in patients with brain tumors [24]. In retrospective cohorts, anticoagulation does not appear to increase the risk of bleeding in patients with brain metastases [25]. While VEGF inhibitors do not appear to increase the risk of ICH in the absence of anticoagulation in primary or metastatic brain tumors, there is concern regarding the safety of anticoagulation in patients with primary brain tumors concurrently receiving VEGF inhibitors [26, 27]. In a retrospective study of 282 patients treated with the VEGF inhibitor bevacizumab for glioma, those receiving concurrent anticoagulation had an ICH rate of 11% compared to 1% in those receiving bevacizumab without anticoagulation [27]. There is limited data on the safety of anticoagulation with VEGF inhibitors in patients with metastatic brain tumors, specifically lung cancer. Reassuringly, among the 18 patients in our study with brain metastases receiving VEGFi and anticoagulation there were no cases of intracranial hemorrhage. However, indications for anticoagulation and other non-cancer therapy medications, which could influence the results, were unavailable and not included in our analyses.

Limitations

There are inherent limitations to a retrospective cohort analysis. To limit potential confounding due to intrinsic differences in outcomes and cancer biology, we only included patients with lung adenocarcinoma. However, we recognize that residual confounding is possible; for instance, we cannot exclude the possibility that patients receiving VEGF inhibitors were only prescribed anticoagulants if they were considered lower risk of bleeding. Additionally, retrospective analyses of bleeding events are often challenging due to the spectrum of bleeding severity and potential human error involved in chart review. We implemented an NLP platform which demonstrated improved sensitivity (83%) and specificity (97%) for identifying overall bleeding events compared to manual chart review alone. We acknowledge that NLP for bleeding is challenging and did not achieve 100% sensitivity. Future work is needed to continue to improve the accuracy of the NLP platform for identifying bleeding events. The study may have been underpowered to detect certain associations due to the sample size, thus we cannot definitively exclude differences in bleeding incidence across the anti-cancer treatment groups. The intention-to-treat approach may also have biased some effect estimates due to the crossover between the VEGFi and EGFRi cohorts. The cohorts also included different systemic agents and different proportions of chemotherapy, targeted therapy, and immunotherapy. It is possible that the differences in these agents between cohorts had varying impacts on bleeding rates.

Conclusions

In conclusion, our findings demonstrate no statistically significant increase in the rate of major bleeding with VEGF inhibitors versus systemic therapy in patients with lung adenocarcinoma on anticoagulation. We observed a numerically higher rate of CRNMB with VEGF inhibitors compared to systemic therapy in this population, with epistaxis being the most common bleeding site. While there were a limited number of patients with brain metastases, it is reassuring that no intracranial hemorrhages were observed in this subset of patients receiving anticoagulation and VEGF inhibitors.