Immune checkpoint inhibitor-associated risk for venous thromboembolism: a comprehensive analysis

Abstract

Background

The relationship between venous thromboembolism (VTE) and immune checkpoint inhibitor therapy (ICI) is unclear. This analysis investigates the incidence of and risk factors for VTE in VTE-naive patients with cancer receiving ICI treatment.

Methods

A retrospective cohort study of patients receiving any type or combination of ICI from 2009 to 2022 at Dana-Farber Cancer Institute was conducted to identify VTE occurring after initiation of ICI treatment. Cumulative incidences of VTE were determined using Fine and Gray’s methods. Associations between VTE, ICI regimens, and clinical risk factors were evaluated using propensity-score stratified, multivariable Cox proportional hazards models.

Results

In 10,638 patients without a prior history of VTE, the 6-month cumulative incidence of VTE was 7.6% (95% CI: 7.1% to 8.1%) and 11.1% (95% CI: 10.5% to 11.8%) at 12 months. Clinical risk factors included: age 15–59 (HR 1.27; 95% CI: 1.12 to 1.43; p=0.002), obesity (HR: 1.41; 95% CI: 1.16 to 1.71), and history of anticoagulation prior to ICI start (HR: 1.43; 95% CI: 1.26 to 1.61). Compared with pembrolizumab, treatment with ipilimumab/nivolumab increased the risk of VTE (HR: 1.36; 95% CI: 1.02 to 1.82), while durvalumab conveyed lower risk (HR: 0.52; 95% CI: 0.31 to 0.87). Treatment with programmed cell death ligand 1 had significantly reduced risk of VTE (HR: 0.79; 95% CI: 0.63 to 0.99) compared with programmed cell death 1 monotherapy. Dual ICI blockade with cytotoxic T lymphocyte antigen 4/PD-1 significantly increased the risk of VTE (HR: 1.43; 95% CI 1.12 to 1.84). Initiation of anticoagulation after starting ICI for indications other than VTE reduced the risk by 40% (HR: 0.60, 95% CI: 0.48 to 0.73).

Conclusions

ICI treatment appears to be independently associated with a high incidence of VTE in patients with cancer warranting further investigation.

WHAT IS ALREADY KNOWN ON THIS TOPIC

• Patients with cancer have an increased risk of venous thromboembolism (VTE) compared with the general population. Many factors contribute to the risk of VTE including cancer treatment (ie, radiation, chemotherapy, surgery), stage, and individual patient risk factors. Immune checkpoint inhibitors are associated with various toxicities, with uncertainty about the incidence of venous thromboembolism.

WHAT THIS STUDY ADDS

• We evaluated risk for VTE and risk factors in patients with cancer receiving immune checkpoint inhibitor therapy in a large cohort of patients. Findings demonstrate that 12.5% (1,328 of 10,638) of VTE-naïve patients developed VTE after starting immune checkpoint inhibitor therapy, with 6-month and 12-month cumulative incidence rates of 7.6% and 11.1%, respectively. Risk factors associated with the incidence of VTE based on multivariable models included age at immune checkpoint inhibitor start, body mass index, cancer type, elevated platelet count, history of anticoagulation use, and type or class of immune checkpoint blockade.

HOW THIS STUDY MIGHT AFFECT RESEARCH, PRACTICE OR POLICY

• Further investigation to more precisely identify patients at highest risk for VTE and to prospectively evaluate the efficacy of prophylactic anticoagulation to prevent VTE in patients receiving immune checkpoint inhibitor therapy is warranted.

Introduction

Patients with cancer have an increased risk of venous thromboembolism (VTE) compared with the general population driven by many factors including cancer type and stage, cancer treatments, and individual patient risk factors. The association of cancer treatment with VTE has been known for many years, with chemotherapy, surgery, and radiation adding to the baseline risk of the cancer itself. The incidence of VTE in patients with cancer is increasing, despite advances in treatments.¹ Immune checkpoint inhibitor therapies (ICI), first approved for use in 2011, have dramatically changed the treatment paradigm for patients with cancer. They promote continued T-cell activity against tumor cells that would otherwise evade immune response. The use of ICI has significantly impacted cancer treatment, resulting in decreased mortality even in those with advanced-stage disease. Four classes of ICI are in use or under investigation in a wide variety of cancer types. They target and inhibit cytotoxic T lymphocyte antigen 4 (CTLA-4), programmed cell death 1 (PD-1), PD ligand 1 (PD-L1), or lymphocyte activating gene 3. Immune-related adverse events (irAEs) resulting from ICI treatment can develop with a unique spectrum of symptoms; however, the risk of VTE associated with ICI is unclear. Conflicting results have been reported, with some suggesting ICI confers an increased VTE risk while others have reported no difference, with cumulative incidences lower than historic rates in ambulatory patients with solid tumors receiving chemotherapy.^{2 3} Some studies have found low incidence rates of 2.1% at 6 months while others note a 6-month cumulative incidence of VTE between 8% and 13%.^{2 4} All reports have been retrospective, frequently with small numbers of patients, single tumor types, single types of ICI, or variable inclusion of participants with past history of thrombosis,⁵ resulting in limited strength of conclusions with the added possibility of under-reporting in the prospective randomized controlled trials of ICI as VTE was not captured as a related adverse event.

Given the conflicting findings, this cohort study aimed to assess the incidence of VTE associated with ICI in a large cohort of VTE-naïve patients with cancer treated at a single institution and to investigate risk factors associated with the development of VTE. Patients with a wide variety of cancer types treated with different classes and types of ICI were included. Baseline characteristics, cancer type and stage, ICI treatments, and known factors associated with the risk of VTE were analyzed. Relevant predictors of VTE were identified using propensity score-weighted, multivariable Cox regression modeling to determine the role of ICI in the development of VTE.

Methods

Design and analysis population

This single-center retrospective cohort study was conducted at Dana-Farber Cancer Institute (DFCI) with Institutional Review Board approval and in accordance with the Declaration of Helsinki. All patients with confirmed cancer and treated with ICI at DFCI between 2009 and 2022 were identified from the OncDRS database, DFCI Protocol 17–624, using listed ICIs from the NCI_PREFERRED_MED_NM field in the Medications data set, and listed ICD codes from the DIAGNOSIS_ICD10_CD field of the electronic health record diagnosis data set (online supplemental file A).⁶ The ICI therapies included pembrolizumab, ipilimumab, nivolumab, combination ipilimumab/nivolumab, atezolizumab, durvalumab, or tremelimumab with or without durvalumab, whether investigational or with regulatory approval. Baseline characteristics of age, sex, race, body mass index (BMI), cancer type, disease stage, type and class of ICI treatment, white blood count (WBC), hemoglobin (HgB), and platelet count were collected at the time of start of ICI treatment. Patients with a history of VTE before starting ICI therapy were excluded from the cohort. Additionally, those receiving anticoagulation therapy at the time of ICI initiation were removed, whereas patients with a history of anticoagulation use were included. Initiation of anticoagulation after starting ICI treatment, but before the development of VTE, was accounted for in the analyses as a time-dependent covariate. Type and dose of anticoagulation were collected. Anticoagulation therapies included apixaban, rivaroxaban, edoxaban, dabigatran, dalteparin, enoxaparin, unfractionated heparin, fondaparinux, and warfarin.

The primary outcome was the cumulative incidence of VTE at 6 and 12 months after initiation of ICI treatment and was estimated using the methods of Gray, and Fine and Gray with death as a competing risk.^{7 8} Development of VTE after the start date of ICI treatment was determined using ICD-10 codes associated with deep vein thrombosis and pulmonary embolus obtained from the patient’s electronic health record (online supplemental file A). Individual chart review for those with a new therapeutic anticoagulant prescription was performed independently by two study team members to manually confirm imaging diagnosis of VTE corresponding to the timing of therapeutic dose anticoagulant initiation on a subset of 50 randomly selected patient charts. The Khorana VTE risk assessment score was calculated using the components of the score collected at the time of initiation of ICI.⁹

Statistical methods

The analysis estimates the cumulative incidence of developing VTE from the time of starting ICI therapy, with a competing risk of death before VTE. The follow-up of patients who did not develop VTE was censored at the date of last follow-up. Equality of cumulative incidence estimates between groups was assessed using Gray’s test.⁸ The cumulative incidence function of VTE was modeled using the technique of Fine and Gray using the subdistribution hazard in the presence of a competing risk.⁷ Candidate predictors that were considered were: age at start of ICI therapy, sex, race, BMI, cancer type, disease stage, ICI therapy, Khorana risk score, WBC, HgB, platelets, prior history of anticoagulation, and use of anticoagulation after ICI start but before VTE event. ICI therapy was explored in two ways: using the actual therapy name or classifying according to antibody type (ie, PD-1 (nivolumab, pembrolizumab), CTLA-4 (ipilimumab, tremelimumab), PD-L1 (atezolizumab, durvalumab), CTLA-4/PD-1 combination). Age at the start of ICI was divided into two groups (15–59 and 60+years). WBC, HgB, and platelets were classified as normal, above the normal range (abnormal-high), below the normal range (abnormal-low), and missing. BMI was classified as underweight (<18.5), normal (18 to <25), overweight (25 to <30), or obese (30+). The Khorana risk score was classified into low (0), moderate,^{1 2} or high (3+). If the time from ICI start to anticoagulation start was negative in the data set, the patient was classified as having a prior history of anticoagulation. Anticoagulation after ICI start, but before a VTE, was included in the modeling as a time-dependent covariate since patients could begin anticoagulation at any time after starting ICI therapy; therefore, there could be some patients who never started anticoagulation, and others who had no anticoagulation for an interval of follow-up and then began anticoagulation therapy.

To allow for differences in patient/disease characteristics among patients who did or did not receive anticoagulation, propensity scores were calculated using multivariable logistic regression. The predictors considered were the previously listed patient and disease risk factors as well as any statistically significant two-way interactions. The Cox modeling for VTE cumulative incidence was stratified by the propensity scores divided into quintiles.

Univariable and multivariable modeling

Univariate assessments of the cumulative incidence of VTE for each patient/disease/risk factor were explored using Gray’s test. Characteristics with p values of 0.1 or less were subsequently carried forward into the multivariable setting as candidate predictors. Multivariable modeling followed a two-step process. The first step included all univariate factors identified using Gray’s test with p values of 0.1 or less. The second model refit the first model after removing any predictor with p value>0.05 in the multivariable setting. Separate sets of models were fit using either ICI therapy or ICI classification. Results are presented as HRs with 95% CIs adjusted for multiple comparisons against a reference group using Dunnett’s correction. All analyses were conducted using SAS V.9.4 (SAS Institute, Cary, North Carolina, USA).

Results

Population characteristics

The initial cohort of patients treated with ICI and not on anticoagulants at the time of starting ICI included 12,007 patients. Of these, 1,343 patients were excluded from subsequent analyses due to prior history of VTE; 25 patients with follow-up durations of 0 days and 1 patient with missing data for sex were also excluded. The final cohort included 10,638 patients (figure 1). Baseline characteristics, cancer type and stage, and types of ICI treatments are summarized in table 1 for the full cohort and for those that developed VTE. Baseline characteristics for censored participants and those that died before developing VTE, as well as the components used to calculate the Khorana score, obtained at the time of initiation of ICI, are shown in online supplemental tables 1s and 2s. Anticoagulation was started in 2,863 patients on or after starting ICI for indications other than a VTE diagnosis (new onset atrial fibrillation, or thromboprophylaxis for prolonged hospital stay, others), with 2422 newly prescribed anticoagulation within the first year of starting ICI treatment. Of patients who received anticoagulation for indications other than VTE, 1,781 (62.2%) received prophylactic dosing and 355 (12.4%) received therapeutic dosing. Further information on the intensity of anticoagulation in this cohort can be found in online supplemental table 3s. Of all checkpoint inhibitors patients received, 49.6% received pembrolizumab and 26.2% nivolumab monotherapy. Of these, 9,512 (89%) had treatment with a single ICI agent at day 0 and 1 year; the remaining 11% received more than one type of ICI at one or more time points.

Figure 1. Consolidated Standards of Reporting Trials diagram of patients included in the study. 12,007 patients with cancer who received immune checkpoint inhibitor therapy. Exclusions: 1,343 patients were excluded due to prior history of VTE; 25 patients were excluded due to follow-up time of zero; one patient was excluded due to missing information on gender. Final patient cohort: 10,638 patients with cancer receiving immunotherapy. Outcomes: 1,328 developed VTE after initiation of ICI; 4,684 died prior to developing VTE, and 4,626 were censored for follow-up. ICI, immune checkpoint inhibitor; VTE, venous thromboembolism.

Table 1. Baseline demographics.

Risk factor	Full cohort		VTE only
	N	%	N	%
Gender	5,066	47.6	661	13.0
Female
Male	5,572	52.4	667	12.0
Age at ICI start (years)	3,481	32.7	502	14.4
15–59
60+	7,157	67.3	826	11.5
Race	341	3.2	38	11.1
Asian
Black	293	2.8	49	16.7
Other	488	4.6	54	11.1
White	9,516	89.5	1187	12.5
BMI	2,570	24.2	218	8.5
Missing
Normal (18.5 to <25)	3,083	29.0	361	11.7
Obese (30+)	1,967	18.5	312	15.9
Overweight (25 to <30)	2,649	24.9	395	14.9
Underweight (<18.5)	369	3.5	42	11.4
History of anticoagulation	6,648	62.5	706	10.6
No
Yes	3,990	37.5	622	15.6
Cancer type	23	0.2	1	4.3
Bone
Brain	369	3.5	27	7.3
Breast	795	7.5	118	14.8
Gastrointestinal	1,248	11.7	173	13.9
Genitourinary	1,237	11.6	167	13.5
Gynecological	383	3.6	61	15.9
Head and neck	590	5.5	54	9.2
Hematological malignancy	398	3.7	38	9.5
Lung	3221	30.3	431	13.4
Other	572	5.4	53	9.3
Sarcoma	122	1.1	22	18.0
Skin	1,680	15.8	183	10.9
Stage	508	4.8	64	12.6
0–1
2	696	6.5	90	12.9
3	1,345	12.6	189	14.1
4	2,464	23.2	388	15.7
Missing	5,625	52.9	597	10.6
ICI type	878	8.3	111	12.6
Atezolizumab
Durvalumab	465	4.4	30	6.5
Ipilimumab	468	4.4	46	9.8
Ipilimumab/nivolumab	719	6.8	117	16.3
Nivolumab	2,784	26.2	327	11.7
Pembrolizumab	5,281	49.6	689	13.0
Treme or durva/treme	43	0.4	8	18.6
ICI class	487	4.6	50	10.3
CTLA-4
CTLA-4/PD-(L)1	743	7.0	121	16.3
PD-1	8,065	75.8	1016	12.6
PD-L1	1,343	12.6	141	10.5
Khorana risk* at treatment start	834	7.8	126	15.1
High (3+)
Low (0)	4,013	37.7	434	10.8
Moderate (1–2)	5,791	54.4	768	13.3

^*Score determined by cancer type, platelet count, hemoglobin level, leukocyte count, and BMI.

BMI, body mass index; CTLA-4, cytotoxic T lymphocyte antigen 4; ICI, immune checkpoint inhibitor; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1; VTE, venous thromboembolism.

Cumulative incidence of VTE

Of all patients, 1,328 developed VTE (12.5%) after starting ICI therapy, with median follow-up time relative to the date of starting ICI therapy of 25.1 months. The overall cumulative incidence rates of VTE are summarized in figure 2. The estimate at 6 months was 7.6% (95% CI: 7.1% to 8.1%) and at 1 year was 11.1% (95% CI: 10.5% to 11.8%). At 2 years the estimated incidence of VTE was 13.8% (95% CI: 13.0% to 14.5%). Among the 1,328 patients who developed VTE, the median time to diagnosis of VTE after starting ICI therapy was 5.4 months (IQR: 1.8–12.0 months). Baseline characteristics according to development of VTE are found in table 1. 25% of patients developed VTE after 12 months of ICI treatment.

Figure 2. Cumulative incidence of VTE in the analysis population. The cumulative incidence of VTE from the time of initiation of ICI treatment with a competing risk of death before VTE. Risk table is embedded within the Kaplan-Meier plot. ICI, immune checkpoint inhibitor; VTE, venous thromboembolism.

Predictors of VTE

Results of univariable analysis of clinical factors associated with the development of VTE found a higher incidence of VTE in patients who were female, age 15–59 years, black race, overweight or obese (BMI>25 kg/m²), who had a history of anticoagulation use, high Khorana score, and high platelet count (online supplemental table 4s). The incidence of VTE also varied significantly by cancer type, with the highest incidences among patients with sarcoma, breast cancer, GI, and GYN cancers (online supplemental figure 1s). In the comparison of ICI therapies, the highest incidence of VTE was observed in patients treated with combination ipilimumab/nivolumab or durvalumab/tremelimumab, and lowest in those treated with ipilimumab or durvalumab as monotherapies (p<0.0001) (online supplemental table 4s). The first multivariable model based on ICI class, found that sex, race, and Khorana risk score were no longer significant, although given that the Khorana score is partially comprised of cancer type, BMI, and platelets, having these three as separate predictors in the model could remove the influence of the Khorana score. The model results were consistent when treatments were grouped by ICI therapy rather than by ICI class (online supplemental tables 4s and 5s). Statistically significant predictors from the first set of multivariable models were then carried forward into the second set.

Results from the multivariable model based on ICI class found an increased risk of VTE was associated with age (15–59 vs 60+years, HR 1.27; 95% CI: 1.12 to 1.43; p<0.0001) and BMI (obese vs normal, HR 1.41; 95% CI: 1.16 to 1.71; overweight vs normal, HR 1.33, 95% CI: 1.11 to 1.60, p<0.0001.) (table 2). The incidence of VTE significantly varied by cancer type (p<0.0001). Compared with patients with lung cancer, there were statistically significant reductions in the risk of VTE for patients with cancers of the brain, skin, head and neck, or hematologic malignancies (table 2). Numerical, but non-significant increases in VTE risk were noted for gynecologic cancers and cancers of the breast, GI tract, or sarcoma (table 2). The risk of VTE was associated with the class of ICI. Compared with PD-1 monotherapy, patients treated with PD-L1 monotherapy had significantly reduced hazards of VTE (HR: 0.79; 95% CI: 0.63 to 0.99). In contrast, patients treated with dual checkpoint inhibitor blockade with a combination of CTLA-4/PD-1 (ipilimumab and nivolumab) had significantly increased hazards of VTE (HR: 1.43; 95% CI: 1.12 to 1.84) (table 2, figures3 4). A high platelet count was also associated with an increased risk of VTE compared with normal platelet count (HR: 1.27; 95% CI: 0.97 to 1.67) while a platelet count lower than normal was associated with a decreased risk (HR: 0.73; 95% CI: 0.57 to 0.93, p=0.001). Although the Khorana score appeared to be associated with VTE in the univariable modeling, no association was observed in the multivariable analysis.

Table 2. Multivariable model of venous thromboembolism risks (based on ICI class).

Predictor	Comparison	HR	95% HR confidence limits (adjusted)		Wald p values
ICI class	CTLA-4 vs PD-1	0.80	0.54	1.20	<0.0001
	CTLA-4/PD-(L)1 vs PD-1	1.43	1.12	1.84
	PD-L1 vs PD-1	0.79	0.63	0.99
Cancer type	Bone vs lung	0.31	0.02	5.61	<0.0001
	Brain vs lung	0.45	0.25	0.81
	Breast vs lung	1.13	0.82	1.54
	GI vs lung	1.13	0.86	1.48
	GU vs lung	0.93	0.70	1.23
	GYN vs lung	1.14	0.75	1.71
	Head and neck vs lung	0.61	0.40	0.93
	Hematological malignancy vs lung	0.61	0.37	1.01
	Other vs lung	0.71	0.47	1.10
	Sarcoma vs lung	1.18	0.63	2.24
	Skin vs lung	0.65	0.49	0.87
BMI	Missing vs normal	0.98	0.72	1.34	<0.0001
	Obese vs normal	1.41	1.16	1.71
	Overweight vs normal	1.33	1.11	1.60
	Underweight vs normal	0.88	0.58	1.33
Age group	15–59 vs 60+	1.27	1.12	1.43	<0.0001
Platelets	Abnormal-high vs normal	1.27	0.97	1.67	0.001
	Abnormal-low vs normal	0.73	0.57	0.93
	Missing vs normal	0.91	0.72	1.15
History of anticoagulation	Yes vs no	1.43	1.26	1.61	<0.0001
Started anticoagulation post-ICI*	Yes vs no	0.60	0.48	0.73	<0.0001

^*Time-dependent covariate.

BMI, body mass index; CTLA-4, cytotoxic T lymphocyte antigen 4; GI, gastrointestinal; GU, genitourinary; GYN, gynecologic; ICI, immune checkpoint inhibitor; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1.

Figure 3. Cumulative incidence of VTE by ICI class. Cumulative incidence of VTE stratified by ICI class (CTLA-4, CTLA-4/PD-(L)1 combination, PD-1 or PD-L1 monotherapy) with death as a competing risk. Patients receiving CTLA-4/PD-(L)1 combination therapy had higher cumulative incidence of VTE while CTLA-4 monotherapy had the lowest incidence of VTE. Risk table is included in table 2. CTLA-4, cytotoxic T lymphocyte antigen 4; ICI, immune checkpoint inhibitor; PD-1, programmed cell death 1; PD-L1, programmed cell death ligand 1; VTE, venous thromboembolism.

Figure 4. Cumulative incidence of VTE by ICI type. Cumulative incidence of VTE stratified by ICI treatment type with death as a competing risk. Patients receiving ipilimumab/nivolumab had the highest cumulative incidence of VTE while ipilimumab had the lowest incidence of VTE. Risk table is included in table 3. ICI, immune checkpoint inhibitor; VTE, venous thromboembolism.

In the multivariable model based on ICI treatment, the predictors of VTE and the magnitudes of the HRs were very consistent with those from the model based on ICI class (table 3). The effect of ICI treatment was significantly associated with the risk of VTE. Using those treated with pembrolizumab as the reference, those receiving ipilimumab/nivolumab had a significantly higher risk of developing VTE (HR: 1.36; 95% CI: 1.02 to 1.82), while those receiving durvalumab had a significantly lower risk (HR: 0.52; 95% CI: 0.31 to 0.87) (table 3).

Table 3. Multivariable analysis of venous thromboembolism risk (based on ICI type).

Predictor	Comparison	HR	95% HR confidence limits (adjusted)		Wald p values
ICI therapy	Atezo vs pembro	0.89	0.67	1.17	0.0002
	Durva vs pembro	0.52	0.31	0.87
	Ipi vs pembro	0.78	0.49	1.23
	Ipi/nivo vs pembro	1.36	1.02	1.82
	Nivo vs pembro	0.90	0.75	1.09
	Treme/durva-treme vs pembro	1.22	0.48	3.10
Cancer type	Bone vs lung	0.32	0.02	5.80	<0.0001
	Brain vs lung	0.45	0.25	0.81
	Breast vs lung	1.09	0.80	1.48
	GI vs lung	1.09	0.83	1.42
	GU vs lung	0.87	0.66	1.14
	GYN vs lung	1.07	0.71	1.61
	Head and neck vs lung	0.59	0.39	0.90
	Hematological malignancy vs lung	0.59	0.36	0.98
	Other vs lung	0.73	0.47	1.12
	Sarcoma vs lung	1.11	0.59	2.11
	Skin vs lung	0.67	0.50	0.89
BMI	Missing vs normal	1.05	0.76	1.44	<0.0001
	Obese vs normal	1.40	1.16	1.70
	Overweight vs normal	1.33	1.11	1.59
	Underweight vs normal	0.88	0.58	1.33
Age group	15–59 vs 60+	1.27	1.12	1.43	<0.0001
Platelets	Abnormal-high vs normal	1.25	0.95	1.64	0.002
	Abnormal-low vs normal	0.73	0.57	0.94
	Missing vs normal	0.91	0.72	1.15
History of anticoagulation	Yes vs no	1.41	1.25	1.59	<0.0001
Started anticoagulation post-ICI*	Yes vs no	0.59	0.48	0.73	<0.0001

^*Time-dependent covariate.

BMI, body mass index; GI, gastrointestinal; GU, genitourinary; GYN, gynecologic; ICI, immune checkpoint inhibitor.

In both multivariable analyses, patients who initiated anticoagulation after starting ICI for indications other than VTE treatment experienced a 40% reduction in risk of developing VTE compared with those who had not yet or never started anticoagulation (tables2 3). Those previously treated with anticoagulation for indications other than VTE, but who were not on anticoagulation at the start of ICI treatment, had an approximate 40% increased risk of developing VTE (tables2 3).

Subanalysis of lung cancer cohort

A separate analysis was performed in the 3,221 patients with primary lung cancer to assess the incidence and VTE risk factors in the largest cohort of patients with a single cancer type. The 6-month cumulative incidence of VTE was 8.5% and 12.5% at 1 year (online supplemental figure 2s and table 6s). In univariate analysis, patients between 23 and 59 years of age had a higher incidence of VTE compared with patients 60+years (p<0.0001). As in the overall analysis population, patients with a history of anticoagulation had an increased incidence of VTE (p<0.0001). The highest incidence of VTE was observed in patients with stage IV disease (p=0.06), and in patients treated with pembrolizumab, combination ipilimumab/nivolumab or tremelimumab/durvalumab (p<0.0001). Analyzing by class of ICI, the incidence of VTE was highest with CTLA-4/PD-1 treatment compared with PD-L1 (p=0.001). In the univariable model, elevated WBC was associated with VTE (p=0.004), while platelet count, HgB levels, and the Khorana risk score (p=0.33) were not. VTE risk was not found to be associated with BMI (p=0.56), race (p=0.40), or gender (p=0.86).

In multivariate analysis of patients with lung cancer, type of ICI, WBC, age, history of prior anticoagulation and starting anticoagulation after ICI remained statistically significant risk factors for VTE (p<0.05 for all). Stage was no longer a statistically significant risk factor in multivariate analysis (p=0.15; online supplemental tables 7s and 8s).

In Cox proportional hazard models, treatment with pembrolizumab monotherapy was associated with an increased VTE risk that was twice as high compared with durvalumab (HR: 2.04; 95% CI: 1.06 to 4.00) and 43% higher compared with nivolumab (HR: 1.43; 95% CI: 1.01 to 2.00).

Patients 60+years of age had a lower risk of VTE than patients 23–59 years (HR: 0.61; 95% CI: 0.49 to 0.76). High WBC trended toward higher risk of VTE than patients with normal WBC (HR: 1.32; 95% CI: 0.98 to 1.78), although not statistically significant. A history of anticoagulation was also associated with increased risk of VTE (HR: 1.31; 95% CI: 1.06 to 1.61). As in the overall population, patients with lung cancer who started anticoagulation after ICI initiation had a 61% lower risk of VTE (HR: 0.39; 95% CI: 0.26 to 0.57) (online supplemental table 9s) compared with patients who had not yet or never started anticoagulation.

Discussion

The results of this analysis of over 10,000 VTE-naïve patients treated with ICI found an increased risk for VTE, with a cumulative incidence of 7.6% at 6 months, higher than previously published incidences. In a large population of unselected ambulatory patients receiving conventional chemotherapy, the incidence of VTE after cancer diagnosis at 6 months was 3.4%¹⁰ while in a national population database analysis the cumulative VTE incidence between 2011 and 2017 was 2.2% at 6 months and 3.0% at 12 months.¹ The 6-month cumulative VTE incidence varied by cancer type: pancreatic cancer 4.4%, Hodgkin’s lymphoma 2.9%, non-Hodgkin’s lymphoma 2.7%, and ovarian cancer 3.1%. These rates are lower than found in this analysis. Just 0.4% of that population was treated with ICI; however, the 6-month VTE incidence of 4.08% was higher than in those treated with conventional chemotherapy.¹ The large number of patients, inclusion of a variety of cancer types, and use of multiple ICI types in this study overcome many of the problems associated with past estimates of VTE risk with ICI treatment and the ambiguity and conflicting results of other published data, and provide specific data on the effects of type and class of ICI in different cancer types. This analysis included approximately three times as many patients as a recently reported study that found a low cumulative incidence of VTE of 2.6% at 6 months in unselected patients with a 5.4% incidence in those identified at higher risk based on Khorana score, and another study of approximately 2,850 patients which found similar 6-month and 12-month cumulative incidences of VTE, but lacked the longer follow-up and other details included in this analysis.^{2 11} The overall cumulative incidence of VTE of 7.6% at 6 months, determined by the Fine and Gray technique using death as a competing risk, found in our study is similar to those in the placebo groups in the primary prophylaxis trials of apixaban and rivaroxaban in which patients were selected for increased VTE risk using a Khorana score of 2 or greater with cumulative incidences of 8.8% and 10.2%, respectively, at 6 months.^{12 13} The results of the multivariable model suggest that ICI appears to be an independent predictor for VTE, along with findings that support BMI, age, and elevated platelet count as independent risks.

Results in the lung cancer cohort demonstrated VTE risk of 8.5% at 6 months and 12.5% at 1 year while receiving ICI, similar to other smaller published studies of ICI use in patients with lung cancer.^{3 14 15} The risk of VTE with PD-1 inhibitor therapy was higher compared with PD-L1 inhibitor therapy, younger age, or receipt of anticoagulation prior to ICI therapy, findings which are consistent with results in the overall cohort.

The risk of VTE with use of ICI treatments was found to be independent of other known risk factors including cancer type and cancer stage using propensity score matching and multivariable modeling. The use of a single ICI is associated with a gradient of risk for VTE. However, dual checkpoint blockade is linked to the highest cumulative incidence of VTE, indicating a potentially additive effect of combined checkpoint inhibition on VTE risk. That other identified comorbid predictors of VTE including BMI, age, and elevated platelet count were found to be predictive of VTE confirms prior findings. Age and increased BMI are characteristics associated with an increased inflammatory state while an elevated platelet count is a known reactive biomarker of increased inflammation. These baseline risks may enhance the inflammatory milieu of ICI therapy and therefore the risk of VTE.

Although the Khorana score was predictive of VTE in the univariable analysis, it was not in the multivariable models in this study. Using the individual components of cancer type, BMI, and platelet count may have removed the predictive influence of the Khorana score. However, as white blood cell count and HgB were not significantly predictive in the multivariable model, these findings may suggest that the true predictive strength of the Khorana score lies in the cancer type and baseline characteristics of increased BMI and platelet count. The Khorana score was derived and validated before the availability of ICI treatments. The findings of this analysis suggest that treatment with ICI should be considered as a predictor in VTE risk assessment models for patients with cancer.

ICI therapy, by releasing the inhibition on endogenous immune effector cells and thereby enhancing their activity against cancer cells, has provided substantial clinical benefits for patients with cancer. While ICI therapy offers significant clinical benefits, it is also associated with immune-related adverse events. These adverse effects arise from the activation of both innate and adaptive immune responses, which can activate coagulation pathways and lead to thrombosis in various clinical settings.¹⁶ It is therefore not surprizing that the use of ICI is associated with increased risk of VTE.

A critically important finding of this study is that the use of anticoagulation decreased the risk of developing VTE in the setting of ICI treatment. Whether patients should be started on prophylactic anticoagulation at the time of initiating ICI treatment is currently unknown. The ASCO guidelines for treating irAEs suggest that VTE prophylaxis in high-risk outpatients may be offered as per their separate VTE guideline, which emphasizes assessing risks and benefits for the individual patient.^{17 18} The results of this study have implications for the design of prospective studies on the utility of VTE prophylaxis by identifying that ICI itself is an independent VTE risk, as well as type of ICI, in addition to known predictors of cancer type, cancer stage, BMI, and platelet count, and past anticoagulation use. While the treatment duration in most VTE prophylaxis trials in patients with cancer is 6 months, the findings in this study found the median time to development of VTE in those treated with ICI was 5.4 months with 25% of patients developing VTE after 12 months, suggesting that longer duration of prophylaxis may be needed.

Limitations of this study include a single center population, no comparison cohort of patients not treated with ICI although the use of propensity score matching to create balanced groups should minimize confounding and bias, and the methodologic challenges associated with database analyses, including missing data and inability to completely account for confounding variables. For example, capturing the stage of disease in this population was challenging as the stage may have changed during treatment and was not coded. Mortality, not unexpected in this cancer population, was high but was adjusted for in the cumulative incidence analyses; VTE occurrence did not influence mortality in this population.

Conclusion

The findings from this large analysis of over 10,000 patients suggest that the use of ICI treatments in patients with cancer without a prior history of VTE appears to be an independent risk factor for VTE, which is mitigated by the use of anticoagulation. Further studies are needed to more precisely identify patients at highest risk for VTE and to evaluate the efficacy of prophylactic anticoagulation in preventing VTE in patients receiving ICI therapy.

Data availability statement

All data relevant to the study are included in the article or uploaded as supplementary information.