Presented in part at the XXIII Congress of the International Society on Thrombosis and Hemostasis; July 23-28, 2011; Kyoto, Japan.
Recent studies suggest that thromboprophylaxis is beneficial in preventing venous thromboembolism (VTE) in cancer outpatients, but this is not widely adopted because of incomplete understanding of the contemporary incidence of VTE and concerns about bleeding. Therefore, the authors examined the incidence and predictors of VTE in ambulatory patients with bladder, colorectal, lung, ovary, pancreas, or gastric cancers.
Data were extracted from a large health care claims database of commercially insured patients in the United States between 2004 and 2009. Demographic and clinical characteristics of the cancer cohort (N = 17,284) and an age/sex-matched, noncancer control cohort were evaluated. VTE incidence was recorded during a 3-month to 12-month follow-up period after the initiation of chemotherapy. Multivariate analyses were conducted to identify independent predictors of VTE and bleeding.
The mean age of the study population was 64 years, and 51% of patients were women. VTE occurred in 12.6% of the cancer cohort (n = 2170) over 12 months after the initiation of chemotherapy versus 1.4% of controls (n = 237; P < .0001); incidence ranged by cancer type from 19.2% (pancreatic cancer) to 8.2% (bladder cancer). Predictors of VTE included type of cancer, comorbidities (Charlson Comorbidity Index score or obesity), and commonly used specific antineoplastic or supportive care agents (cisplatin, bevacizumab, and erythropoietin).
The association between cancer and venous thromboembolism (VTE), including deep vein thrombosis (DVT) and pulmonary embolism (PE), is well established.1-3 Malignancy is characterized by activation of the coagulation system, and this prothrombotic state is exacerbated further by chemotherapy, hormone therapy, and surgery.4-6 Cancer is 1 of the most common and important acquired risk factors for VTE,7 and patients with active malignancy have a 4-fold to 7-fold higher incidence of symptomatic VTE than the general population.8, 9
Historically, VTE has been observed most frequently in hospitalized cancer patients admitted for surgery or acute medical illness.10, 11 The risk of cancer-associated VTE varies markedly; however, population-based case-control studies indicate a 2-year cumulative incidence of 0.6% to 7.8%, depending on the population studied.12-14 This wide variation reflects the multitude of influences on VTE risk in cancer patients, including disease-related factors (eg, tumor type, disease stage), patient-related factors (eg, comorbidity), and treatment-related factors (eg, chemotherapy, antiangiogenic therapy).9, 15 Systemic chemotherapy increases the risk of VTE 6-fold to 7-fold,16 and the rise in cancer-associated VTE over recent decades16, 17 may have been caused in part by the introduction of therapies with direct effects on the vascular endothelium.18 It has been demonstrated that thromboprophylaxis is beneficial in reducing VTE in hospitalized, acutely ill medical patients, including those with cancer.19 Recent large studies have suggested that outpatient thromboprophylaxis also is beneficial in cancer outpatients.20, 21 However, outpatient prophylaxis has not been widely adopted clinically because of an incomplete understanding of contemporary rates of VTE.16, 22 In addition, concerns regarding the risk of bleeding may lead to low compliance with prophylaxis in the cancer population, and little is known about the contemporary prevalence of bleeding in cancer patients or the risk factors for bleeding. The primary objective of this large, retrospective cohort analysis was to examine the real-world incidence and predictors of VTE in a contemporary cohort of ambulatory patients undergoing current chemotherapy regimens for select advanced or metastatic solid tumors compared with an age-matched and sex-matched control population of noncancer patients. We chose to focus on 6 common solid tumors at high risk of VTE that were included in recent large studies of thromboprophylaxis.20, 21 We also examined the incidence and predictors of all-cause bleeding events in these patients.
MATERIALS AND METHODS
This retrospective, observational cohort study was based on health care data collected from the IMS/PharMetrics Patient-Centric database (IMS Health, Inc., New York, NY) for the period from January 2004 through December 2009. This large database provides integrated enrollment, medical, and prescription claims information from more than 90 managed care organizations and Medicare, and it represents the health services of over 58 million patients across the United States. It includes both inpatient and outpatient diagnoses (in International Classification of Diseases, Ninth Revision, Clinical Modification [ICD-9-CM] format) and procedures (in Current Procedural Terminology, Fourth Edition [CPT-4] and Health Care Common Procedure Coding System [HCPCS] formats) and outpatient prescription records. Additional data elements include patient demographics (age, sex, US Census Bureau geographic region), health plan type, payer type, provider specialty, and health plan enrollment dates. All patient records that we used were deidentified in compliance with the Health Insurance Portability and Accountability Act of 1996; therefore, the study was exempt from institutional review board overview.
Patients aged ≥18 years who had an inpatient diagnosis of malignant neoplasm of the lung (ICD-9-CM codes 162.0, 162.2-162.5, 162.8, and 162.9), pancreas (ICD-9-CM codes 157.0-157.4, 157.8, and 157.9), gastrointestinal system (ICD-9-CM codes 151.0-151.6, 151.8, and 151.9), colon/rectum (ICD-9-CM codes 153.0-154.3 and 154.8), bladder (ICD-9-CM codes 188.x), or ovary (ICD-9-CM codes 183.0, 183.2-183.5, 183.8, and 183.9) and who received cytotoxic chemotherapy between January 2005 and December 2008 (the “index event identification period”) were selected from the database. Receipt of chemotherapy was indicated by health care claims with relevant National Drug Code (NDC) or HCPCS codes. For study inclusion, patients also were required to have continuous medical and prescription drug coverage for ≥12 months before and 3 to 12 months after their earliest (index) cycle of chemotherapy during the index event identification period. Patients who received biologic agents alone, in the absence of chemotherapy, were excluded from the study. Patients who had a diagnosis of VTE (ICD-9-CM codes 415.1, 451.1, 451.2, 451.81, 451.83, 451.84, 451.9, 453.4, 453.8, and 453.9), severe renal impairment, hemorrhagic stroke, or thrombocytopenia during the 12-month preindex period, or major bleeding (including gastrointestinal or ulcer-related bleeding) within the 3-month period immediately before the index date were excluded from the study. In addition, patients who received antithrombotic/thrombolytic treatment, as indicated by health care claims with NDC and HCPCS codes for heparin, enoxaparin, dalteparin, tinzaparin, fondaparinux, bivalirudin, warfarin, alteplase, reteplase, streptokinase, or urokinase <2 weeks before the index date were excluded.
Matched control cohort
Noncancer patients who were selected from the same database were matched individually (1:1) to cancer patients on age, sex, geographic region, and medical and prescription coverage status to form a matched control cohort. The control cohort served to demonstrate the magnitude of VTE burden in real-world patients initiating chemotherapy for solid tumors. Each control patient was assigned the index date of the corresponding matched cancer patient. Control patients also were required to have continuous medical and prescription drug coverage data for ≤12 months before and ≥3 months after the index date.
Data Collection and Outcome Measures
Patient characteristics and treatment patterns
Information on patient demographics (sex, age, geographic region, and health plan) and clinical characteristics (medical conditions/comorbidities, Charlson Comorbidity Index [CCI] [an overall composite measure of the severity of illness23]), during the 12-month preindex period was collected for both patient cohorts. Medical conditions/comorbidities of interest included hypertension, stroke/transient ischemic attack, type 2 diabetes, congestive heart failure, pulmonary disease, hepatic disease, atrial fibrillation/flutter, and obesity (all identified by ICD-9-CM codes).
Receipt of anticoagulation and cancer therapy over the 3-month to 12-month postindex period was determined for both patient cohorts. Receipt of anticoagulants was indicated by the presence of an outpatient health care claim with an NDC or HCPCS code for heparin, warfarin, enoxaparin, dalteparin, tinzaparin, or fondaparinux. Receipt of cancer therapy was indicated by the corresponding NDC or HCPCS code for cisplatin, doxorubicin, carboplatin, oxaliplatin, bevacizumab, cetuximab, erythropoietin, or myeloid growth factors.
Incidence rates of VTE and bleeding events over the postindex follow-up period were determined from documented outpatient or inpatient claims with relevant ICD-9CM, CPT-4, and HCPCS codes. VTE events were separately classified as any VTE event (DVT or PE), DVT only (ICD-9-CM codes 451.1, 451.2, 451.8, 451.9, 453.4, 453.8, and 453.9), PE only (ICD-9-CM code 415.1), and DVT + PE, as previously defined.24 Major bleeding was defined based on clinical input as the presence of appropriate ICD-9-CM codes (431, 432, 767, 767.11, 362.81, 376.32, 377.42, 430, or 719.1-719.19) or CPT-4 codes (36,430 or 36,455). Minor bleeding was defined as all other bleeding. Patients who had transfusion codes alone without diagnostic codes for bleeding events (based on HCPCS codes) were excluded from counts of bleeding. Predictors of VTE and bleeding in cancer patients were determined using a multivariate regression model that included demographic variables (age, sex, health plan type) and clinical variables (preindex surgery, comorbidity, CCI score, cancer type, and index cancer therapy).
Demographic and clinical characteristics for each patient cohort were summarized with descriptive statistics. Intercohort comparisons of patient demographic, clinical, and treatment characteristics and clinical outcomes were performed with Student t tests (continuous variables) and chi-square tests (categorical variables). A P value < .05 was considered statistically significant. Logistic regression was performed for a predictor analysis. Statistical analyses were conducted using SAS version 9.2 (SAS Institute Inc., Cary, NC).
Patient Demographics and Clinical Characteristics
In total, 63,453 patients who had a diagnosis of bladder, colorectal, lung, ovarian, pancreatic, or gastric cancer who were receiving chemotherapy during the index event identification period were identified from the database. Of these, 17,284 patients with cancer fulfilled the study inclusion criteria, with the majority having either lung (38.9%) or colorectal (26.3%) cancer (Fig. 1). The control cohort comprised 17,284 noncancer patients who matched the cancer cohort on age, sex, geographic region, and enrollment status.
The mean age in both cohorts was 63.7 years, and 51.1% of patients were women (Table 1). The mean preindex CCI score was significantly higher in the cancer cohort than in the control cohort (6.3 vs 0.6; P < .0001), and the majority of individuals in the cancer cohort had a CCI score ≥5 (59.8% vs 1.8% in noncancer controls; P < .0001). Compared with the noncancer cohort, patients in the cancer cohort also had a significantly (P < .0001) higher prevalence of hypertension, stroke, diabetes mellitus, congestive heart failure, pulmonary disease, hepatic disease, atrial fibrillation, and obesity (Table 1).
Table 1. Patient Demographic and Preindex Clinical Characteristics
No. of Patients (%)
Cancer Cohort, N = 17,284
Noncancer Cohort, N = 17,284
Abbreviations: AF, atrial fibrillation; CCI, Charlson Comorbidity Index; SD, standard deviation.
Cancer was included as a comorbidity in the calculation of CCI scores.
VTE occurred significantly more commonly in the cancer cohort than in the noncancer cohort over 12 months after the initiation of chemotherapy (12.6% vs 1.4%; P < .0001) (Fig. 2). Similarly, the cancer cohort experienced a significantly higher incidence of DVT only (7.7% vs 1%; P < .0001), PE only (2.4% vs 0.2%; P < .0001), and DVT plus PE (2.5% vs 0.2%; P < .0001) compared with the control cohort. The incidence of VTE differed significantly (P < .0001) across cancer types, with the highest rates observed in patients with pancreatic cancer (19.2%) and the lowest rates observed in patients with bladder cancer (8.2%) (Fig. 3). Although DVT was most frequent in patients with pancreatic cancer (12.6%), PE was most frequent in those with lung (3.6%) and gastric (3.3%) cancer. The majority of VTE events in the cancer cohort occurred shortly after initiation of chemotherapy: among the patients with VTE, 18.1% had their first event within the first month, 47% had their first event within the first 3 months, and 72.5% had their first event within the first 6 months (Fig. 4).
In multivariate analysis, pancreatic cancer (odds ratio [OR], 5.39; 95% confidence interval [CI], 4.26-6.80), gastric cancer (OR, 4.00; 95% CI, 3.04-5.25), lung cancer (OR, 3.15; 95% CI, 2.55-3.89), a CCI score of 3 or 4 (OR, 3.60; 95% CI, 2.72-4.77), a CCI score ≥5 (OR, 4.38; 95% CI, 3.33-5.78), obesity (OR, 1.49; 95% CI, 1.23-1.82), and the receipt of erythropoietin (OR, 1.56; 95% CI, 1.41-1.73), bevacizumab (OR, 1.43; 95% CI, 1.24-1.65), or cisplatin (OR, 1.36; 95% CI, 1.19-1.55) were associated with the greatest risk of VTE (Fig. 5, top).
Bleeding Incidence and Predictors
All-cause bleeding occurred in a significantly greater proportion of the cancer cohort compared with the noncancer cohort (17.7% vs 7%; P < .0001) during follow-up. Both major bleeding events (0.9% vs 0.1%; P < .0001) and minor bleeding events (16.8% vs 6.9%; P < .0001) were more common in the cancer cohort than in the control cohort. Predictors of bleeding differed from predictors of VTE, and bladder cancer (OR, 4.09; 95% CI, 3.58-4.68; P < .0001), gastric cancer (OR, 2.16; 95% CI, 1.77-2.65; P < .0001), atrial fibrillation/flutter (OR, 1.46; 95% CI, 1.29-1.65; P < .0001), receipt of erythropoietin (OR, 1.60; 95% CI, 1.48-1.74; P < .0001), receipt of cetuximab (OR, 1.47; 95% CI, 1.15-1.87; P < .05), and receipt of bevacizumab (OR, 1.39; 95% CI, 1.23-1.57; P < .0001) were associated with the greatest risk of bleeding (Fig. 5, bottom).
The principal finding of the current population-based study is the high overall incidence of VTE (12.6%) documented in a large, contemporary, nationally representative, real-world cohort of ambulatory patients who were initiating conventional chemotherapy for common cancers (lung, gastric, pancreas, colon/rectum, bladder, and ovary). In this cohort, we identified multiple clinical risk factors, including specific cancer-associated drug treatments, for VTE (erythropoietin, bevacizumab, and cisplatin) and bleeding (erythropoietin, bevacizumab, and cetuximab). The study is important because it provides a contemporary estimate of VTE and bleeding rates in the ambulatory setting, which is where the vast majority of cancer care now occurs.
The incidence of cancer-associated VTE reported in previous epidemiologic studies varies from 0.6% to 7.8%,3, 12, 13 reflecting differences in patient population, clinical setting, duration of follow-up, receipt of chemotherapy, method of detecting VTE, and time frame of the analysis, with a higher incidence reported in more recent studies. Nevertheless, differences in the time frame do not fully explain the markedly greater overall incidence of VTE (12.6%) noted in our study. The PROTECHT study,21 which enrolled ambulatory cancer patients between 2003 and 2007, reported an incidence of thromboembolic events (a composite of symptomatic venous or arterial thromboembolic events) of 3.9% in the placebo arm (n = 381) with a treatment duration of ≤4 months. Another, more recent clinical trial (SAVE-ONCO), in which selection criteria similar to those in the current study were used, reported a rate of thromboembolic events (a composite of symptomatic DVT, nonfatal PE, and VTE-related death) of 3.4% in the placebo arm (n = 1604) with a median treatment period of 3.5 months.20 In our real-world study, the incidence of VTE among ambulatory patients with cancer over a similar timeframe (the first 4 months after initiating chemotherapy) was markedly higher at 7.2%, which may reflect the selection bias inherent in clinical trials (ie, cancer patients with better performance status and fewer comorbidities) or perhaps geographic differences in surveillance/treatment (frequency and resolution of scanning, chemotherapeutic regimens). Alternatively, it is possible that the use of ICD-9 diagnosis codes for the identification of patients with VTE may result in an overestimate of the incidence of clinical disease. It should be noted, however, that our choice of ICD-9 diagnosis codes for the identification of VTE was more selective than the methods used in several previous database studies of VTE epidemiology.25-28 In addition, the overwhelming majority of DVT events that we identified in the database were site-specific, and <5% of events fell into the “DVT with unspecified site” category (ICD-9 code 451.9). Furthermore, there is a large body of published literature to substantiate the use of ICD-9 codes as surrogate diagnostic markers for clinically documented VTE27-31 and major bleeding32-35 in retrospective claims data analyses.
The SAVE-ONCO and PROTECHT studies also provided clear evidence that prophylactic pharmacologic anticoagulation is effective in reducing clinically important VTE in selected cancer outpatients who are receiving chemotherapy. The SAVE-ONCO study, which was a randomized, placebo-controlled trial of an ultra-low-molecular-weight heparin (semuloparin) in patients who were initiating chemotherapy for common solid tumors, indicated a significant reduction in the thromboembolic event rate (1.2% vs 3.4%; P < .001) yet similar rates of major bleeding (1.2% vs 1.1%) with semuloparin compared with placebo.20 In the PROTECHT study, a low-molecular-weight heparin (nadroparin) significantly reduced thromboembolic events compared with placebo in ambulatory cancer patients who were receiving chemotherapy (2.0% vs 3.9%; P = .02). Although PROTECHT was not powered to assess differences in major bleeding events between the 2 treatment arms, 5 patients (0.7%) in the nadroparin group and no patients in the placebo group had major bleeding events.21
Our observations that the risk of VTE depends heavily on the primary site of cancer and that pancreatic, gastric, and lung cancers were significant predictive factors for developing VTE in this patient cohort are consistent with earlier epidemiologic findings.10, 12, 13, 17 Similarly, our identification of erythropoietin, bevacizumab, and cisplatin therapies as significant predictors for the development of VTE is in accordance with earlier evidence implicating platinum-based therapy,36, 37 erythropoiesis-stimulating agents,38, 39 and antiangiogenic agents18, 40, 41 in the pathogenesis of VTE. Furthermore, several risk factors, including obesity, type of cancer, and the receipt of erythropoietin, further validate variables that were included in a previously developed predictive model for chemotherapy-associated VTE.42
In the case of bevacizumab, the impact on the risk of VTE remains controversial based on data from clinical trials and meta-analyses.43-45 In the current study, bevacizumab was received by 10.6% of patients across all tumor types, and it was associated independently with a 1.43-fold greater risk of VTE. Bevacizumab was received by 25% of patients (n = 1135) in the colorectal cancer subcohort from this study, in whom it was associated with a 1.66-fold increase in the risk of VTE.46 In contrast to the more controlled environment of clinical trials, our study addresses the real-world impact of initiating bevacizumab in ambulatory patients with solid tumors while controlling for other risk factors. It is noteworthy that the US Food and Drug Administration recently amended the prescribing information for bevacizumab to include concerns regarding an increased risk of VTE.47
Like VTE, bleeding in cancer patients arises through the complex interplay of disease-related and treatment-related factors. In the current study, the high incidence of bleeding noted in the cancer cohort (approximately 3-fold higher than in the control cohort) suggests causes other than anticoagulation. Unfortunately, we were not able to adjust for anticoagulation, because antiplatelet therapy was not accurately captured. To the best of our knowledge, this is the first study to provide contemporary, real-world risk factors for bleeding in cancer patients. Predictors for bleeding differed from predictors for VTE. For instance, the type of cancer with the greatest risk of bleeding was bladder cancer, which had the lowest risk of VTE. The association of bevacizumab with bleeding is well documented.47, 48 The association of erythropoietin with bleeding has not previously been reported, but it may be indicative of the receipt of erythropoietin to treat cancer-associated anemia rather than indicating causality. The association of cetuximab with bleeding is novel and may be related to the cancer stage or site (typically, colorectal or head and neck) for which this agent is used; however, the association was significant even after adjusting for cancer site. In the context of thromboprophylaxis, it is encouraging to note that major bleeding rates were not elevated in recent large clinical trials with the use of anticoagulation.20, 21, 49, 50
Like all studies that use administrative claims databases, the current investigation has several limitations. The IMS/PharMetrics Patient-Centric database covers the commercially insured population and may be unrepresentative of those with government-funded health care and the uninsured. Information on clinical risk factors for bleeding and thrombosis (eg, tumor stage, histologic subtype, and leukocyte and platelet counts), duration of chemotherapy, and anticoagulant/chemotherapy dose was unavailable. Similarly, the specific details of bleeding events could not be determined, and it was not possible to differentiate between symptomatic versus asymptomatic VTE events. The use of claims data precludes verification of diagnoses and possibly introduces coding errors. Despite these limitations, retrospective analyses using administrative claims data provide valuable information that reflects contemporary, real-world clinical practice.
In conclusion, in the current large, contemporary analysis, we observed high rates of VTE in patients initiating chemotherapy for solid tumors. The risk of VTE was 9-fold greater among cancer patients who were receiving chemotherapy than in noncancer controls. Furthermore, the incidence of VTE (approximately 8%-19%, depending on cancer type) reported in this real-world analysis of patients with solid tumors was considerably higher than the incidence reported in recent studies of thromboprophylaxis. Improved awareness on the part of oncologists of the risks and clinical consequences of cancer-associated VTE is an essential first step toward reducing the clinical burden of this condition. Predictors of VTE differ from predictors of bleeding. Results from large, randomized studies suggest that outpatient thromboprophylaxis is feasible, safe, and effective. Targeting cancer patients who have the greatest risk of VTE with appropriate prophylaxis may reduce the clinical burden of VTE and its consequences.
We acknowledge editorial/writing support provided by Paul Lane, PhD of UBC Scientific Solutions and funded by Sanofi-Aventis US.
This study was supported by Sanofi-Aventis US.
CONFLICT OF INTEREST DISCLOSURES
Dr Khorana has received honorararia for consulting and research funding from Sanofi-Aventis US. Dr. Dalal is an employee of Sanofi-Aventis US. The other authors made no disclosures of conflict of interests.