923
Are Venous Thromboembolic Events Associated with Subsequent Breast and Colorectal Carcinoma Diagnoses in the Elderly? A Case–Control Study of Medicare Beneficiaries
Blase N. Polite, M.D., M.P.P.1 Elizabeth B. Lamont, M.D., M.S.2,3
BACKGROUND. Multiple epidemiologic studies have reported associations between venous thromboembolic events and subsequent cancer diagnoses, but the published results have not suggested clear cancer screening approaches.
1
METHODS. Using data from the National Cancer Institute’s Surveillance, Epidemiology, and End Results-Medicare Program, the authors identified patients who were diagnosed with breast and colorectal carcinoma (n ⫽ 7166 patients) and compared them with a noncancer control group (n ⫽ 126,668 patients) according to their history of hospitalization for deep vein thrombosis (DVT) or pulmonary embolism (PE) in Medicare claims files over the previous 24 months. Using logistic regression analysis, the authors calculated the odds of receiving a diagnosis of breast carcinoma or colorectal carcinoma in the 24 months after admission for DVT or PE. RESULTS. Patients who were hospitalized for DVT or PE had nearly 3.0 times the odds of being diagnosed with colorectal carcinoma (odds ratio [OR], 2.83; 95% confidence interval [95% CI], 1.92– 4.17) and ⬎ 1.5 times the odds of being diagnosed with breast carcinoma (OR, 1.78; 95% CI, 1.05–3.02) in the subsequent 24 months. CONCLUSIONS. Because hospitalization for DVT or PE is associated with an increased risk of a breast or colorectal carcinoma diagnosis in the subsequent 2 years, physicians should be vigilant in assessing the cancer screening status of patients with new DVT and/or PE to be certain that they are up to date with recommended breast and colorectal screening guidelines. Cancer 2006;106: 923–30. © 2006 American Cancer Society.
Section of Hematology-Oncology, Department of Medicine, University of Chicago, Chicago, Illinois.
2
Department of Medicine and Institute of Technology Assessment, Massachusetts General Hospital, Boston, Massachusetts.
3
Department of Healthcare Policy, Harvard Medical School, Boston, Massachusetts.
Supported by a grant from the National Institutes of Health (grant K07 CA93892 to E.B.L.). For this study, the linked Surveillance, Epidemiology, and End Results-Medicare data base was used. The interpretation and reporting of these data are the sole responsibility of the authors. The authors acknowledge the efforts of the Applied Research Program of the National Cancer Institute; the Office of Information Services and the Office of Strategic Planning, Centers for Medicare and Medicaid Services; Information Management Services, Inc.; and the Surveillance, Epidemiology, and End Results (SEER) Program tumor registries in the creation of the SEER-Medicare data base. They also thank Dr. Francis Cook of the Brigham and Women’s Hospital; participants of the Dana Farber/ Harvard Cancer Center Cancer Outcome Research Program Seminar for their helpful suggestions; and Dr. Diane Lauderdale of the University of Chicago and Dr. Nicholas Christakis of Harvard Medical School for their careful review of the article. Address for reprints: Blase N. Polite, M.D., M.P.P., Section of Hematology-Oncology, Department of Medicine, The University of Chicago Medical Center, 5841 South Maryland Avenue, MC 2115, Chicago, IL 60637-1470; Fax: (773) 702-3002; Email:
[email protected] Received March 22, 2005; revision received August 16, 2005; accepted September 20, 2005.
KEYWORDS: venous thrombosis, pulmonary embolism, colorectal neoplasms, breast neoplasms.
D
espite its origin in the 19th century, the Virchow triad remains a clear and logical way to organize the mechanisms that contribute to thrombosis in patients with cancer. Stasis, endothelial damage, and hypercoagulability each can contribute to the development of blood clots in patients along the cancer continuum, from ambulatory patients with occult malignancies, to patients undergoing aggressive antitumor therapies (e.g., surgery, chemotherapy, hormone therapy, radiotherapy), and to patients who are debilitated and at the end of life. The causative link between cancer and hypercoagulability is multifactorial and includes tumor elaboration of tissue factor (a transmembrane glycoprotein believed active in angiogenesis and metas-
© 2006 American Cancer Society DOI 10.1002/cncr.21672 Published online 12 January 2006 in Wiley InterScience (www.interscience.wiley.com).
924
CANCER February 15, 2006 / Volume 106 / Number 4
tasis formation) and cancer procoagulant (a cysteine protease that directly activates factor X).1 Because of the potential for thromboses to serve as markers for underlying malignancy, numerous groups have sought to characterize the potential association between deep vein thrombosis (DVT) or pulmonary embolus (PE) and an occult malignancy in patients. These range from a small, prospective cohort study that involved ⬍ 100 patients to a much larger, population-based, retrospective analysis that involved ⬎ 60,000 patients.2,3 Several of the prospective cohort studies have shown an increased risk of subsequent cancer in the first 1–2 years after an idiopathic DVT or PE.4 –7 This increased risk was confirmed in two large, population-based European studies.3,8 However, several other studies either failed to show any significant link between DVT or PE and a subsequent diagnosis of cancer or demonstrated only that the link existed for a 6-month period.9 –12 The heterogeneity of results may relate, in part, to the heterogeneity of tumors and patients included in the analyses. Because of the heterogeneity of tumors involved in these studies, no clear tumor-specific cancer screening recommendations for ambulatory patients who have a new DVT or PE have emerged from these studies. We used population-level data to evaluate the strength of the association between development of a DVT or PE and the subsequent diagnosis of two common and screening-amenable malignancies, breast and colorectal carcinoma, among a population sample of elderly American patients to inform screening practices in this group of patients.
MATERIALS AND METHODS Data Sources We used data from the National Cancer Institute’s (NCI) Surveillance, Epidemiology, and End Results (SEER)-Medicare Program to study the association between thromboembolic events (i.e., DVT or PE) and subsequent cancer diagnoses in elderly Medicare beneficiaries. The SEER-Medicare data base is an NCIsponsored linkage of the clinical data collected by the SEER registries with health services billing claims collected by Medicare for administrative purposes. The SEER Program collects information regarding the diagnosis and treatment of patients with cancer from 11 geographically diverse tumor registries to monitor trends in incidence and survival. It is estimated that approximately 14% of the American population with cancer is represented in these data.13 Prior research has shown that, in the aggregate, patients in these registries are representative demographically of the general population.14 The SEER Program collects demographic information as well as
detailed information regarding initial diagnoses, including the date of diagnosis (i.e., month and year), disease site, histology, stage of tumor at diagnosis, and date of death. To facilitate comparisons between Medicare enrollees with and without cancer, the NCI has created a data file that identifies a 5% random sample of Medicare beneficiaries who reside in SEER areas but who do not have cancer files in SEER. We used that 5% random sample of Medicare beneficiaries to identify elderly patients without histories of cancer to serve as our control group. Medicare is a federally sponsored health insurance program that is administered by the Centers for Medicare and Medicaid Services (CMS), the beneficiaries of which include ⬎ 96% of all American citizens age 65 years and older.15 CMS maintains billing records of outpatient, inpatient, home health, hospice, and other claims for all beneficiaries who are not enrolled in risk contract health maintenance organizations (HMOs). In 1993, approximately 17% of Medicare beneficiaries were enrolled in a Medicare managed-care plan.16 To determine the study population’s inpatient and outpatient medical services (regardless of whether they were in the SEER Program or not), we used three types of Medicare files for the analyses: the Medicare Provider Analysis and Review (MEDPAR) file, the Outpatient Standard Analytic File, and the National Claims History file.
Definition of Cases and Controls (Disease ascertainment) We used the SEER file to identify male and female patients who had been diagnosed with pathologically confirmed, Stage I–IV colorectal adenocarcinoma and female patients who had been diagnosed with pathologically confirmed, Stage I–IV breast carcinoma using a modified AJCC stage. We included all patients who were diagnosed at or after age 67 years during a 12month period (i.e., between January 1, 1993 and December 30, 1993), who were entitled to Medicare Part A, and who were not enrolled in an HMO. The date of disease ascertainment was considered the date of diagnosis for patients in the case group, and the date of diagnosis was operationalized as the first day of the month (i.e., January through December) and year (i.e., 1993) of diagnosis, as reported in SEER. We used the SEER-Medicare 5% random sample of Medicare patients to identify appropriate noncancer control patients. Similar to the patients in our case group, we required that patients in the control group were at least age 67 years during the period between January 1, 1993 and December 30, 1993, were entitled to Medicare Part A, and were not enrolled in an HMO. The 5% random sample of Medicare patients contains patients who reside in the SEER regions but who do
Thromboemboli and Cancer/Polite and Lamont
not have SEER records for cancer. However, to further ensure that patients in the control group did not include patients who had the diseases of interest (i.e., breast and colorectal carcinoma) diagnosed in nonSEER regions, we removed from the control group those patients who had any International Classification of Diseases, ninth revision, clinical modification (ICD-9-CM) diagnoses of breast carcinoma (ICD-9 code 174) or colorectal carcinoma (ICD-9 codes 153– 154.0, and 197.5) in inpatient or outpatient Medicare files in the 2 years prior to their inclusion in the analysis (n ⫽ 256 patients). The date of disease ascertainment was January 1, 1993 for all patients in the control group.
Exposure Ascertainment For each case and control, we used the MEDPAR file to identify a discharge diagnosis of DVT or PE (PE: ICD-9 codes 415.11– 415.19; DVT: ICD-9 codes 451.1, 451.11, 451.19, 451.81, 453.2 451.2, 453.8, and 453.9) in the 2 years prior to cancer diagnosis in the case group and in an equivalent calendar period (i.e., 2 yrs prior to January 1, 1993) for the control group, in a manner consistent with prior investigations.17,18 Prior work using these codes has estimated that the annual incidence among all Medicare beneficiaries ages 65– 69 years is 1.8 per 1000 beneficiaries for DVT and 1.3 per 1000 beneficiaries for PE,18 and the annual incidence of DVT and/or PE among Medicare beneficiaries is 2.5 per 1000 beneficiaries for colorectal carcinoma and ⱖ 1.0 per 1000 beneficiaries for breast carcinoma.17
Definition of Other Explanatory Variables The additional variables patient age and race were taken from the Medicare denominator file for cases and controls. Because individual-level indicators of socioeconomic status are not available in the Medicare files, we used median income for ZIP code of residence according to U.S. Census data, transformed into quartiles, and treated categorically.19,20 Patients’ disease-related variables of tumor site, histology, and stage and the dates of diagnoses were ascertained from the SEER file.
925
However, there are at least two alternative hypotheses. Rather than the malignancy itself causing the clot, the hospital admission for treatment of the clot may have led to the cancer diagnosis, either through the process of the admission history (e.g., reports of weight loss, black tarry stool, breast discharge or mass) and physical examination (e.g., breast examination, rectal examination, stool guaic, laboratory studies) or, in the case of colorectal carcinoma, through clinically evident hemorrhage of an occult gastrointestinal tumor during hospitalization related to anticoagulation for the clot. In the first instance, the association between DVT or PE and the subsequent diagnosis of occult malignancy would be the result of a “hospitalization effect”; and, in the second case, it would be the result of an “anticoagulation effect.” To evaluate the extent to which any associations that we observed between a DVT or PE and a subsequent diagnosis of cancer were the result of hospitalization or anticoagulation effects (i.e., nonmechanistic), we sought to determine associations between cancer diagnoses and other antecedent discharge diagnoses that may serve as proxies for these alternative hypotheses (i.e., pneumonia [ICD-9 codes 481 and 483] and atrial fibrillation [ICD-9 code 427.3]) for each case and control patient (in the 2 yrs prior to cancer diagnosis in the case group and in an equivalent calendar period for the control group). We reasoned that hospitalization for “pneumonia” would measure the “hospitalization effect” and, because it often is associated with anticoagulation, that hospitalization for atrial fibrillation would measure the “anticoagulation effect.” The positive predictive value of ICD-9 coding for community-acquired pneumonia was reported previously as 93%.21 The sensitivity and specificity of ICD-9 coding for atrial fibrillation have been estimated at 87% and 100%, respectively.22 If the variance related to a subsequent cancer diagnosis could be captured similarly by these alternative hypothesis effects (i.e., coefficient on DVT or PE hospitalization the same in size as pneumonia hospitalization and/or atrial fibrillation hospitalization), then the mechanistic relation between thrombosis and occult cancer would be less clear.
Evaluation of Alternative Hypotheses Although we were interested primarily in estimating the association between hospitalization for DVT or PE and subsequent cancer diagnosis, we undertook two analyses to evaluate the possible mechanistic relation underlying the association. We hypothesized that the malignancy caused an increased propensity of blood to clot, leading to a DVT or PE, and the associated clinical symptoms that would lead the patient to seek medical attention and subsequent hospitalization.
Sensitivity Analyses To further evaluate possible hospitalization effects and, by extension, to estimate the duration of any association, we undertook sensitivity analyses through which we reestimated our logistic regression models after serially excluding those patients whose exposure (i.e., hospitalization for a DVT and/or PE) occurred within 1 month, 3 months, 6 months, 12 months, 18 months, and 21 months of disease ascertainment (i.e.,
926
CANCER February 15, 2006 / Volume 106 / Number 4
TABLE 1 Characteristics of Patients with Colorectal Carcinoma and Breast Carcinoma and Respective Noncancer Control Groups Colorectal carcinoma
Variable Mean age (yrs) Race (proportion) White Black Other Unknown Female (proportion) Mean income No. of patients (cases/ 1000 beneficiaries) with antecedent hospitalization for DVT or PE Pneumonia Atrial fibrillation
Patients with colorectal carcinoma (n ⴝ 3601 cases)
Noncancer group (n ⴝ 126,668 controls)
77.1
76.6
0.89 0.06 0.05 ⬍ 0.01 0.55 $36,7401
0.85 0.07 0.07 0.01 0.63 $35,655
28 (7.8/1000) 23 (6.4/1000) 147 (41.0/1000)
328 (2.6/1000) 1019 (8.0/1000) 4136 (33.0/1000)
Breast carcinoma Patients with breast carcinoma (n ⴝ 3565 cases)
Female noncancer group (n ⴝ 80,347 controls)
76.5
77.3
⬍ 0.001 ⬍ 0.001
0.91 0.05 0.03 0.01 1.00 $37,299
0.85 0.07 0.07 0.01 1.00 $35,528
NA ⬍ 0.001
⬍ 0.001 0.271 0.007
15 (4.2/1000) 12 (3.4/1000)) NA
191 (2.4/1000) 591 (7.4/1000) NA
0.031 0.006 NA
P value ⬍ 0.001 ⬍ 0.001
P value ⬍ 0.001 ⬍ 0.001
DVT: deep vein thrombosis; PE: pulmonary embolism; NA: not available.
the diagnosis date for the case group, January 1, 1993 for the control group). Specifically, we sought to determine whether the majority of cancer diagnoses occurred coincident with or close in time to the hospitalization (consistent with a hospitalization effect or nonmechanistic relation) or whether the association was longer lasting and persisted after discharge for DVT or PE (more consistent with a mechanistic relation). In these analyses, we were interested affirmatively in stability of the coefficient describing the association between DVT or PE hospitalization and subsequent cancer diagnosis, not in the statistical significance of the series of estimates. That is, because these sequential analyses necessarily resulted in sequential loss of exposed cases and controls, the confidence intervals (CI) necessarily increased in these sensitivity analyses.
years. All analyses were performed using STATA software (version 8.0; StataCorp LP, College Station, TX).
RESULTS Of the 133,834 elderly Medicare beneficiaries who were included in the current analysis, 3601 beneficiaries (3%) had a diagnosis of colorectal carcinoma, 3565 beneficiaries (3%) had a diagnosis of breast carcinoma, and 126,668 beneficiaries (94%) constituted the noncancer control group. The characteristics of the patients in the colorectal carcinoma case group and the corresponding control group (both genders) and of the patients in the breast carcinoma case group and the corresponding (female) control group are shown in Table 1. On average, for both disease types, patients in the case groups more often were white and had a higher mean income compared with their respective control groups.
Statistical Analyses In a series of analyses, we calculated the odds ratios (OR) for a cancer diagnosis in the 24 months after an admission for DVT or PE, atrial fibrillation, and pneumonia for the cases and controls by using logistic regression analysis. Estimates were adjusted for age, gender (as appropriate), race, and income quartile. To make inferences regarding the potential utility of cancer screening in patients whose cancers were associated with thromboembolic events, we used the SEER file to determine the stage at presentation for cases according to history of DVT or PE in the previous 2
DVT and/or PE and Colorectal Carcinoma Overall, during the 24-month ascertainment period, 28 of 3601 patients with colorectal carcinoma (annual incidence, 3.9 per 1000), compared with 328 of 126,668 noncancer control patients (annual incidence, 1.3 per 1000), had ICD-9-CM codes that indicated an antecedent hospitalization for a DVT or PE (P ⬍ 0.001). The unadjusted odds ratio (OR) of an antecedent thromboembolic event for patients who had newly diagnosed colorectal carcinoma, compared with the control group, was 3.02 (95% CI, 2.05– 4.45); and, when it
Thromboemboli and Cancer/Polite and Lamont
927
TABLE 2 Association between Hospitalization for DVT/PE, Atrial Fibrillation, or Pneumonia and a Diagnosis of Breast Carcinoma or Colorectal Carcinoma after 24 Months OR (95% CI) Colorectal carcinoma
DVT/PE AFIB Pneumonia
Breast carcinoma
Crude OR
Adjusted ORa
Crude OR
Adjusted ORa
3.02 (2.04–4.45) 1.23 (0.97–1.55) 0.79 (0.52–1.20)
2.83 (1.92–4.17) 1.16 (0.98–1.38) 0.75 (0.50–1.14)
1.87 (1.05–3.00) 0.98 (0.81–1.19) 0.46 (0.26–0.81)
1.78 (1.05–3.02) 1.01 (0.83–1.23) 0.48 (0.27–0.85)
DVT: deep vein thrombosis; PE: pulmonary embolism; OR: odds ratio; 95% CI: 95% confidence interval; AFIB: atrial fibrillation. a The odds ratio was adjusted for age, gender (for colorectal carcinoma patients), race, and income.
was adjusted for demographics, the OR was 2.83 (95% CI, 1.92– 4.17) (Table 2).
DVT and/or PE and Breast Carcinoma Overall, during the 24-month ascertainment period, 15 of 3565 patients with breast carcinoma (annual incidence, 2.1 per 1000), compared with 191 of 80,347 female noncancer control patients (annual incidence, 1.2 per 1000), had ICD-9-CM codes that indicated an antecedent hospitalization for a DVT or PE (P ⫽ 0.031). The unadjusted OR of an antecedent thromboembolic event for patients who had with newly diagnosed breast carcinoma, compared with the female control group, was 1.87 (95% CI, 1.05–3.00); and, when it was adjusted for demographics, the OR was 1.78 (95% CI, 1.05–3.02) (Table 2).
Evaluation of Alternative Hypotheses An antecedent admission for pneumonia was found in 23 of the 3601 patients with colorectal carcinoma (0.6%) compared with 1019 of the 126,668 patients in the noncancer control group (0.8%; P ⫽ 0.271). The unadjusted OR of an antecedent admission for pneumonia in patients who had newly diagnosed colorectal carcinoma, compared with the control group, was 0.79 (95% CI, 0.52–1.20); and, when it was adjusted for demographics, the OR was 0.75 (95% CI, 0.50 –1.14). An antecedent pneumonia was found in 12 of the 3565 patients with breast carcinoma (0.3%). compared with 591 of the 80,347 women in the female noncancer control group (0.7%; P ⫽ 0.006). The unadjusted OR of an antecedent admission for pneumonia in patients who had newly diagnosed breast carcinoma, compared with the control group, was 0.46 (95% CI, 0.26 – 0.81); and, when it was adjusted for demographics, the OR was 0.48 (95% CI, 0.27– 0.85). An antecedent discharge for atrial fibrillation was found in 4.1% of patients with colorectal carcinoma
(147 of 3601 patients), compared with 3.3% of patients in the noncancer control group (4136 of 126,668 patients; P ⫽ 0.007). The unadjusted OR of an antecedent admission for atrial fibrillation in patients who had with newly diagnosed colorectal carcinoma, compared with the control group, was 1.23 (95% CI, 0.97– 1.55); and, when it was adjusted for demographics, the OR decreased to 1.16 (95% CI, 0.98 –1.38) (Table 2). When indicators for prior admission for DVT/PE and prior admission for atrial fibrillation both were included in the same logistic regression that modeled the diagnosis of colorectal carcinoma, only a history of antecedent DVT and/or PE was an important and significant predictor of a diagnosis of colorectal carcinoma. The adjusted coefficient on prior DVT/PE admission was 2.76 (OR, 2.76; 95% CI, 1.87– 4.08), and the coefficient on prior atrial fibrillation admission was 1.13 (OR, 1.13; 95% CI, 0.96 –1.34). Confidence intervals on the coefficients did not overlap, and a formal test of equality of coefficients had a chi-square value of 15.8, corresponding to a ⬍ 0.1% chance that the coefficients were the same. Finally, in an attempt to determine whether all cancer diagnoses occurred coincident with or close in time to the hospitalization and whether the association persisted after discharge, we conducted a series of logistic regression analyses that modeled cancer diagnosis in which we excluded patients in the case and control groups who were found to have the exposure (i.e., DVT or PE) within 1 month, 3 months, 6 months, 12 months, 18 months, and 21 months prior to their disease ascertainment (i.e., case or control status) (Table 3). For the patients with breast carcinoma, the magnitude of the association remained fairly stable across the observation period, suggesting that a hospitalization effect was not etiologic and that the duration of the risk was not fleeting. For the patients with colorectal carcinoma, the magnitude of the associa-
928
CANCER February 15, 2006 / Volume 106 / Number 4
TABLE 3 Temporal Stability of the Association between a Diagnoses of Colorectal Carcinoma and Breast Carcinoma and Prior DVT/PEa Exposure to disease ascertainment interval
Colorectal carcinoma
Breast carcinoma
Look-back start time (mos)
Look-back duration (mos)
Adjusted OR (95% CI)
No. of patients at risk
Adjusted OR (95% CI)
No. of patients at risk
T0
24
2.83 (1.92–4.17)
130,267
1.78 (1.05–3.02)
83,911
23 21 18 12 6 3
2.87 (1.93–4.27) 2.23 (1.40–3.56) 2.20 (1.32–3.66) 2.30 (1.28–4.14) 1.99 (0.80–4.93) 2.00 (0.62–6.43)
130,248 130,210 130,165 130,091 129,994 129,962
1.86 (1.10–3.16) 1.53 (0.83–2.83) 1.64 (0.86–3.12) 1.39 (0.61–3.18) 1.47 (0.46–4.75) 1.76 (0.42–7.43)
83,902 83,878 83,852 83,807 83,752 83,733
(disease ascertainment) T⫺1 T⫺3 T⫺6 T ⫺ 12 T ⫺ 18 T ⫺ 21
DVT: deep vein thrombosis; PE: pulmonary embolism; OR: odds ratio; 95% CI: 95% confidence interval. a This table describes the temporal stability of the association between a diagnosis of colorectal and breast carcinoma and a prior deep vein thrombosis/pulmonary embolism. To do this, cases and controls with exposure noted within 1 month, 3 months, 6 months, 12 months, 18 months, and 21 months of ascertainment of disease status were dropped sequentially from the sample, and the odds were reestimated in each successively smaller sample with logistic regression. All analyses were adjusted for age, gender (in patients with colorectal carcinoma), race, and income similar to the primary analyses reported in Table 2.
tion decreased by approximately 20% over the first 12 months to 2.30, suggesting that there is a persistent and important association between DVT or PE hospitalization and a subsequent diagnosis of colorectal carcinoma but that some part of the magnitude of the original estimate of the association (i.e., OR, 2.83) may have been related to a hospitalization effect.
Stage at Disease Presentation Of the patients who had colorectal carcinoma with a prior thromboembolic event, 86% (24 of 28 patients) presented with curable stages of disease (i.e., Stage I, Stage II, or Stage III). Among the patients with colorectal carcinoma who had no prior thromboembolic events, the same proportion (86%; 3081 of 3573 patients) presented with curable stages of disease (i.e., Stage I, Stage II, or Stage III). Similarly, among the patients who had breast carcinoma with a prior thromboembolic event, 100% (15 of 15 patients) presented with curable stages of disease (i.e., Stage I, Stage II, or Stage III).
DISCUSSION In this case– control study, we analyzed data regarding ⬎ 130,000 elderly Medicare beneficiaries and found that those who were hospitalized for a DVT or PE had nearly 3 times the odds of being diagnosed with mostly curable stages of colorectal carcinoma and nearly 2 times the odds of being diagnosed with curable stages of breast carcinoma in the subsequent 2 years. Therefore, in the current study, we identified subgroups of patients who should be screened aggressively for colorectal carcinoma and breast carcinoma. One of the strengths of this study is that using the
unique SEER-Medicare data set allowed us to strengthen the mechanistic association between DVT or PE and subsequent cancer diagnosis by subordinating the likelihood of a hospitalization effect (i.e., the greater diagnostic scrutiny given to hospitalized patients) or an anticoagulation effect (i.e., the effect of anticoagulation on a previous occult colorectal malignancy) as an explanation for our main findings. That is, we found no significant, positive correlation between a discharge diagnosis of pneumonia and a subsequent diagnosis of colorectal or breast carcinoma and no significant correlation between a discharge diagnosis of atrial fibrillation and a subsequent diagnosis of colorectal carcinoma. The reduced OR noted for the correlation between an admission for pneumonia and a subsequent diagnosis of breast carcinoma may have been the result of competing comorbidities (and associated mortality) in the study population and of the significant rates of in-hospital and 30-day mortality, all of which may make it less likely that subsequent malignancies would be found.23 Why the same effect was not observed for colorectal carcinoma is unclear; however, because the OR was in the same direction, the lack of significance may be a statistical artifact. Although this somewhat limits the use of a pneumonia discharge to measure the hospitalization effect, other surrogate discharge measures suffer similar problems. It is noteworthy that the lack of a positive association combined with the atrial fibrillation data, which measure both the hospitalization and anticoagulation effects, strengthens the mechanistic correlation between a DVT or PE and a subsequent diagnosis of cancer. Finally, the relative stability of the measure of association across these periods for both
Thromboemboli and Cancer/Polite and Lamont
types of cancer also argues against the possibility that a hospitalization effect explains most of the association reported. Several investigators have explored the question of whether and how a patient who has a thromboembolic event should be screened for an underlying malignancy.4 – 6,8,11,12 Although the recommendations vary, in general, the findings indicate that most cancers can be detected by history, physical, basic laboratory studies, and a chest X-ray with subsequent testing as dictated by any abnormal results.12 However, findings from the only prospective, randomized study of extensive screening versus standard screening indicated that extensive screening, which included mammography and the option of colonoscopy, was very sensitive for detecting occult cancers and for allowing cancers to be diagnosed in a shorter time frame and at an earlier stage, including two lymph node-negative colorectal and breast carcinomas among the extensive screening group versus one lymph node-positive breast carcinoma and two metastatic colorectal carcinomas found in the control group.24 Unfortunately, because of poor accrual, that study was underpowered to demonstrate an effect on cancer-related mortality. Previous population-based analyses from Europe have shown up to a four-fold increased risk of developing colorectal carcinoma and a two-fold increased risk of developing breast carcinoma among patients who had a previous DVT/PE.3,8 The current study adds further population-based evidence from the U.S., confirming this increased risk among elderly Medicare beneficiaries; and we believe that our results support screening these patients for colorectal and breast carcinoma. In practice, this means that any patient with a DVT or PE who has not received appropriate screening for breast or colorectal carcinoma within the recommended time intervals should be screened. Although it may be argued that all elderly Americans who have a reasonable life expectancy should be screened anyway, in practice, a substantial proportion of elderly Americans are deficient in screening for breast and colorectal carcinoma.25 However, for colorectal carcinoma, the physician workforce that would be needed to remedy this through endoscopic screening simply does not exist.26 Therefore, we propose that physicians use this comorbidity of DVT/PE as a heuritistic to determine which patients should be screened both immediately and in accordance with guideline recommendations and that consideration should be given to the use of colonoscopy in this higher risk patient subset.27,28 There are limitations to the current study related to our data source. In claims data, race coding was limited, information on HMO enrollees was available
929
inconsistently, and many clinical variables were lacking.29 –32 Specifically, the limited clinical information regarding each patient hindered our ability to account for potentially important differences between the two groups, such as the use of estrogens, recent trauma, and systemic disorders such as antiphospholipid syndrome or congenital deficiencies in protein function, which may have predisposed one group versus the other to develop a DVT or PE. In addition, the socioeconomic status measure we employed was suboptimal, because the ZIP code represents an indeterminate combination of both individual and neighborhood socioeconomic characteristics.33–35 In summary, we observed a strong and statistically significant correlation between hospitalization for DVT and/or PE and subsequent diagnoses of breast and colorectal carcinoma in the elderly. These results suggest that physicians should be vigilant in assessing the cancer screening status of patients with new DVT and/or PE to be certain that they are up to date with recommended breast and colorectal screening guidelines.
REFERENCES 1.
Gale JA, Gordon SG. Update on tumor cell procoagulant factors. Acta Haematol. 2001;106:25–32. 2. Monreal M, Casals A, Boix J, Olazabal A, Montserrat E, Mundo MR. Occult cancer in patients with acute pulmonary embolism. Chest. 1993;103:816 – 819. 3. Baron JA, Gridley G, Weiderpass E, Olof N, Linet M. Venous thromboembolism and cancer. Lancet. 1998;351:1077–1080. 4. Prandoni P, Lensing A, Buller HR, et al. Deep-vein thrombosis and the incidence of subsequent symptomatic cancer. N Engl J Med. 1992;327:1128 –33. 5. Bastounis EA, Karayiannakis AJ, Makri GG, Alexiou D, Papalambros EL. The incidence of occult cancer in patients with deep venous thrombosis: a prospective study. J Intern Med. 1996;239:153–156. 6. Monreal M, Llamazares J, Perandreu J, Urrutia A, Sahuquillo JC, Contel E. Occult cancer in patients with venous thromboembolism: which patients, which cancers. Thromb Haemost. 1997;78:1316 –1318. 7. Hettiarachchi R, Lok J, Prins MH, Buller HR, Prandoni P. Undiagnosed malignancy in patients with deep vein thrombosis. Cancer. 1998;83:180 –185. 8. Sorensen HT, Mellemjaer L, Steffensen FH, Olsen JH, Nielsen GL. The risk of diagnosis of cancer after primary deep venous thrombosis or pulmonary embolism. N Engl J Med. 1998;338:1169 –1173. 9. Rajan R, Levine M, Gent M, Geerts W, Skingley P, Julian J. The occurrence of subsequent malignancy in patients presenting with deep vein thrombosis: results from a historical cohort study. Thromb Haemost. 1998;79:19 –22. 10. Griffin MR, Stanson AW, Brown ML, et al. Deep venous thrombosis and pulmonary embolism. Arch Intern Med. 1987;147:1907–1911. 11. Nordstrom M, Lindblad B, Anderson H, Bergqvist D, Kjellstron T. Deep venous thrombosis and occult malignancy: an epidemiological study. BMJ. 1994;308:891– 894.
930
CANCER February 15, 2006 / Volume 106 / Number 4
12. Cornuz J, Pearson S, Creager M, Francis EF, Goldman L. Importance of findings on the initial evaluation for cancer in patients with symptomatic idiopathic deep venous thrombosis. Ann Intern Med. 1996;125:785–793. 13. Ries LAG, Kosary CL, Hankey BF, et al. SEER cancer statistics review, 1973–1994. NIH Pub. No. 97-2789. Bethesda, MD: National Cancer Institute, 1997. 14. Nattinger AB, McAuliffe TL, Schapira MM. Generalizability of the Surveillance, Epidemiology, and End Results registry population: factors relevant to epidemiologic and health care research. J Clin Epidemiol. 1997;50:939 –945. 15. Hatten J. Medicare’s common denominator: the covered population. Health Care Financ Rev. 1980;2:53– 64. 16. Morgan RO, Virnig BA, DeVito CE, Persily NA. The Medicare-HMO revolving door—the healthy go in and the sick go out. N Engl J Med. 1997;337:169 –175. 17. Levitan N, Dowlati A, Remic SC, et al. Rates of initial and recurrent thromboembolic disease among patients with malignancy versus those without malignancy: risk analysis using Medicare claims data. Medicine. 1999;78:285–291. 18. Kniffin WD, Baron J, Barret J, Birkmeyer JD, Anderson FA Jr. The epidemiology of diagnosed pulmonary embolism and deep vein thrombosis in the elderly. Arch Intern Med. 1994; 154:860 – 866. 19. Krieger N. Overcoming the absence of socioeconomic data in medical records: validation and application of a Censusbased methodology. Am J Public Health. 1992;92:703–710. 20. Hofer TP, Wolfe RA, Tedeshci PJ, McMahon LF, Griffith JR. Use of community versus individual socioeconomic data in predicting variation in hospital use. Health Serv Res. 1998; 33:243–259. 21. Metersky M, Tate J, Fine MJ, Petrillo MK, Meehan TP. Temporal trends in outcomes of older patients with pneumonia. Arch Intern Med. 2000;160:3385–3391. 22. Yuan, Z, Bowlin S, Einstadter D, Cebul RD, Conners AR Jr., Rimm AA. Atrial fibrillation as a risk factor for stroke: a retrospective cohort study of hospitalized Medicare beneficiaries. Am J Public Health. 1998;88:395– 400. 23. Houch PM, Bratzler DW, Nsa W, Ma A, Bartlett JG. Timing of antibiotic administration and outcomes for Medicare pa-
24.
25.
26.
27.
28.
29.
30. 31.
32.
33.
34. 35.
tients hospitalized with community-acquired pneumonia. Arch Intern Med. 2004;164:637– 644. Piccioli A, Lensing AWA, Prins MH, et al. Extensive screening for occult malignant disease in idiopathic venous thromboembolism: a prospective randomized clinical trial. J Thromb Haemost. 2004;2:884 – 889. Swan J, Breen N, Coates RJ, Rimer BK, Lee NC. Progress in cancer screening practices in the United States. Cancer. 2003;97:1528 –1540. Vijan S, Inadomi J, Hawyard RA, Hofer TP, Fendrick AM. Projections of demand and capacity for colonoscopy related to increasing rates of colorectal cancer screening in the United States. Aliment Pharmacol Ther. 2004;20:507–515. Leitch AM, Dodd GD, Costanza M, et al. American Cancer Society guidelines for the early detection of breast cancer: update 1997. CA Cancer J Clin. 1997;47:150 –153. Winawer S, Fletcher R, Rex D, et al. Colorectal cancer screening and surveillance: clinical guidelines and rationale— update based on new evidence. Gastroenterology. 2003;124:544 –560. Lauderdale DS, Goldberg J. The expanded racial and ethnic codes in the Medicare data files: their completeness of coverage and accuracy. Am J Public Health. 1996;86:712–716. Mitchell JB, Bubolz T, Paul JE, et al. Using Medicare claims for outcomes research. Med Care. 1994;32:JS38 –JS51. Fisher ES, Baron JA, Malenka DJ, Barrett J, Bubolz TA. Overcoming potential pitfalls in the use of Medicare data for epidemiologic research. Am J Public Health. 1990;80:1487– 1490. Lauderdale D, Furner SE, Miles TP, Goldberg J. Epidemiological uses of Medicare data. Epidemiol Rev. 1993;15:319 – 327. Geronimus AT, Bound J, Neidert LJ. On the validity of using Census geocode characteristics to proxy individual socioeconomic characteristics. J Am Stat Assoc. 1996;91:529 –537. Robinson WS. Ecological correlations and the behavior of individuals. Am Sociol Rev. 1950;15:351–357. Diez Roux AV, Merkin SS, Arnett D, et al. Neighborhood of residence and incidence of coronary heart disease. N Engl J Med. 2001;345:99 –106.
