Aspirin and the Risk of Hodgkin s Lymphoma in a Population-Based Case Control Study

Transcription

1 Aspirin and the Risk of Hodgkin s Lymphoma in a Population-Based Case Control Study Ellen T. Chang, Tongzhang Zheng, Edward G. Weir, Michael Borowitz, Risa B. Mann, Donna Spiegelman, Nancy E. Mueller Background: Regular use of nonsteroidal anti-inflammatory drugs (NSAIDs) is associated with decreased risk of several malignancies. NSAIDs may prevent cancer development by blocking the cyclooxygenase-catalyzed synthesis of proinflammatory prostaglandins. Aspirin may also protect against Hodgkin s lymphoma by inhibiting transcription factor nuclear factor B (NF- B), which is necessary for immune function and the survival of Hodgkin s lymphoma cells. We examined the association between regular analgesic use and the risk of Hodgkin s lymphoma. Methods: A populationbased case control study of 565 case patients with Hodgkin s lymphoma and 679 control subjects was conducted in the metropolitan area of Boston, Massachusetts, and in the state of Connecticut. Participants reported their average use of aspirin, non-aspirin NSAIDs, and acetaminophen over the previous 5 years. Regular analgesic use was defined as consumption of at least two tablets per week on average over the preceding 5 years; non-regular use was defined as consumption of fewer than two tablets per week. Results: The risk of Hodgkin s lymphoma associated with regular aspirin use was statistically significantly lower (odds ratio [OR] 0.60, 95 confidence interval [CI] 0.42 to 0.85) than that associated with non-regular aspirin use. The risk was not associated with use of other non-aspirin NSAIDs (OR 0.97, 95 CI 0.73 to 1.30). However, the risk associated with regular acetaminophen use was statistically significantly higher (OR 1.72, 95 CI 1.29 to 2.31) than that associated with non-regular use. Conclusion: The inverse association between aspirin, but not other NSAIDs, and Hodgkin s lymphoma suggests that NF- B signaling may play a key role in Hodgkin s lymphoma pathogenesis. [J Natl Cancer Inst 2004;96:305 15] Epidemiologic studies in human populations (1,2) have provided evidence that regular use of nonsteroidal anti-inflammatory drugs (NSAIDs), particularly aspirin, reduces cancer risk. The decreased risk has been most consistently detected in association with colorectal cancer (3 15), for which the risk of disease is reportedly lower for regular aspirin users than for non-users. An inverse association has been found, although less uniformly, with esophageal and gastric cancers and other malignancies (1,13,16 19). However, to our knowledge, no studies have examined the association between NSAID use and Affiliations of authors: Harvard School of Public Health, Boston, MA (ETC, DS, NEM); Yale University School of Medicine, New Haven, CT (TZ); The Johns Hopkins Medical Institute, Baltimore, MD (EGW, MB, RBM). Correspondence to: Ellen T. Chang, ScD, Harvard School of Public Health, Department of Epidemiology, 677 Huntington Ave., Boston, MA ( echang@hsph.harvard.edu). See Notes following References. DOI: /jnci/djh038 Journal of the National Cancer Institute, Vol. 96, No. 4, Oxford University Press 2004, all rights reserved. Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004 ARTICLES 305

2 Hodgkin s lymphoma, a cancer characterized by chronic immune dysfunction and inflammation. NSAIDs appear to act primarily by inhibiting cyclooxygenase 1 and 2, two isoforms of the enzyme prostaglandin endoperoxide synthase, and thus reducing prostaglandin synthesis (1,2,20). NSAIDs bind to the cyclooxygenase active site, disabling the enzyme (21,22). As potent mediators of inflammation and potential cofactors in carcinogenesis, prostaglandins stimulate tumor growth and dramatically increase DNA and RNA synthesis in cancer cells (23 26). Elevated prostaglandin levels are detected in various tumors, including Hodgkin s lymphoma (1,27 29). Cyclooxygenase expression is also increased in the neoplastic Hodgkin s and Reed Sternberg cells and some reactive cells in lymph nodes from Hodgkin s lymphoma patients (30). Certain NSAIDs may also act through other cyclooxygenaseindependent pathways (31 38). In particular, aspirin and its metabolite sodium salicylate inhibit the transcription factor nuclear factor B (NF- B), which regulates the expression of multiple cellular and viral genes (39,40). Most of the genes activated by NF- B are involved in immune and inflammatory responses, including cytokines and cell adhesion molecules, or regulate cell growth and apoptosis (39 44). Activated NF- B has been detected almost universally in both primary Hodgkin s and Reed Sternberg cells and cell lines, and inhibition of active NF- B decreases proliferation and causes spontaneous apoptosis of Hodgkin s and Reed Sternberg cells (45 47). Thus, NF- B appears to have a vital role in Hodgkin s and Reed Sternberg cell survival and apoptosis resistance. An etiologic link between the Epstein Barr virus (EBV) and Hodgkin s lymphoma pathogenesis is supported by strong epidemiologic and molecular biologic evidence, including detection of monoclonal viral DNA and gene products in the Hodgkin s and Reed Sternberg cells in of patients with Hodgkin s lymphoma (48 50). EBV is a ubiquitous herpesvirus that infects more than 90 of the adult population worldwide, most often during childhood when it typically causes no serious clinical disease (51); however, EBV is also associated with various serious diseases and malignancies, including Hodgkin s lymphoma (49,52). Although EBV activates and exploits the NF- B pathway to replicate and to evade the human immune system (47,53), the constitutive expression of active NF- B in Hodgkin s and Reed Sternberg cells is independent of EBV status (45,54 56). Given the important role of NF- B in Hodgkin s lymphoma, aspirin may have a unique protective effect against the development of Hodgkin s lymphoma. Consequently, we conducted a population-based case control study to investigate whether the risk of Hodgkin s lymphoma was associated with the reported use of analgesics. PATIENTS AND METHODS Study Population We initiated a population-based case control study in 1997 designed to investigate the viral etiology of Hodgkin s lymphoma in the greater Boston, Massachusetts, metropolitan area and in the state of Connecticut. Eligible case patients were individuals between ages 15 and 79 years at diagnosis who were living in the described geographic area and were without evidence of human immunodeficiency virus infection. Between August 1, 1997, and December 31, 2000, 743 eligible case patients with Hodgkin s lymphoma were diagnosed within the study base and identified by participating hospitals and cancer registries. Permission for contacting these patients was sought from the treating physician and was granted for 685 (92) of 743 patients. After physician consent, each patient was sent an initial explanatory letter inviting him or her to participate in the study. Of the eligible patients who were contacted for an interview by study staff, 335 (84) of 398 case patients in Massachusetts and 240 (86) of 280 case patients in Connecticut participated. Ten patients were excluded by study pathologists as not having Hodgkin s lymphoma, leaving 565 case patients with Hodgkin s lymphoma. A total of 679 population control subjects were selected, enrolled from the study base, and frequency matched to the distribution of the case patients by age (within a 5-year interval), sex, and state of residence. Control subjects were defined as residents living in the study area without a prior history of Hodgkin s lymphoma. From the Boston region, population control subjects were randomly identified from the current Town Books for the 132 cities and towns within the study area. The Town Books are annually compiled records including the name, sex, street address, and year of birth of all town or city residents aged 17 years or older. These records are more than 90 complete (57). Within our study, 135 (93) of a random subset of 145 case patients were listed in the Town Books. When a selected control subject refused to participate or could not be reached despite telephone and address searches for individuals who had moved from the addresses listed in the Town Books, the next eligible person listed was selected as a replacement control subject. As with the case patients, all control subjects were initially contacted via an introductory letter. Of the 835 Massachusetts residents who were selected as potential control subjects, 115 (14) could not be contacted. Of the 720 who were contacted, 238 (33) did not respond to multiple telephone calls; 31 (4) did not respond to attempts at contact by mail; four cited a language barrier; two were incapacitated; and one was deceased. Seventy-seven residents (11) refused to participate, and 367 (51) consented. Among the enrolled control subjects from Massachusetts, 178 (49) were initial selections from the Town Books, and the other 189 were replacements for refusing or noncontacted selections. In Connecticut, where 98.9 of residents have home telephone service (58), control subjects between ages 18 and 65 years were identified by random digit dialing (59). All case patients from Connecticut aged 65 years or younger also had home telephone numbers. Control subjects between ages 66 and 79 years were randomly selected from files provided by the Health Care Financing Administration. All case patients from Connecticut between ages 66 and 79 years were also listed in the Health Care Financing Administration files. Only one control subject was recruited per household to avoid overlap of subject responses and clustering by social class. To prevent geographic clustering, the maximum number of households screened by random digit dialing within a block of 100 telephone numbers was limited to eight. Among 5632 phone numbers attempted by random digit dialing, 4747 were successfully screened; of these, 450 (9) corresponded to eligible Connecticut residents of the targeted age range and sex. Twelve persons could not be reached by telephone. Of the remaining 438 individuals, 21 (5) did not respond to multiple telephone calls, four (1) were incapacitated, 128 (29) refused, and 276 (63) completed the survey. 306 ARTICLES Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004

3 Among the 65 contacted individuals of 69 eligible Health Care Financing Administration members, eight (12) did not respond to telephone calls, two (3) were incapacitated, 19 (29) refused, and 36 (55) consented. All study participants or, if they were minors, their guardians gave written informed consent for enrollment in the study. This research protocol was approved by the institutional review boards of the Harvard School of Public Health, the Yale University School of Medicine, and the Johns Hopkins Medical School, as well as all participating hospitals, the Massachusetts Cancer Registry, and the Connecticut Tumor Registry located in the Connecticut Department of Public Health. Data Collection The introductory letter was followed by a telephone request to schedule a structured telephone interview. Next-of-kin interviews were performed for two case patients and two control subjects. For two case patients and 29 control subjects who could not be reached by telephone, an abbreviated questionnaire was mailed; in rare instances, the questionnaire was administered in person. All participants were asked to report their average aspirin, ibuprofen and/or other non-aspirin NSAID, and acetaminophen use separately over the previous 5 years. Exposure was calculated in terms of the number of days per month, the number of tablets taken per day, and the medication strength. Reasons for use were not recorded. For statistical analyses, use of aspirin, other NSAIDs, and acetaminophen was expressed as the average number of tablets taken per week. This value was calculated by multiplying average monthly use by both average strength and average number of tablets per use and then dividing the product by 4.3 (the approximate number of weeks per month). If subjects reported variable use over time, a time-weighted mean was calculated to yield the average number of tablets taken per week over the previous 5 years. Baby-strength, extra-strength, and prescription-strength medications were considered to be multiples of regular-strength tablets. For example, extra-strength aspirin tablets (500 mg) were considered to be the equivalent of 1.54 regular-strength aspirin tablets (325 mg), and baby-strength aspirin tablets (80 mg) were considered to be the equivalent of 0.25 regular-strength aspirin tablets. If subjects reported a frequency of nonprescription medication use but were unsure of the tablet strength, regular strength was assumed. Use of each medication was coded as a binary variable, with regular users defined as those taking two or more tablets per week and non-users defined as those taking fewer than two tablets per week. This coding scheme is in accordance with previous observational studies of NSAID use in human populations (7,8). Case information, including stage of disease and presence of B symptoms at diagnosis, was abstracted from patients medical records by study staff and medical personnel. Data on median household income in census tract and percentage of census tract below poverty level were obtained for each participant based on his or her residential street address (60). Histopathology The study pathologists (M. Borowitz, R. B. Mann, and E. G. Weir) reviewed all available pathology material to verify the diagnosis of Hodgkin s lymphoma. Archived tissue blocks were retrieved and examined for 411 (71) of 575 case patients. For those case patients without adequate, available, or consented diagnostic specimens, the original pathology reports were reviewed to determine whether a diagnosis of Hodgkin s lymphoma was probable, possible, unlikely, or unable to be confirmed. Pathology report based diagnoses were made for 156 case patients (27); pathology material was unavailable for eight case patients (1). Histopathologic classification of tumors was based on the 1994 Revised European-American Classification of Lymphoid Neoplasms (REAL Classification) and the 2001 World Health Organization (WHO) Classification of Hematopoietic and Lymphoid Tumors (61,62). These pathologic classification schemes distinguish among several subtypes that are considered broadly to be classical Hodgkin s lymphoma: nodular sclerosis, mixed cellularity, lymphocyte depletion, and lymphocyte-rich classic Hodgkin s lymphoma. Nodular lymphocyte predominance Hodgkin s lymphoma was considered to be a distinct entity and was grouped separately from the classical subtypes. Study pathologists also confirmed Hodgkin s lymphoma case patients who could not be further subclassified, including patients for whom there was an insufficient amount of diagnostic tissue for subtyping, as well as those in whom nodular sclerosis could not be differentiated from mixed cellularity due to early involvement of a lymph node in an interfollicular pattern. The presence of EBV in Hodgkin s lymphoma tissue was determined by in situ hybridization for EBV-encoded RNA transcripts and/or by immunohistochemical assay for the viral latency membrane protein 1 in the malignant Hodgkin s and Reed Sternberg cells (63,64). A Hodgkin s lymphoma tumor was considered to be positive for the EBV genome if results were positive for either assay; a tumor was considered to be negative for the EBV genome if both assays were negative or only a single assay was performed and its result was negative (65). EBV assays were interpreted by consensus of all three study pathologists. Statistical Methods All exposures with more than two levels were analyzed first as categorical variables and then as ordinal variables. Likelihood ratio tests were performed to determine whether these exposures should be entered into statistical models as single ordinal terms. From the results of these tests, monotonic trends for association with the log odds of Hodgkin s lymphoma were observed for educational level (subject, maternal, and paternal), number of siblings, number of people in childhood home, number of regular childhood playmates, census tract median household income, and percentage of census tract population below poverty level. All of these exposures were coded as ordinal variables in statistical models. Associations (odds ratios [ORs], 95 confidence intervals [CIs], and corresponding P values from two-sided chi-square tests) were estimated with multiple logistic regression to control for age, sex, and state of residence. Age was controlled for as a categorical variable with three levels (15 39 years, years, and years) from previous findings that Hodgkin s lymphoma risk is associated nonlinearly with age, with incidence peaks in the younger and older groups. Potential confounders were considered from prior knowledge, as well as change-in- Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004 ARTICLES 307

4 estimate criteria (66) and likelihood ratio tests comparing models with and without additional variables. To detect possible effect modification, the data were stratified by additional factors to examine differences in associations across strata. The presence of effect modification was assessed with likelihood ratio and Wald tests for the statistical significance of interaction terms. Distributions of demographic characteristics and main exposures among case patients with pathologically unconfirmed disease were compared with those among case patients with confirmed disease. Among case patients from Massachusetts, distributions for those who were not listed in the Town Books were compared with distributions for those who were listed. After ensuring that there were no statistically significant differences between the compared groups, all eligible case patients were combined in the analyses. All statistical analyses were performed with SAS System software, release 8.01 ( ; SAS Institute, Cary, NC). All statistical tests were two-sided. Sensitivity Analysis for Selection Bias To evaluate the magnitude of bias from an overselection of control subjects with high socioeconomic status, we used the method of Kleinbaum et al. (67) to calculate the control subject selection odds ratio. The prevalences of high and low median household income in subjects residential census tracts, as indicators of socioeconomic status, were compared between enrolled Massachusetts control subjects and all Massachusetts residents initially selected from the Town Books. After adjusting for the degree of misclassification between median household income and regular analgesic use, which was our main exposure of interest, we estimated the percent difference between the true and observed odds ratios for the associations between Hodgkin s lymphoma risk and regular compared with non-regular use of aspirin, other NSAIDs, and acetaminophen. RESULTS The initial study population consisted of 565 Hodgkin s lymphoma case patients and 679 control subjects frequencymatched to the case patients by age, sex, and state of residence. The distribution of demographic characteristics among case patients and control subjects is shown in Table 1. Table 2 details the prevalence of regular aspirin, non-aspirin NSAID, and acetaminophen use among Hodgkin s lymphoma case patients and control subjects. Younger case patients and control subjects were statistically significantly less likely than older case patients and control subjects to use aspirin, and younger case patients were more likely than older case patients to use non-aspirin NSAIDs regularly. Neither case patients nor control subjects differed by age group in frequency of regular acetaminophen use. Male control subjects were more likely than female control subjects to use aspirin, whereas males in both the case patient and control subject groups were less likely than females in these groups to use other NSAIDs and acetaminophen. Use of particular analgesics also differed statistically significantly by state, race, religion, census tract median household income, smoking history, and use of other medications (Table 2). Because of the associations of age, sex, state of residence, smoking history, and use of other analgesics with analgesic use in the control population, the Table 1. Distribution of demographic characteristics among Hodgkin s lymphoma case patients and control subjects Characteristic Case patients (n 565) Control subjects (n 679) Age, y y y y y y 9 11 Median age, y Sex, Male Female State of residence, Massachusetts Connecticut Race, White Black 4 5 Hispanic 5 5 Other 3 4 Religion, Catholic Protestant Jewish 7 4 Other/none Subject s educational level, Less than high school 10 7 High school College Advanced degree Maternal educational level, Less than high school High school College Advanced degree 7 9 Median household income strata, * $ $ $70, $ $85, $ Median household income, $ $ $ Below poverty level strata, * Median below poverty level, *U.S. Census data were based on participants census tract of residence. Quartiles were based on the distribution of median household income and percent below poverty level among control subjects. independent associations of these factors with Hodgkin s lymphoma risk, and the frequency matching in our study design, we controlled for these factors in all analyses. Past studies (68,69) have shown that risk factors for Hodgkin s lymphoma differ between younger and older populations; therefore, we initially analyzed data separately for subjects aged years and those aged 55 years or older. However, the associations between Hodgkin s lymphoma risk and aspirin, other NSAID, and acetaminophen use did not differ by age group (P for heterogeneity by age.26,.50, and.64, respectively); consequently, subjects of all ages were combined for further analyses. The overall adjusted odds ratio for 308 ARTICLES Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004

5 Table 2. Prevalence of regular analgesic use (two or more times per week) within subgroups of case patients with Hodgkin s lymphoma and control subjects and two-sided chi-square tests for differences in prevalence across subgroups of participants Characteristic case patients P Regular aspirin use Regular other NSAID* use Regular acetaminophen use control subjects P case patients P control subjects P case patients P control subjects P Overall Age y y y Sex Male Female State of residence Massachusetts Connecticut Race/ethnicity White Black Other Religion Catholic Protestant Jewish Other/none Subject s educational level Less than high school High school College Advanced degree Median income $ $ $ $ $ $ Cigarette smoking history Yes No Regular aspirin use Yes No 0.0 Undefined 0.0 Undefined Regular other NSAID use Yes No Undefined 0.0 Undefined Regular acetaminophen use Yes No Undefined 0.0 Undefined *NSAID nonsteroidal anti-inflammatory drug. P value from two-sided chi-square test for difference in prevalence of regular analgesic use across subgroups of case patients or control subjects U.S. Census data were based on participants census tract of residence. Quartiles were based on the distribution of median household income among control subjects. Hodgkin s lymphoma risk associated with regular versus nonregular aspirin use was statistically significantly decreased (OR 0.60, 95 CI 0.42 to 0.85) (Table 3). This association was not substantially affected by adjustment for other potential confounders, including demographic characteristics and socioeconomic status. Hodgkin s lymphoma risk was not statistically significantly associated with regular use of non-aspirin NSAIDs (OR 0.97, 95 CI 0.73 to 1.30), compared with nonregular use of non-aspirin NSAIDs. However, an elevated risk was statistically significantly associated with regular acetaminophen use (OR 1.72, 95 CI 1.29 to 2.31), compared with non-regular acetaminophen use. Again, adjustment for other potential confounders did not alter the associations between Hodgkin s lymphoma risk and use of non-aspirin NSAIDs or acetaminophen. We assessed the possibility of a dose response relationship by dividing individuals into three groups instead of two: those who never used analgesics (never users), those who used analgesics fewer than two times per week on average (occasional users), and those who used analgesics two or more times per week on average (regular users). Occasional aspirin use was associated with a statistically significantly decreased risk of Hodgkin s lymphoma (OR 0.71, 95 CI 0.51 to 0.99) compared with never use; regular aspirin use had an even lower risk (OR 0.58, 95 CI 0.41 to 0.83) Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004 ARTICLES 309

6 Table 3. Stratified, adjusted odds ratios (ORs) and 95 confidence intervals (CIs) for association between Hodgkin s lymphoma risk and regular versus nonregular analgesic use (two or more times per week versus fewer times per week) and P values from two-sided tests for heterogeneity across subgroups of participants Subgroup Regular vs. non-regular aspirin use Regular vs. non-regular acetaminophen use OR* (95 CI) P OR* (95 CI) P Overall 0.60 (0.42 to 0.85) 1.72 (1.29 to 2.31) Age y 0.67 (0.44 to 1.01) 1.75 (1.27 to 2.41) y 0.46 (0.24 to 0.88) (0.74 to 3.21).64 Sex Male 0.58 (0.36 to 0.91) 1.74 (1.13 to 2.68) Female 0.65 (0.38 to 1.12) (1.14 to 2.53).87 State of residence Massachusetts 0.54 (0.33 to 0.91) 2.33 (1.54 to 3.55) Connecticut 0.62 (0.38 to 1.00) (0.85 to 1.96).04 Race White 0.59 (0.41 to 0.85) 1.59 (1.16 to 2.18) Black 2.67 (0.37 to 19.25) 0.94 (0.23 to 3.75) Other 0.23 (0.02 to 2.17) (2.15 to 30.91).17 Religion Catholic 0.71 (0.45 to 1.12) 1.44 (0.98 to 2.10) Protestant 0.30 (0.13 to 0.70) 2.91 (1.59 to 5.32) Jewish 0.30 (0.04 to 2.21) 0.26 (0.06 to 1.18) Other/none 0.85 (0.32 to 2.29) (1.01 to 8.07).05 Educational level Subject s Less than high school 0.13 (0.03 to 0.68) 2.09 (0.74 to 5.93) High school 0.60 (0.31 to 1.16) 1.92 (1.10 to 3.36) College 0.69 (0.42 to 1.14) 1.45 (0.94 to 2.23) Advanced degree 0.43 (0.16 to 1.20) (0.73 to 4.23).53 Mother s Less than high school 0.30 (0.13 to 0.70) 3.03 (1.44 to 6.34) High school 0.72 (0.42 to 1.24) 1.71 (1.09 to 2.69) College 0.91 (0.46 to 1.79) 1.42 (0.78 to 2.58) Advanced degree 0.24 (0.03 to 2.19) (0.17 to 2.11).05 Median household income $ (0.48 to 2.11) 2.11 (1.20 to 3.70) $ $ (0.31 to 1.50) 1.86 (1.00 to 3.47) $ $ (0.16 to 0.61) 2.01 (1.10 to 3.66) $ (0.33 to 1.36) (0.58 to 2.00).10 Cigarette smoking Yes 0.60 (0.38 to 0.93) 1.73 (1.17 to 2.57) No 0.62 (0.35 to 1.08) (1.09 to 2.62).98 Regular analgesic use Aspirin Yes Undefined 1.49 (0.76 to 2.92) No Undefined Undefined 1.79 (1.29 to 2.48) 1.00 Non-aspirin NSAIDs Yes 0.72 (0.36 to 1.45) 2.85 (1.62 to 5.03) No 0.57 (0.38 to 0.85) (1.03 to 2.06).05 Acetaminophen Yes 0.59 (0.32 to 1.09) Undefined No 0.62 (0.41 to 0.95) 1.00 Undefined Undefined Histologic subtype Nodular sclerosis 0.49 (0.32 to 0.78) 1.63 (1.16 to 2.29) Mixed cellularity 0.94 (0.47 to 1.91) (0.72 to 2.80).72 Tumor EBV genome status Positive 0.42 (0.21 to 0.84) 1.81 (1.06 to 3.08) Negative 0.64 (0.42 to 0.99) (1.11 to 2.24).68 *Odds ratio for regular versus non-regular use of analgesics among Hodgkin s lymphoma case patients versus control subjects, adjusted for age, sex, state of residence, smoking history, and use of other analgesics. P value from two-sided likelihood ratio test for heterogeneity of odds ratio across subgroups of participants. U.S. Census data were based on participants census tract of residence. Quartiles were based on distribution of median household income among control subjects. P value from two-sided Wald test for heterogeneity of odds ratio across subgroups of participants. compared with never use (P trend.001). Neither occasional nor regular ibuprofen use was associated with the risk of Hodgkin s lymphoma (for occasional versus never users, OR 0.72, 95 CI 0.53 to 0.97; for regular versus never users, OR 0.87, 95 CI 0.63 to 1.19) (P trend.24). Occasional acetaminophen use, compared with never use, was statistically significantly associated with a higher risk of Hodgkin s lymphoma (OR 1.88, 95 CI 1.40 to 2.53), 310 ARTICLES Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004

7 and regular use, compared with never use, was associated with a still higher risk (OR 2.17, 95 CI 1.58 to 2.98) (P trend.001). For clarity and because the comparisons of regular use with never use, as opposed to non-regular use, did not differ appreciably, we performed remaining analyses with the dichotomous definition of regular analgesic use (two or more tablets per week versus fewer than two tablets per week). To detect possible heterogeneity of the associations between analgesic use and the risk of Hodgkin s lymphoma, we performed stratified analyses among subject subgroups (Table 3). The inverse association between aspirin use and the risk of Hodgkin s lymphoma did not vary by any characteristic except income: the apparent decrease in risk associated with regular use, compared with non-regular use, was more marked among subjects living in middle-income census tracts than among those living in lower income and higher income areas (P for heterogeneity by census tract median household income.05). However, the association with aspirin was inverse for all but the lowest quartile of income. The statistically nonsignificant association between regular use of other NSAIDs, compared with non-regular use of other NSAIDs, and Hodgkin s lymphoma risk was consistent across levels of all subgroups examined (data not shown). The association with regular acetaminophen use, compared with nonregular acetaminophen use, varied by state (P.04), with a statistically significantly positive association among Massachusetts but not among Connecticut residents. In addition, the association between acetaminophen use and Hodgkin s lymphoma risk varied by religion (P.05), with an inverse association among Jewish individuals and a positive association among those of other religions. The association between acetaminophen use and Hodgkin s lymphoma risk also varied by maternal educational level (P.05), with a decreasing association across increasing educational levels. This association also varied by use of non-aspirin NSAIDs (P.05), with a stronger positive association with regular use than with non-regular use of other NSAIDs. A fully saturated statistical model, including all two-way interactions between main effects, did not fit the data better than the reduced model with only main effects (P.51). We therefore constructed a model, by use of stepwise selection, that included only statistically significant two-way interactions: acetaminophen use by state, religion, and maternal educational level. Additional interactions were statistically nonsignificant after including these three interactions. In this model, the associations between Hodgkin s lymphoma risk and regular versus nonregular aspirin use (OR 0.65, 95 CI 0.45 to 0.93) and regular versus non-regular use of other NSAIDs (OR 1.00, 95 CI 0.94 to 1.17) were not statistically significantly different from the estimates from the reduced model. Stratified odds ratios for acetaminophen use across state, religion, and maternal educational levels are provided in Table 3. To determine whether the relationships between regular analgesic use and Hodgkin s lymphoma differed by whether tumor cells contained the EBV genome (EBV genome status), we stratified case patients with Hodgkin s lymphoma by EBV genome status. Overall, 96 (24) of 407 case patients with known EBV status were positive for the EBV genome (EBV-positive). Tumor EBV status did not modify the associations between Hodgkin s lymphoma risk and regular versus non-regular use of aspirin (P.31), of non-aspirin NSAIDs (P.11), or of acetaminophen (P.68). The homogeneity of odds ratios between EBV-positive and EBV-negative case patients, compared with control subjects, suggests that the associations between analgesic use and Hodgkin s lymphoma risk were independent of EBV genome status in the tumor. To address the concern that analgesic use may have been affected by Hodgkin s lymphoma diagnosis or the onset of symptoms, we stratified case patients by presence of B symptoms (fever, night sweats, and weight loss) and by time from diagnosis to interview. Among asymptomatic patients, the inverse association between Hodgkin s lymphoma risk and regular versus non-regular aspirin use (OR 0.54, 95 CI 0.33 to 0.87), the lack of association with regular versus non-regular non-aspirin NSAID use (OR 0.87, 95 CI 0.60 to 1.28), and the positive association with regular versus non-regular acetaminophen use (OR 1.69, 95 CI 1.16 to 2.46) did not differ from the associations among patients who exhibited B symptoms (P.68 for aspirin, P.37 for other NSAIDs, P.65 for acetaminophen). Likewise, there was also no difference in the prevalence of rheumatoid arthritis, a condition for which aspirin and other NSAIDs are often prescribed, between case patients (7) and control subjects (6). The median time between Hodgkin s lymphoma diagnosis and interview was 7.2 months, with a range of months. There was no difference in the associations between Hodgkin s lymphoma risk and regular versus non-regular aspirin use (P.54), regular versus non-regular non-aspirin NSAID use (P.22), or regular versus non-regular acetaminophen use (P.97), comparing the odds ratio for case patients who were diagnosed within 7.2 months of interview with that for case patients diagnosed longer before interview. We performed a series of analyses to evaluate whether there was any evidence of selection bias in our method of control subject selection. Among the Massachusetts residents initially selected as potential control subjects, those who lived in census tracts with higher median household income were more likely to participate than those from lower income areas (for higher versus lower income, comparing consenting with nonconsenting initially identified control subjects, OR 1.26, 95 CI 1.03 to 1.54), adjusting for age and sex. However, non-enrolled individuals were replaced with control subjects from the same residential area so that participants were eventually enrolled from all originally selected census tracts. Although a higher number of consecutive refusals within a residential area was associated with both lower income and higher poverty, the income distribution of the final Massachusetts control population was representative of that of the source population (for higher versus lower income, comparing finally consenting with all initially identified control subjects, OR 1.03, 95 CI 0.91 to 1.18). A sensitivity analysis of the potential effects of selection bias revealed that differences in socioeconomic status between the control subjects and the source population would have resulted in a 14, 7, and 12 bias of the odds ratios for regular versus non-regular use of aspirin, regular versus non-regular use of other NSAIDs, and regular versus non-regular use of acetaminophen, respectively, if there was no selection bias among the case patients and no other sources of bias or confounding (67). We compared the distribution of several demographic characteristics within our Connecticut control group to the current Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004 ARTICLES 311

8 distribution across the state of Connecticut (58). Study control subjects were generally more likely to have attended college than high school only (OR 1.35, 95 CI 1.07 to 1.70) and were more likely to live in single-family rather than multifamily housing units (OR 1.40, 95 CI 1.10 to 1.80); however, educational level, number of people in childhood home, and other socioeconomic status indicators were not confounders of the association between analgesic use and Hodgkin s lymphoma risk and thus would not have affected our results. DISCUSSION In this population-based case control study, regular aspirin use was associated with a 40 reduced risk of Hodgkin s lymphoma, compared with non-regular aspirin use. This association was consistent across subgroups of age, sex, race, religion, smoking history, and analgesic use. Unlike aspirin, use of other NSAIDs such as ibuprofen was not statistically significantly associated with Hodgkin s lymphoma risk. Regular acetaminophen use, compared with non-regular use, however, was associated with a consistently higher risk of Hodgkin s lymphoma. A dose response analysis showed that the associations of both aspirin and acetaminophen use with the risk of Hodgkin s lymphoma were stronger with regular use than with occasional use. That is, there was a statistically significant trend for decreasing Hodgkin s lymphoma risk with increasing frequency of aspirin use and a trend in the opposite direction for acetaminophen use. The association with Hodgkin s lymphoma risk did not vary among regular, occasional, and never use of non-aspirin NSAIDs. The statistically significant reduced risk associated with regular versus non-regular use of aspirin, but not other NSAIDs, may indicate that properties exclusive to aspirin are responsible for its relationship with Hodgkin s lymphoma. Extensive laboratory studies have demonstrated that aspirin effectively inhibits NF- B activation at physiologic concentrations measured in the serum of patients treated with aspirin for chronic inflammatory diseases. In contrast, neither acetaminophen (a poor cyclooxygenase inhibitor and a poor NSAID) nor indomethacin (an active cyclooxygenase inhibitor and an effective NSAID) has a substantial effect on NF- B activity (70 72). Aspirin is unique among common NSAIDs because it forms an irreversible covalent bond with cyclooxygenase, whereas other NSAIDs bind reversibly to the protein (73). The extent of enzymatic inhibition also differs among NSAIDs (74 76). We detected some modification of the association between analgesic use and Hodgkin s lymphoma risk by several factors. The variation in the association of analgesic use and the risk of Hodgkin s lymphoma by socioeconomic factors, including income and maternal educational level, could be the result of selection bias by socioeconomic status, misclassification, or chance. However, there could also be a real biologic interaction between the effects of socioeconomic environment and analgesic use on the immune system because poverty is associated with declining nutritional status and host immune competency (77,78). This variation is in accord with the findings that the inverse association of aspirin use and the positive association of acetaminophen use with risk of Hodgkin s lymphoma were apparently less pronounced in lower socioeconomic status strata. However, neither association was modified by subject educational level. The heterogeneity of the odds ratio for regular versus non-regular acetaminophen use by state of residence and religion is difficult to explain but may result from differences in environment and lifestyle or from chance or misclassification. The lack of similar interactions with use of aspirin and other NSAIDs makes chance a more likely explanation. Our results should be interpreted with caution for the following reasons. In the study interview, questions to evaluate analgesic use were limited and addressed use only over the last 5 years. Therefore, the effect of duration of analgesic use could not be assessed. Most subjects reported that they did not use any analgesics regularly, leading to a limited range of medication use and insufficient power for detailed dose response analyses. In addition, indications for medication use were not recorded, so it is possible that for some case patients the reported use of aspirin, other NSAIDs, or acetaminophen was for analgesic and/or antipyrrhetic purposes during treatment for Hodgkin s lymphoma. However, this problem would have biased the odds ratio above 1.0, because protopathic bias that is, medication use for the purpose of alleviating disease- and treatment-related symptoms would have produced a positive association between case patient status and analgesic use (79). The possibility of protopathic bias is also countered by the fact that the associations between Hodgkin s lymphoma risk and regular analgesic use did not vary according to the presence of B symptoms among case patients or by the period between diagnosis and interview. The association between aspirin and Hodgkin s lymphoma risk was the same among regular and non-regular users of acetaminophen or other NSAIDs, suggesting that case patients with Hodgkin s lymphoma did not use these other analgesics in place of aspirin. The finding of a statistically significantly elevated odds ratio between Hodgkin s lymphoma risk and the regular versus nonregular use of acetaminophen does raise some concerns, however. It is consistent with the possibility that case patients reported increased analgesic use after the onset of diseaserelated symptoms. It is, however, also possible that a true association exists between acetaminophen use and Hodgkin s lymphoma. Several, but not all, epidemiologic studies (80 85) have reported an increased risk of renal and bladder cancer after regular acetaminophen use. Positive associations with acetaminophen use appear to be limited to genitourinary tract cancers, and an inverse association with ovarian cancer may exist (86,87). The increased risk of Hodgkin s lymphoma associated with regular acetaminophen use observed in this study is, therefore, more likely explained by uncontrolled confounding by other factors than by a true association. The uncontrolled confounding may or may not have affected the observed associations between Hodgkin s lymphoma risk and the use of aspirin or other NSAIDs. Another possible source of bias in our data is selection bias, because individuals who participate as control subjects are often of higher educational level and socioeconomic status than those who do not participate. Indeed, an analysis of U.S. Census data revealed that persons from higher income census tracts were more likely than persons from lower income census tracts to participate in our study. Our study protocol specified that all non-enrolled persons were replaced with control subjects selected from the same residential area. However, although our final control population appeared to be representative of the source population in terms of socioeconomic status, selection bias within census tracts may still have existed, as well as selection bias by other socioeconomic status determinants and unmeasurable factors, such as health consciousness. The incom- 312 ARTICLES Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004

9 plete response level among case patients may also have contributed to selection bias. Still, our sensitivity analysis using censusbased income data (data not presented) suggested that selection bias did not explain the observed associations between Hodgkin s lymphoma risk and analgesic use. Our findings require confirmation in other study populations, ideally with prospective data to ensure that medication use precedes diagnosis, as well as with more detailed data on dose and duration of analgesic use. Validation of our findings will require either many years of follow-up in prospective cohorts to identify enough case patients with Hodgkin s lymphoma for adequate statistical power or analyses in other case control studies of Hodgkin s lymphoma that have data on analgesic use. Despite these caveats, the finding that regular aspirin use is associated with a decreased risk of Hodgkin s lymphoma is intriguing. Given the extensive inflammation and immune dysregulation accompanying Hodgkin s lymphoma, an attractive hypothesis is that anti-inflammatory drugs guard against the onset of disease. This hypothesis is supported by the fact that aspirin specifically inactivates NF- B, which is required for the survival of malignant Hodgkin s lymphoma cells, and also triggers myriad signaling cascades in the immune system. Our finding that aspirin use is associated with a reduced risk of both EBV-positive and EBV-negative Hodgkin s lymphoma is in accord with the observation that NF- B is constitutively active in Hodgkin s lymphoma regardless of tumor EBV status. If aspirin use is indeed found to protect against Hodgkin s lymphoma, this relationship could afford insight into the pathogenesis of the disease and offer possible clues toward its prevention. REFERENCES (1) IARC Working Group on the Evaluation of Cancer Preventive Agents. Non-steroidal anti-inflammatory drugs. IARC Handbooks of Cancer Prevention. Vol. 1. New York (NY): Oxford University Press; p (2) Thun MJ, Henley SJ, Patrono C. Nonsteroidal anti-inflammatory drugs as anticancer agents: mechanistic, pharmacologic, and clinical issues. J Natl Cancer Inst 2002;94: (3) Kune GA, Kune S, Watson LF. Colorectal cancer risk, chronic illnesses, operations, and medications: case control results from the Melbourne Colorectal Cancer Study. Cancer Res 1988;48: (4) Isomaki HA, Hakulinen T, Joutsenlahti U. Excess risk of lymphomas, leukemia and myeloma in patients with rheumatoid arthritis. J Chronic Dis 1978;31: (5) Gann PH, Manson JE, Glynn RJ, Buring JE, Hennekens CH. Low-dose aspirin and incidence of colorectal tumors in a randomized trial. J Natl Cancer Inst 1993;85: (6) Garcia-Rodríguez LA, Huerta-Alvarez C. Reduced risk of colorectal cancer among long-term users of aspirin and nonaspirin nonsteroidal antiinflammatory drugs. Epidemiology 2001;12: (7) Giovannucci E, Rimm EB, Stampfer MJ, Colditz GA, Ascherio A, Willett WC. Aspirin use and the risk for colorectal cancer and adenoma in male health professionals. Ann Intern Med 1994;121: (8) Giovannucci E, Egan KM, Hunter DJ, Stampfer MJ, Colditz GA, Willett WC, et al. Aspirin and the risk of colorectal cancer in women. New Engl J Med 1995;333: (9) Giovannucci E. The prevention of colorectal cancer by aspirin use. Biomed Pharmacother 1999;53: (10) Greenberg ER, Baron JA, Freeman DH Jr, Mandel JS, Haile R. Reduced risk of large-bowel adenomas among aspirin users. The Polyp Prevention Study Group. J Natl Cancer Inst 1993;85: (11) Gridley G, McLaughlin JK, Ekbom A, Klareskog L, Adami HO, Hacker DG, et al. Incidence of cancer among patients with rheumatoid arthritis. J Natl Cancer Inst 1993;85: (12) Suh O, Mettlin C, Petrelli NJ. Aspirin use, cancer, and polyps of the large bowel. Cancer 1993;72: (13) Thun MJ, Namboodiri MM, Calle EE, Flanders WD, Heath CW Jr. Aspirin use and risk of fatal cancer. Cancer Res 1993;53: (14) Sandler RS, Halabi S, Baron JA, Budinger S, Paskett E, Keresztes R, et al. A randomized trial of aspirin to prevent colorectal adenomas in patients with previous colorectal cancer [published erratum appears in N Engl J Med 2003;348:1939]. N Engl J Med 2003;348: (15) Baron JA, Cole BF, Sandler RS, Haile RW, Ahnen D, Bresalier R, et al. A randomized trial of aspirin to prevent colorectal adenomas. N Engl J Med 2003;348: (16) Schreinemachers DM, Everson RB. Aspirin use and lung, colon, and breast cancer incidence in a prospective study. Epidemiology 1994;5: (17) Sharp L, Chilvers CE, Cheng KK, McKinney PA, Logan RF, Cook- Mozaffari P, et al. Risk factors for squamous cell carcinoma of the oesophagus in women: a case-control study. Br J Cancer 2001;85: (18) Farrow DC, Vaughan TL, Hansten PD, Stanford JL, Risch HA, Gammon MD, et al. Use of aspirin and other nonsteroidal anti-inflammatory drugs and risk of esophageal and gastric cancer. Cancer Epidemiol Biomarkers Prev 1998;7: (19) Funkhouser EM, Sharp GB. Aspirin and reduced risk of esophageal carcinoma. Cancer 1995;76: (20) Ohki S, Ogino N, Yamamoto S, Hayaishi O. Prostaglandin hydroperoxidase, an integral part of prostaglandin endoperoxide synthetase from bovine vesicular gland microsomes. J Biol Chem 1979;254: (21) Picot D, Loll PJ, Garavito RM. The X-ray crystal structure of the membrane protein prostaglandin H2 synthase-1. Nature 1994;367: (22) Loll PJ, Picot D, Garavito RM. The structural basis of aspirin activity inferred from the crystal structure of inactivated prostaglandin H2 synthase. Nat Struct Biol 1995;2: (23) Lupulescu A. Effect of prostaglandins on protein, RNA, DNA and collagen synthesis in experimental wounds. Prostaglandins 1975;10: (24) Lupulescu AP. Cytologic and metabolic effects of prostaglandins on rat skin. J Invest Dermatol 1977;68: (25) Lupulescu A. Enhancement of carcinogenesis by prostaglandins in male albino Swiss mice. J Natl Cancer Inst 1978;61: (26) Lupulescu A. Enhancement of carcinogenesis by prostaglandins. Nature 1978;272: (27) Goodwin JS, Messner RP, Bankhurst AD, Peake GT, Saiki JH, Williams RC Jr. Prostaglandin-producing suppressor cells in Hodgkin s disease. N Engl J Med 1977;297: (28) Sébahoun G, Maraninchi D, Carcassonne Y. Increased prostaglandin E production in malignant lymphomas. Acta Haematol 1985;74: (29) Cayeux SJ, Beverley PC, Schulz R, Dorken B. Elevated plasma prostaglandin E2 levels found in 14 patients undergoing autologous bone marrow or stem cell transplantation. Bone Marrow Transplant 1993;12: (30) Hsu SM, Hsu PL, Lo SS, Wu KK. Expression of prostaglandin H synthase (cyclooxygenase) in Hodgkin s mononuclear and Reed-Sternberg cells. Functional resemblance between H-RS cells and histiocytes or interdigitating reticulum cells. Am J Pathol 1988;133:5 12. (31) Santoro MG, Philpott GW, Jaffe BM. Inhibition of tumour growth in vivo and in vitro by prostaglandin E. Nature 1976;263: (32) Tutton PJ, Barkla DH. Influence of prostaglandin analogues on epithelial cell proliferation and xenograft growth. Br J Cancer 1980;41: (33) Craven PA, Saito R, DeRubertis FR. Role of local prostaglandin synthesis in the modulation of proliferative activity of rat colonic epithelium. J Clin Invest 1983;72: (34) de Mello MC, Bayer BM, Beaven MA. Evidence that prostaglandins do not have a role in the cytostatic action of anti-inflammatory drugs. Biochem Pharmacol 1980;29: (35) Elder DJ, Hague A, Hicks DJ, Paraskeva C. Differential growth inhibition by the aspirin metabolite salicylate in human colorectal tumor cell lines: enhanced apoptosis in carcinoma and in vitro-transformed adenoma relative to adenoma relative to adenoma cell lines. Cancer Res 1996;56: (36) Hanif R, Pittas A, Feng Y, Koutsos MI, Qiao L, Staiano-Coico L, et al. Effects of nonsteroidal anti-inflammatory drugs on proliferation and on induction of apoptosis in colon cancer cells by a prostaglandin-independent pathway. Biochem Pharmacol 1996;52: Journal of the National Cancer Institute, Vol. 96, No. 4, February 18, 2004 ARTICLES 313