Use of Nonsteroidal Antiinflammatory Drugs and Non-Hodgkin Lymphoma: A Population-based Case-Control Study

Abstract

The association between long-term use of nonsteroidal antiinflammatory drugs (NSAIDs) and non-Hodgkin lymphoma (NHL) was examined using data collected between October 2001 and May 2004 in an ongoing population-based case-control study in the San Francisco Bay Area. NHL cases were identified using rapid case ascertainment and Surveillance, Epidemiology, and End Results registry data. Control participants were frequency-matched to cases by age, sex, and county of residence. Participants completed in-person interviews designed to measure potential NHL risk factors. Questions were asked regarding use during the past 20 years of aspirin, prescription and over-the-counter nonselective NSAIDs, and cyclooxygenase-2 (COX-2) inhibitors. A total of 1,000 cases and 1,060 controls contributed data for these interim analyses. Analyses were carried out for men and women and for both sexes combined. After adjustment for age and sex, there was no consistent association between long-term use and NHL for all NSAIDs combined, aspirin, nonselective NSAIDs, and COX-2 inhibitors. For women, long-term aspirin use may be associated with a decreased risk of NHL (for 3–<9 years of use, odds ratio = 0.41, 95% confidence interval: 0.18, 0.94). Conversely, although the confidence intervals were wide, the adjusted odds ratios for COX-2 inhibitor use were nearly twofold for women, indicating a possible increase in NHL risk associated with regular use of COX-2 inhibitors.

Non-Hodgkin lymphoma (NHL) is the seventh most common cancer diagnosis and the sixth most common cause of cancer death in the United States (1). It is estimated that in 2006, 58,870 new cancer cases and 18,840 cancer deaths will be attributable to NHL (1). The disease is more common in men than in women, and incidence and mortality rates have been consistently higher for Whites than for Blacks/African Americans and persons in other race/ethnicity groups (2, 3). Surveillance, Epidemiology, and End Results data show that incidence rates for all NHLs increased at a rate of 4 percent per year for men and 3 percent per year for women from the early 1970s to the 1990s, although data collected during the 1990s indicated that these increases may be diminishing slightly (4, 5).

The utility of nonsteroidal antiinflammatory drugs (NSAIDs), especially cyclooxygenase-2 (COX-2) inhibitors, as cancer chemopreventive agents has been the focus of numerous experimental and observational studies. Experimental research has demonstrated the ability of these agents to inhibit tumorigenesis through proapoptotic, antiangiogenic, antimetastatic, and immune-modulating processes. These results have been supported by a wealth of observational evidence indicating that long-term NSAID use is associated with decreased risks of many cancers, including cancer of the colon/rectum (6–11), pancreas (12), prostate (13–15), lung (6), breast (6, 16–18), esophagus (19, 20), stomach (19, 20), bladder (21), and ovary (22). The antitumorigenic effects of NSAIDs have been attributed primarily to their ability to inhibit the activity of the COX-2 enzyme. However, in-vivo and epidemiologic studies have found that regular users of non-COX-2-selective NSAIDs, including aspirin, also have a lower incidence of colorectal (23–25), esophageal (19, 20), lung (6, 26, 27), and bladder cancers (21). Even NSAIDs having no affinity for cyclooxygenase have demonstrated antitumorigenic effects (28).

COX-2 levels are upgraded during periods of sustained inflammation, and inflammation is a critical part of most tumorigenic processes, including lymphomagenesis (29). The mechanistic relation between COX-2 expression and NHL is unclear, but it may involve activated macrophages and the release of cytokines and prostaglandins (30, 31). Therefore, exploring the association between NSAID use and NHL could enhance our understanding of lymphomagenesis and other tumorigenic processes.

Few epidemiologic studies have examined the association between NSAID use and NHL. In 1999, a population-based case-control study we conducted in the San Francisco Bay Area showed a decreased risk of NHL among users of nonselective NSAIDs (32). A recent case-control study showed similar results among men who were regular users of aspirin (33). However, other observational studies showed null associations (6, 34–36) or an increased risk among regular users of NSAIDs, primarily among women (37–40). These inconsistencies necessitate additional research to clarify the NSAID-NHL association and to examine whether the association differs between women and men. In addition, the withdrawal of rofecoxib from the market in September 2004 because of adverse cardiovascular events has generated concern regarding the use of any NSAIDs, further emphasizing the importance of clarifying the risks or benefits of long-term NSAID use. Consequently, we examined the association between long-term NSAID use and NHL using interim data collected from the first 2,060 participants in an ongoing population-based case-control study of NHL in the San Francisco Bay Area.

MATERIALS AND METHODS

Case ascertainment

The research protocol was approved by the Committee for Human Research at the University of California, San Francisco. Participants were persons diagnosed with incident NHL between October 2001 and May 2004 in the California counties of San Francisco, Alameda, Marin, Contra Costa, San Mateo, and Santa Clara. Eligible cases were those who were San Francisco Bay Area residents at the time of diagnosis, aged 21–85 years at the time of diagnosis, and able to complete an interview in English. The Northern California Cancer Center used rapid case ascertainment to identify incident cases within 1 month of diagnosis in the six-county Bay Area hospitals. Cases not identified through rapid case ascertainment were identified through the Northern California Cancer Center's Surveillance, Epidemiology, and End Results registry (median time from diagnosis to interview, 140 days). Eligible cases provided written consent. A total of 415 women and 585 men contributed valid information and were included in the final analyses.

A response rate for the cases was not computed because of outstanding interviews at the time of these analyses. An estimate of the response rate would range between 45 percent for a worst-case scenario and 75 percent for a best-case scenario. If none of the cases who were assigned were interviewed, the response rate would be approximately 45 percent, whereas if all of the cases who were assigned were interviewed, the response rate would be approximately 75 percent. The response rate estimate is based upon current data and data from our previous NHL study (32).

Controls

Control participants were a random sample of persons identified by random digit dialing or from Centers for Medicare and Medicaid Services files. Eligible controls had no history of NHL, resided in the same six counties as the cases at the time of diagnosis, were aged 21–85 years, and were able to complete an interview in English. Details of the random digit dialing procedure used to identify controls have been reported previously (41). Control participants were frequency-matched to cases by sex, age (within predefined 5-year age groups), and county of residence to ensure that they were from the same study base as the cases. By May 2004, valid information had been collected from 501 women and 559 men.

Exposure assessment

The methods of data collection emulated those used in our earlier study (32). In-person interviews were conducted in participants' homes by experienced interviewers. Data were collected on medical, diet, exercise, family, and lifestyle factors prior to 1 year before diagnosis. Detailed questions regarding the use of aspirin, prescription nonselective NSAIDs, over-the-counter nonselective NSAIDs, and COX-2 inhibitors were included in the questionnaire. Participants were asked whether they had ever used prescription nonselective NSAIDs in the past 20 years (yes/no) and whether they had ever used a COX-2 inhibitor since 1999 (yes/no). Use of aspirin and over-the-counter nonselective NSAIDs was ascertained from a series of questions that asked about use of these products for at least 2 days per week for 3 months or longer during the past 20 years. An affirmative answer to any of these questions was followed by several questions regarding the name(s) of any drug(s) used, years of use, and frequency of use. A list of brand and generic names was provided to enhance recall. We asked participants to exclude NSAID use within 1 year prior to diagnosis (cases) or interview (controls) to minimize bias that could have been introduced by use of NSAIDs for early, subclinical NHL symptoms.

Covariate assessment

Established or potential risk factors for NHL that potentially could be correlated with long-term NSAID use included various lifestyle factors, comorbid conditions, and use of other medications. Potential confounders evaluated in these analyses included: race/ethnicity (non-Hispanic White, African-American/Black, Hispanic, Asian, or other); body mass index (weight (kg)/height (m)²; World Health Organization categories: <18.5, 18.5–24.9, 25–29.9, and ≥30); education (less than high school, high school graduation or some post-high-school education, and college graduation); human immunodeficiency virus (HIV) status; immune-system-related conditions (e.g., rheumatic diseases (ever/never)); history of chronic or recurrent infections (e.g., shingles, bladder infections (ever/never)) (3); marital status; physical activity (low, moderate, or high); history of osteoarthritis, hypertension, heart disease, type 2 diabetes, or other chronic diseases; plant allergies; hormone replacement therapy; and use of acetaminophen, antihistamines (for allergies and/or insomnia), antiviral agents, ulcer medications, and cholesterol-lowering drugs (3, 32, 34, 35, 37, 38).

Statistical analyses

We performed unconditional logistic regression analyses using SAS 9.1 (SAS Institute, Inc., Cary, North Carolina) to estimate odds ratios and 95 percent confidence intervals. Evidence from previous studies suggested that sex modifies the NSAID-NHL association (32–34, 38). Therefore, analyses were conducted both for men and women combined and stratified by sex. For all analyses except those for the COX-2 inhibitors, the reference group (hereafter referred to as “<90-day users”) consisted of persons who reported never having used NSAIDs or having used them for fewer than 90 days over the past 20 years. For the COX-2 inhibitor analyses, the reference group consisted of persons who reported never having used any COX-2 inhibitors. However, these persons may have used aspirin or nonselective NSAIDs for fewer than 90 days over the past 20 years. The exposed were categorized by tertiles of total days of use among the controls. In the first set of analyses, tertiles of all NSAID use were compared with use for less than 90 days. If the odds ratio estimates increased or decreased linearly, we conducted a linear test for trend based on the beta estimate computed for the exposure when it was analyzed as an ordinal variable in an adjusted unconditional logistic regression analysis. In a second set of analyses, we used similar methods to examine the associations between specific categories of NSAIDs and NHL.

Using contingency-table analyses, we examined the confounder-exposure associations among the controls and the confounder-disease associations among the unexposed for each covariate considered. On the basis of the univariate results, factors with cell sizes greater than 5 and two-sided p values less than 0.10 were explored further in the multivariable models. In all models, results were adjusted for age and sex, with age grouped into the categories <35, 35–39, 40–44, …, 75–79 years and ≥80 years. We fitted a full model containing all potential confounders to the data using backward elimination by removing variables one at a time according to the highest Wald p value, after determining that removal of a given variable did not change the exposure coefficients by more than 10 percent.

HIV-infected persons may be more likely to use NSAIDs (42), so we addressed the potential for confounding by indication. Our data included few HIV-positive study participants (45 cases and three controls). Thus, the potential for uncontrolled confounding by HIV status was small, and restriction to HIV-negative participants did not change the odds ratio estimates. Therefore, we included HIV-positive participants in the final models. We also examined confounding by indication for other conditions potentially associated with NHL, including shingles and rheumatic disease.

Prior studies have examined whether a minimum duration of NSAID exposure is associated with NHL risk (34, 39). Therefore, we examined whether exposure levels defined as ≤2 years and >2 years, ≤5 years and >5 years, and ≤10 years and >10 years of NSAID use would change the effect estimates for all NSAIDs combined, for aspirin users, and for nonselective NSAID users.

RESULTS

The study population consisted of 415 NHL case women and 501 control women and 585 NHL case men and 559 control men (table 1). The majority of the participants were at least 60 years of age (65 percent of women, 59 percent of men) and non-Hispanic White (approximately 76 percent of women and 80 percent of men). Approximately 60 percent of women and 54 percent of men reported having taken NSAIDs in the past 20 years. A greater proportion of men than women reported the use of aspirin, whereas a greater proportion of women than men reported the use of nonselective NSAIDs and COX-2 inhibitors. The use of specific types of NSAIDs was similar in cases and controls, except for aspirin use. A larger proportion of controls than cases reported using aspirin, especially among women (22 percent of cases, 32 percent of controls).

None of the covariates considered as potential confounders changed the odds ratio estimates by more than 10 percent; thus, these factors were ultimately removed from the models. In general, total NSAID use was not associated with NHL among women or men or among both sexes combined (table 2). Among women, a 28 percent reduction in risk for NHL was found among those who reported >8–20 years of total NSAID use as compared with <90-day users, although the 95 percent confidence interval included unity. The adjusted odds ratios for total NSAID use among women and men combined decreased linearly, although all estimates could have occurred by chance and the two-sided p value for trend was 0.44.

Age- and sex-adjusted analyses were stratified by use of aspirin, prescription and over-the-counter nonselective NSAIDs, and COX-2 inhibitors (table 3). Adjustment for the covariates of interest did not appreciably change the odds ratio estimates. Each NSAID analysis was restricted to persons who reported use of that type of NSAID as compared with <90-day users. Because 17 percent of the study participants reported having taken multiple types of NSAIDs over the past 20 years, there was a decrease in sample size for each set of analyses by NSAID type.

Among aspirin users, the adjusted odds ratio estimates were null, with one exception among women. Women who reported using aspirin for approximately 3–<9 years had a 59 percent reduction in NHL risk (table 3), although there was no clear duration response. No association was found between the use of nonselective NSAIDs and NHL risk in comparison with <90-day users in any analysis. Analyses of COX-2 inhibitor use were limited by small sample sizes, especially among men, despite our combining tertile groups. As a result, wide confidence intervals containing 1.0 were present in all of the analyses. Unlike aspirin, the adjusted odds ratio for COX-2 inhibitor use was nearly twofold for women, indicating a possible increase in NHL risk associated with regular COX-2 inhibitor use. The results of the 2-year, 5-year, and 10-year sensitivity analyses were similar to the overall results and are not presented here.

DISCUSSION

No consistent associations were observed in our analyses between long-term NSAID use and NHL for all NSAIDs combined or for aspirin, nonselective NSAIDs, or COX-2 inhibitors. Long-term aspirin use (3–<9 years) was associated with a decreased risk of NHL in women, whereas COX-2 inhibitor use was found to possibly increase NHL risk only in women. However, biologic inferences are limited given that results were based on few exposed cases and the dose-response trend was nonlinear. Results from these more detailed analyses regarding NSAID use differ from our previously published results (32, 43) that showed a consistently reduced risk of NHL with regular use of nonselective NSAIDs. In the earlier study, NSAID exposure was ascertained from an open-ended question about use of any medications for 4 consecutive weeks or longer. Data regarding type of NSAID used and duration of use were not collected in the earlier study.

Results have been mixed from observational studies that examined the association between NSAID use and NHL. To our knowledge, only our earlier observational study has demonstrated a reduced risk of NHL with NSAID use that has differed from chance (32, 43), although decreased risk estimates have been reported by other investigators. A case-control study conducted in Buffalo, New York, showed that regular aspirin use may reduce the risk of NHL among men but not among women. However, the confidence intervals included unity (33). A nested case-control study conducted in the Netherlands showed a reduced risk with ever use of NSAIDs compared with never use, although estimates were not different from chance after adjustment for comorbidity and follow-up time (34). A reduced risk of lymphoma was also observed in a large, prospective study of men and women that compared persons who reported aspirin use within the past 30 days with persons who reported no use (6). Case-control studies conducted in Yorkshire, United Kingdom (35) and the US state of Connecticut (36) reported no consistent association between use of antiinflammatory drugs and NHL.

Conversely, several studies have found increased risk estimates for NHL associated with NSAID use that differed from unity. Recent results from the Scandinavian Lymphoma Etiology Study showed an increased risk of NHL among persons who reported using NSAIDs daily for more than 10 years (40). Investigators in the Iowa Women's Health Study reported more than a doubling of risk of NHL among postmenopausal women associated with the use of aspirin only and more than a threefold increased risk for nonaspirin NSAIDs only in comparison with never use of any NSAIDs, after excluding the first 2 years of follow-up (37). Similarly, an increased risk of NHL was observed among NSAID users in a population-based case-control study conducted among women in Upstate New York (39). Women who used NSAIDs and acetaminophen daily for more than 10 years had two- to threefold increased risks of B-cell and extranodal lymphoma as compared with women who took these agents less than once per month. Results from a case-control study conducted in Los Angeles, California, showed a similar doubling of risk for NHL associated with use of aspirin or other pain relievers for 1 month or longer among women only and among men and women combined (38).

There are several possible explanations for the inconsistency in results across studies. First, NSAIDs represent a chemically and pharmaceutically diverse group of agents with varying abilities to bind to cyclooxygenase-1 and COX-2 receptors. Effect estimates may have been altered in studies that grouped nonselective NSAIDs into one category or acetaminophen use into the NSAID category, especially if COX-2 inhibition is associated with increased risk of NHL. Additionally, a large proportion of people take two or more types of NSAIDs over a long period of time, and reported results are likely to reflect a limited ability to distinguish and compare specific NSAID exposures. Second, because of changes in drug availability over time, the time period of data collection is pertinent to the use of specific NSAIDs and to possible duration of use by participants across studies. With the exception of aspirin, indomethacin, and ibuprofen, most NSAIDs became commercially available only within the past 25 years, and the COX-2 selective agents became available only within the past 7 years. Third, there has been little consistency in dose, frequency, and duration categories used across studies, making direct comparisons of dose effects difficult. Finally, few investigators have determined the NSAID-NHL association by NHL subtype. If the effect varies by subtype, the proportion of a given NHL subtype in a study population may affect whether NSAID use appears beneficial, deleterious, or null.

Our study possessed several strengths. Rapid case ascertainment enabled us to identify newly diagnosed patients at a relatively early stage of illness, which was likely to have improved participation rates. Additionally, frequency matching by age, sex, and county of residence ensured that cases and controls were selected from the same study base. We also ascertained NSAID exposure over a period of 20 years, a length of time relevant to cancer risk. Finally, the large sample size and detailed questionnaire enabled us to examine the NSAID-NHL association by NSAID type while adjusting for a number of potential confounders.

A limitation of this study was the potential for bias associated with recall of exposures incurred up to 20 years in the past. However, recall of regular long-term NSAID use should be relatively accurate, because regular use often is associated with a health condition or injury. Errors in recall are more likely to be related to rare or occasional use that is less likely to be related to cancer risk. We combined <90-day users with nonusers to alleviate this concern. We also made an effort to minimize missing data and exposure misclassification by conducting in-person interviews in the participants' homes, where participants had access to medications, and by providing lists of brand names and generic names of medications to enhance recall.

We found that long-term NSAID use was not associated consistently with NHL in women or men. These results are an important contribution to the literature, because in our earlier study we found a statistically significantly reduced risk of NHL and NHL subtypes among NSAID users (32, 43). Using the same population-based case-control study design, we were able to reassess the NSAID-NHL association in a new study population using detailed questions on NSAID use that included brand names and generic names. Furthermore, the recent wave of public concern regarding the safety of long-term use of NSAIDs, particularly COX-2 inhibitors, makes this result particularly timely. In comparison with randomized clinical trials, observational studies have the capability to assess an exposure-outcome association over a longer time period and provide an opportunity to explore factors that might distort the association between exposure and disease. Analyses by histologic subtype and later pooling of these data with those of the International Consortium on Lymphoma (InterLymph) will add substantially to our knowledge of the association between NSAID use and NHL.

This work was supported by grants CA87014, CA89745, and CA45614 from the National Cancer Institute (NCI).

E. Dawn Flick and Dr. Arnold Chan received partial support from the Harvard School of Public Health Pharmacoepidemiology Program. E. Dawn Flick also received an honorarium for this project from the University of California, San Francisco (UCSF), Program for Hematopoietic Cancers in the UCSF, NCI-sponsored Comprehensive Cancer Center.

Conflict of interest: none declared.

References