Non-steroidal Anti-inflammatory Drug Use and Risk of Adult Leukemia¹

Leukemia accounts for ∼5% of all newly diagnosed malignancies in the United States (1, 2). In adults, the most common types of leukemia are AML,³ chronic myeloid leukemia, and CLL; acute lymphoblastic leukemia is quite rare. Little is known regarding the etiology of leukemia in adults. Benzene, chemotherapeutic drugs, ionizing radiation, and cigarette smoking are known to increase the risk of certain subtypes of leukemia (1, 3); however, taken together, these factors likely account for only a very small proportion of leukemia cases.

Regular use of aspirin or other NSAIDs has been associated with a decreased risk of colon cancer (4, 5, 6, 7, 8) and possibly of other sites (8, 9, 10, 11). To date, few studies have regarded possible associations between NSAID use and the development of leukemia. This prospective study explored whether regular use of aspirin or other NSAIDs may protect against the development of leukemia in older adult women.

Detailed descriptions of the IWHS have been published elsewhere (12, 13, 14). Briefly, participants in the IWHS were a random sample of women ages 55–69 years listed on the 1985 State of Iowa driver’s license list. A total of 41,836 women responded to a mail survey in January 1986 (42.7% response rate).

The baseline mail survey assessed medical history, anthropometric data, diet, and risk factors for cancer. Use of aspirin and other NSAIDs were assessed in a 1992 follow-up mail survey. Respondents were asked to complete two questions regarding NSAID use: (a) “How often do you take aspirin?” (several common brand names were provided as examples); and (b) “How often do you take other non-steroidal anti-inflammatory drugs or arthritis medicines?” (Nonaspirin NSAID product names were provided, including common brand names for ibuprofen, naproxen, sulindac, and piroxicam. Respondents were instructed not to include aspirin, acetaminophen, or cortisone-type drugs.) Response options included six categories ranging from never to six or more times per week. Diagnoses of rheumatoid arthritis and osteoarthritis were also assessed in the 1992 follow-up questionnaire. Women were excluded who did not complete the 1992 follow-up questionnaire (n = 8819), did not complete at least one of the NSAID questions (n = 258), or were diagnosed with any malignancy (other than nonmelanoma skin cancer) before 1993 (n = 4535). After these exclusions, 28,224 women comprised the at-risk cohort for leukemia.

Incident cancer was determined using the Health Registry of Iowa, part of the National Cancer Institute’s Surveillance, Epidemiology, and End Results Program (15). Incident cases were identified through computer matching between the 1993–2000 Registry cases and study participants. A total of 81 women in the at-risk cohort was identified as developing leukemia during the 8-year follow-up period. The leukemias included 35 CLL, 28 AML, 5 chronic myeloid leukemia, 4 chronic myelomonocytic leukemia, 4 basophilic, 3 acute not otherwise specified, 1 acute lymphoblastic leukemia, and 1 hairy cell (16).

Person-years of follow-up were calculated from completion of the 1992 follow-up questionnaire (January 1, 1993) until one of the following events: (a) date of diagnosis of leukemia; (b) date of death (if death occurred in Iowa); (c) date of move from Iowa (exact date, if known) or interval midpoint, if move date uncertain; and (d) the midpoint between the last contact and date of death (if death occurred outside of Iowa). Emigration was ascertained via the follow-up questionnaire in 1987 by an address correction request or via the National Death Index. If none of these events occurred, the women were assumed to be living in Iowa at the end of the follow-up date on December 31, 2000. χ² and t tests were performed to determine whether characteristics differed according to leukemia status. Associations of aspirin and nonaspirin NSAIDs with incident leukemia were calculated from age-adjusted and multivariate RRs, along with 95% CIs using Cox proportional hazards regression (17). Preliminary analyses were adjusted for age only; subsequent analyses were adjusted for age, smoking (ever/never and pack-years), and education. A Wald χ² test for trend was performed by treating the categorical variables of intake as a continuous variable in the model.

Women who developed leukemia reported significantly less frequent aspirin use than those who did not develop leukemia (Table 1). In contrast, those who developed leukemia reported more frequent nonaspirin NSAID use than those who did not develop leukemia, although this difference was not statistically significant. No statistically significant differences between cases and noncases were observed for mean age, education level, or smoking.

Significant inverse associations were observed between aspirin use and incident leukemia, when aspirin use was categorized into three groups based on frequency of use (Table 2). Those who reported using aspirin more than or equal to two times per week were 0.45 times as likely to develop leukemia as those who reported never using aspirin (95% CI: 0.27–0.75). Similar inverse associations, which were statistically significant, were observed when the two major subgroups of leukemia were analyzed separately, i.e., AML and CLL. We also excluded cases (n = 20) diagnosed in the first 2 years of follow-up (1993–1994). Although fewer relationships approached statistical significance, there were no notable differences in results (data not shown).

In contrast, for nonaspirin NSAID use, there was a nonsignificant positive association observed with all leukemias (more than or equal to two times per week versus none, RR = 1.31, 95% CI: 0.77–2.22; Table 3). Analyses by subgroups revealed that this was accounted for primarily by AML. For all analyses, the fully adjusted models did not differ significantly from models that were adjusted for age only, nor did they differ from models that included terms for rheumatoid arthritis, osteoarthritis, body mass index, or energy intake.

Finally, we examined the relation between incident leukemia and NSAID usage categorized as none, aspirin only, nonaspirin NSAIDs only, and both kinds of NSAIDs, with non-NSAID users as the reference group. Women reporting usage of both types of NSAIDs were 0.52 times as likely to develop leukemia as those reporting no usage of any kind of NSAIDs (95% CI: 0.28–0.95), similar to women who reported aspirin usage only (RR = 0.4, 95% CI: 0.22–0.71). In contrast, there was no inverse association observed for women who reported nonaspirin NSAID use only.

In our prospective study, the observed lower risk of leukemia with aspirin use is consistent with numerous studies reporting an inverse association of aspirin with other cancers (4, 5, 6, 7, 8, 9, 10, 11). No reduction in leukemia risk was associated with use of NSAIDs other than aspirin, contrary to expectations that NSAIDs might have similar effects as a drug class. It should be noted, however, that many studies have not explored NSAIDs separately. Our observation might suggest that properties unique to specific NSAIDs may be etiologically important.

All aspirin-like drugs have antipyretic, analgesic, and anti-inflammatory properties, but there are important differences in their activities (18, 19, 20). Aspirin (acetylsalicylic acid) and other NSAIDs act as COX inhibitors, thereby inhibiting prostaglandin synthesis and prostaglandin-induced cellular immunity. The COX enzyme exists as two isoforms: (a) COX-1 and (b) COX-2. COX-1 is constitutively expressed in several tissues and is the only isoform expressed in the gastrointestinal tract and platelets. It is primarily responsible for the biosynthesis of eicosanoids protecting the gastric mucosa and the maintenance of platelet function. COX-2 is up-regulated typically in response to inflammation and injury. Aspirin, but not other NSAIDs, irreversibly inhibits COX-1 activity in platelets and platelet aggregation.

Although aspirin has potentially unique properties that may confer a protective effect, it is unclear why nonaspirin NSAIDs had either no effect or the suggestion of a slight increased risk of leukemia, particularly AML. This category comprised several different NSAIDs, including ibuprofen, sulindac, piroxicam, and naproxen, so subgroup analysis was impossible. Although all of these NSAIDs have properties similar to aspirin, they each also have different half-lives, various effects on COX inhibition, and differential effects on specific p450 enzymes (18, 20). Moreover, people tend to be less familiar with some of these drugs as compared with aspirin and therefore may report them poorly. Nevertheless, further investigation of this potential association is needed.

Given the preliminary nature of this analysis, several limitations need to be considered. The results of this study are based on only 81 incident leukemia cases that occurred over an 8-year period. Although the prospective nature of this study avoids the problem of differential recall bias, analysis in a larger population is warranted. Although the relatively homogenous nature of our cohort may decrease residual confounding, it may also limit the generalizability of our findings. Additionally, exposure was assessed by self-report; therefore, some misclassification may have occurred. No information was collected regarding duration of use, which some previous studies have suggested may be important (6, 7). There is also the possibility that underlying disease could influence NSAID use, particularly aspirin (e.g., one possible explanation for our findings is that persons with compromised platelet function caused by premalignant hematopoeitic dysfunction may avoid aspirin because of bruising or bleeding problems). Analyses that excluded cases diagnosed before 1995, however, did not notably alter the results. Lastly, the similar associations observed for both AML and CLL might be unexpected given that these leukemias have distinctly different biological and clinical behaviors (1). The biological mechanism(s) by which NSAIDs exert their anticarcinogenic effects, however, is still unclear. In addition to the inhibition of COX-2, other suggested mechanisms include inhibition of angiogenesis, induction of apoptosis, disruption of signal-transduction pathways, and inhibition of oxidative DNA damage (21, 22), all of which may influence risk of any malignancy. A strength of this study was the ability to separately examine the effects of aspirin and nonaspirin NSAIDs. Several studies have grouped aspirin and nonaspirin NSAIDs together, preventing observation of potentially differing associations. In addition, we were able to consider several potential confounders in the analyses, including age, education, previous blood transfusion, smoking status, rheumatoid arthritis, and osteoarthritis. In summary, given the growing body of evidence implicating aspirin as a chemopreventive agent and the lack of etiological data with respect to adult leukemia, it will be necessary for other large cohort studies to evaluate NSAID use and leukemia risk to confirm or refute these findings.