In our study, regular/extra strength aspirin as well as other anti-inflammatory analgesic use was associated with a decreased risk of myeloid leukemia in women but not men. In contrast, acetaminophen use was associated with an increased risk of myeloid leukemia for women only. We found no associations with low dose aspirin nor ibuprofen with either sex or leukemia subtype.
We are aware of only three studies that have explored the association between over-the-counter analgesic use and risk of adult leukemia. In a prospective cohort study of over 28,000 postmenopausal women in Iowa, Kasum et al(10
) reported a 55% (95% CI= 0.27–0.75) decreased risk of leukemia (81 cases, mostly AML and CLL) in women who reported using aspirin two or more times weekly compared to women who never used aspirin. In contrast, the authors found a slight 31% (95% CI=0.77–2.22) increased risk of leukemia with use of non-aspirin NSAIDs. In a case-control study of acute leukemia (169 cases, 676 controls) in New York, Weiss et al(11
) reported a 16% (95% CI=0.59–1.21) non-statistically significant decreased risk of acute leukemia with aspirin use; results were stronger for ALL than AML and were found for both sexes. Further, they reported a statistically significant 53% (95%CI=1.03–2.26) increased risk of acute leukemia associated with use of acetaminophen. Lastly, in a prospective cohort study, Walter et al(12
) reported a statistically significant 130% increased risk (95% CI=1.12–4.73) of AML and myelodysplastic syndromes combined (N=90) with high (≥4 days/week for ≥4years) acetaminophen use compared to lower use. They found no evidence of an association with NSAID use, including aspirin. Of note, the increased risk of hematologic malignancy associated with acetaminophen in the overall cohort appeared confined mostly to women.
Although based on a small number of cases, we also found an increased risk of CML associated with use of COX-2 inhibitors in women. Only three studies to our knowledge have evaluated the risk of adult leukemia and use of prescription NSAIDs. In the United Kingdom, Bhayat et al(17
) linked data from over 330 general practices, which included medical and prescribed-drug histories on over five million patients. Over 3200 leukemia patients were identified along with four age, gender, and practice matched controls per case. In an examination of NSAID prescription rate prior to leukemia diagnosis and risk of leukemia, the authors found no overall associations with the possible exception of an increased risk of CLL with 2–5 NSAID prescriptions per year. Pogoda et al(18
) reported a 50% decreased risk of AML in a case-control study (299 matched pairs) with use of a prescription NSAID at least 2 years prior to diagnosis. Lastly, Traversa et al(19
), in a linkage study in Italy of adult leukemia (202 cases, 2020 controls) and prescription drugs, reported a non-statistically significant 60% decreased risk of leukemia associated with prior use of prescription NSAIDs for more than 180 days. Unfortunately, none of these studies provided NSAID-specific analyses.
In addition to anti-inflammatory properties, aspirin, ibuprofen, COX-2 inhibitors, and other NSAIDs possess anti-pyretic and analgesic properties(20
). Acetaminophen also possesses the latter two properties, but it is considered only a weak anti-inflammatory drug(21
). All of these drugs target the COX enzymes, COX-1, -2, -3, although to varying degrees. For example, aspirin irreversibly inhibits both COX-1 and COX-2, whereas binding by other NSAIDs is reversible(22
); acetaminophen preferentially inhibits COX-3. COX enzymes produce prostaglandins that are formed at sites of inflammation. While it is unclear why cancer risk might be decreased by regular use of certain analgesics, analgesic-associated reductions in prostaglandin synthesis and oxidative cell damage, inhibition of angiogenesis, and disruption of signal transduction pathways have all been speculated to influence risk of malignancy(23
). Inverse associations we observed with aspirin and other anti-inflammatory analgesics, in light of positive associations observed with COX-2 and COX-3 inhibitors, may suggest that COX-1 inhibition is most relevant. However, inhibition appears to vary by drug class as well as target tissue (e.g., brain, gastrointestinal tract), which adds to the complexity in understanding mechanisms(21
Specific to hematopoiesis, there is evidence that certain drugs have differential effects on leukemia cells. For example, aspirin has been shown to induce apoptosis in AML cell lines(25
), while COX-2 inhibitors induce apoptosis in CML cell lines(24
). The increased risk of myeloid leukemia with acetaminophen could be due to the genotoxic effects of a metabolite, N
-benzoquinone imine, which has been shown to be a DNA topoisomerase II poison(26
). DNA topoisomerase II chemotherapy poisons, including etoposide, are associated with secondary leukemias, particularly AML(27
). Additionally, a few experimental studies suggest that aceteminophen is genotoxic to bone marrow and could increase risk of leukemia (reviewed in (28
)). Taken in toto, however, it is difficult to speculate on how each of these drugs may influence myeloid leukemia risk, especially in the context of a population-based study.
It is also not clear why we observed significant associations for women but not men. It is possible that these are chance findings, but given the differing directions by analgesic type (i.e., aspirin versus acetaminophen), this seems less likely. In studies of the cardioprotective effects of aspirin, there is evidence that men and women respond differently, which could be related to metabolic and/or hormonal differences (reviewed in (29
)). For example, in contrast to effects observed in men, aspirin has been shown to significantly reduce the risk of stroke but has little influence on risk of myocardial infarction or mortality in women(30
). Related to our findings, sex-specific differences have been reported with regard to analgesic use and risk of certain malignancies such as lung cancer, non-Hodgkin lymphoma, and most recently, hematopoeitic malignancies (31
). Moreover, due to the discovery of the cardioprotective benefits identified in clinical trials decades ago, at-risk men have been (and continue to be) advised by their physicians to begin an aspirin regimen early. However, women have lagged behind in these global recommendations and still have a 5–10% lower overall prevalence of aspirin use compared to men(33
). It is possible that long term increasing aspirin prevalence among men could diminish the ability to detect an association assuming that at least some of these men will now never develop leukemia. We acknowledge that we performed numerous comparisons, and thus our statistically significant results could be false positive findings. In a few instances, there was a lack of dose-response. It will be important to confirm these sex specific differences, ideally through pooled cohort studies.
Recall bias is a potential limitation to our study, which could further be influenced by conditions associated with disease status. Given the broad scope of the questionnaire, and the lack of readily available public information regarding associations between NSAIDs and leukemia, it seems unlikely that recall would be differential amongst cases and controls. Further, our observations differed by sex and leukemia subtype, and it would be unlikely that selective recall would be associated with specific groups. In order to diminish the possibility that disease status affected analgesic use, we focused on a period at least two years prior to study enrollment. Nevertheless, there is the possibility of non-differential misclassification. Further, the primary reason for using each type of analgesic was collected using an open-ended question, which resulted in a variety of responses along with some missing values. Adjustment for broad categories of reasons for analgesic use did not materially alter our risk estimates, although some residual confounding by indication could remain.
While our participation rates (58% for cases and 64% for controls) are comparable if not better than other case-control studies of rapidly fatal malignancies including leukemia and pancreatic cancer(18
), selection bias is still a concern. Our controls were recruited from the population and represent an accurate base from which cases arose; importantly, participating cases and controls were comparable in education and income. Moreover, the prevalence of reported aspirin use by controls in our study is comparable to that reported in a national survey(33
). Nevertheless, controls or cases who participated could still be fundamentally different with regard to analgesic use compared to those who did not participate. We controlled for age, body mass index, and other analgesic use in our analyses, but cannot rule out residual confounding. It is also important to note that no proxy interviews were conducted in our study. Proxy interviews are prone to misclassification and bias, which is difficult to quantify, particularly if the proportion of proxy interviews between cases and controls differs(37
). Thus, our study results may be difficult to compare to others, such as Pogoda et al, where 49% of interviews were completed by proxy(18
In conclusion, this is the largest population-based study to evaluate the use of NSAIDs and risk of adult myeloid leukemia. We found further evidence of a protective association with aspirin (at least in women) and we confirm previous findings of a positive association with acetaminophen. Given the interest in understanding the potential benefits and harm associated with over-the-counter analgesics, and the fact that leukemia is ranked fifth in person years of life lost due to cancer, further studies of these associations are warranted.