Use of Nonsteroidal Anti-Inflammatory Drugs and Pancreatic Cancer Risk in the Women’s Health Initiative

Pancreatic cancer is among the most fatal human cancers and the fourth leading cause of cancer death in the United States (1). Although incidence rates remain relatively low, they have slowly increased since the turn of the century (1). Evidence suggests that chronic inflammation may play a role in pancreatic carcinogenesis (2) and survival (3). Relatedly, chronic pancreatitis, smoking, and obesity, each inflammatory or proinflammatory states, are risk factors for pancreatic cancer (4–7). It therefore stands to reason that anti-inflammatory medications could reduce pancreatic cancer risk. Large, randomized chemoprevention trials of nonsteroidal anti-inflammatory drugs (NSAID) have been equivocal, although such studies remain limited to trials of aspirin—typically at daily doses with only marginal inhibition of cyclooxygenase-2 (COX-2) and downstream prostaglandins (8, 9), the primary hypothesized biological pathway of effect—and constrained to fewer than 100 pancreatic cancer cases (10–12). However, in a pooled analysis of six aspirin trials with daily aspirin doses ranging from 75 to 500 mg, Rothwell and colleagues (13) reported strong reductions in pancreatic cancer mortality based upon <50 pancreatic cancer deaths. Data from prospective observational studies, which have examined an array of NSAIDs either grouped as a class or individually, have been inconsistent (14–16), yet results from meta-analyses indicate inverse associations for aspirin rather than non-aspirin NSAIDs (15, 17–19), and a possible period effect of inverse associations for aspirin among more modern studies (20, 21).

In a previous analysis among postmenopausal women participating in the Women’s Health Initiative (WHI), Brasky and colleagues (22) observed statistically nonsignificant, inverse associations between consistent use of NSAIDs, aspirin in particular, with pancreatic cancer risk after 10 years of follow-up and 378 incident cases in an analysis focused on total cancer incidence. Herein, we update that analysis with an additional 7 years of follow-up and inclusion of an additional 349 incident cases in a focused examination on NSAIDs and pancreatic cancer incidence.

The WHI is an ongoing prospective study of postmenopausal women consisting of a multifactorial clinical trial (CT; Trial registration: clinicaltrials.gov identifier, NCT00000611) and an observational study (OS). The study’s methods are detailed elsewhere (23–25). Briefly, 161,808 postmenopausal women, aged 50 to 79 years, were recruited at 40 U.S. clinical centers between 1993 and 1998. The CT included two placebo-controlled hormone therapy trials [estrogen-alone (n = 10,739) and estrogen plus progestin (n = 16,608)], a low-fat diet modification compared to usual diet trial (n = 48,836), and a calcium/vitamin D placebo-controlled trial (n = 36,282; refs. 26–28). Participants in the OS were 93,676 women who were screened for participation in the CT but were ineligible or unwilling to participate or who were recruited directly (29). The WHI did not collect data on personal histories of pancreatic cancer prior to baseline, however study participants were excluded from WHI if they had a medical condition predictive of a survival time <3 years, meaning there were unlikely to have been prevalent pancreatic cancer cases at baseline. After the original WHI studies ended in 2005, several extension studies were conducted to collect additional follow-up data. Participants provided written informed consent for participation in both the original and extension studies. Human Subjects Review Committees at all participating institutions approved the WHI study protocol.

We excluded participants with missing medication data at either WHI study’s original baseline or third year of follow-up, including those whose study follow-up ended before the third year of the WHI study (n = 21,703), those with a diagnosis of pancreatic cancer between the WHI study’s original baseline and third year of follow-up (n = 21), those with baseline age outside the WHI age range of 50 to 79 year (n = 8), and those with missing data on characteristics used for adjustments in our analyses (n = 22,678). After exclusions, there were 117,452 postmenopausal women available for inclusion in the analysis, including n = 727 incident pancreatic cancer cases.

WHI participants attended baseline screening visits, during which they completed self-administered questionnaires that collected detailed information on demographics, medical and reproductive history, family history of cancer, physical activity, and other risk factors. Trained clinical staff measured participants’ height and weight, from which body mass index (BMI; kg/m²) was calculated.

A computerized medication inventory was developed to capture usual current medication use (24). Participants were asked to bring prescription and over-the-counter medications used ≥2 times/week over the prior 2 weeks to their clinic visit to facilitate completion of a computer-assisted interview about current medication use. Medications were classified by their therapeutic class and National Drug Code, associating each medication with its active ingredients. Using this ingredient information, we summarized participants use of NSAIDs, including aspirin, ibuprofen, naproxen, and other NSAID preparations (e.g., indomethacin, meloxicam), as well as non-NSAID analgesics including acetaminophen. Medication data were queried at baseline and year 3 in the OS and CT, and in years 6 and 9 in the CT only. To minimize measurement error, we categorized medication use as none, inconsistent, and consistent, corresponding to non-use at baseline and the year 3 visit, use at baseline or year 3 only, and use at both baseline and year 3, respectively. Duration of medication use was reported at year 3 and was dichotomized (<5 and ≥5 years). Analyses of medication duration were restricted to consistent users. Summary variables were created to account for uses of any NSAID (including prescription and over-the-counter preparations), any aspirin, and any non-aspirin NSAID. We defined low-dose aspirin as preparations containing ≤100 mg.

Incident pancreatic cancer cases were reported annually by questionnaire in the OS and semi-annually in the CT. Physician adjudicators reviewed participants’ medical records to confirm the diagnosis. Only confirmed pancreatic cancer diagnoses (ICD-O 25.0–25.9) after year 3, the adjusted “baseline”, were considered cases. After a median of 17.6 years of follow-up, 727 pancreatic cancer cases were identified.

Participants were right-censored from the analysis at the earliest of the following events: end of original follow-up for participants who were not enrolled in the WHI extension study (n = 20,836), end of extension 1 for participants who were not enrolled in the second WHI extension study (n = 9,724), withdrawal or loss of contact (n = 1,167), death (n = 19,679), or September 30, 2015, to represent the originally intended endpoint for the second WHI extension study (n = 65,785).

Distributions of participants’ baseline characteristics were calculated with means and standard deviations, and frequencies, stratified by their baseline NSAID use. The t tests and χ² tests were used to examine statistical differences between consistent NSAID use and non-use, as appropriate. Cox proportional hazards regression models with participants’ age at year 3 (the adjusted “baseline”) as the time metric were used to estimate adjusted hazard ratio (HR) and 95% confidence interval (CI) for associations between medication use and pancreatic cancer incidence. We adjusted for participant characteristics that were associated with NSAID use (see Table 1). Characteristics with more than 5% missing data across participants were recoded to include a “missing” category rather than excluding participants missing these characteristics. To address the potential for confounding of associations of one medication with pancreatic cancer risk by use of another, regression analyses for any one NSAID exposure were further adjusted for the use of other NSAIDs. Tests for linear trend (P trend) across categories of NSAID duration were calculated by including a three-level ordinal variable for NSAID duration (non-use, consistent use <5 years, consistent use ≥5 years) in regression models. Last, given recent evidence of interaction of aspirin with diabetes (16), we examined the possibility of interaction of associations of NSAID use with pancreatic cancer risk may be modified by factors associated with pancreatic cancer or inflammation, including prevalent diabetes, smoking, alcohol intake, and BMI. Tests of interaction (P interaction) were calculated by comparing the model that included an interaction between the factor and NSAID use with the non-interaction model using a likelihood-ratio test. The proportional hazards assumption of regression models was confirmed using Schoenfeld residual tests and visual inspection of plots of the scaled Schoenfeld residuals against time. All statistical analyses were performed using R Statistical Software (v4.1.2; R Core Team 2021). All statistical tests are two-sided with the probability of type I error set at 5%.

The data that support the findings of this study are available from the WHI (http://www.whi.org). Restrictions apply to the availability of these data that were used under permission for this analysis.

Participants’ baseline characteristics overall and stratified on total NSAIDs are given in Table 1. Overall, participants were aged 63 years on average and 41% participated in the WHI CT; 79% had at least a high school education and 12% identified as Asian, Black, American Indian, Alaska Native, or bi/poly-racial. Among the 117,452 postmenopausal women, 54,695 (46%), 34,996 (30%), and 27,761 (24%) were identified as non-users, inconsistent users, and consistent users of NSAIDs, respectively. Compared to non-users, consistent users of NSAIDs tended to be White and non-Hispanic, obese, and less physically active. NSAID users smoked more, ingested multivitamins, and were more likely to use unopposed menopausal hormones. They were also more likely to have prevalent hypertension, cardiovascular disease, diabetes, arthritis, and to use cholesterol-lowering medications.

Associations of NSAIDs with pancreatic cancer risk are given in Table 2. Consistent use of any NSAID was associated with 29% (HR 0.71, 95% CI, 0.59–0.87) reduced pancreatic cancer risk; however, there was no exposure-response gradient with increasing categories of NSAID duration, possibly reflecting a threshold effect. Among individual medications, associations were the strongest for any aspirin (HR 0.67, 95% CI, 0.52–0.86) and low-dose aspirin (HR 0.54, 95% CI, 0.33–0.89); the latter of which exhibited evidence of benefit with increased duration of use (≥5 years, HR 0.44, 95% CI, 0.21–0.93; P trend = 0.01). Consistent uses of non-aspirin NSAIDs (individually or as a class) and acetaminophen were not associated with pancreatic cancer risk.

We additionally stratified associations of any NSAID and any aspirin with pancreatic cancer by prevalent diabetes (Table 3). Although numbers of pancreatic cancer cases among participants with diabetes were small and CIs were wide, consistent use of any NSAID was associated with 72% (HR 0.28, 95% CI, 0.10–0.75) reduced pancreatic cancer risk among participants with diabetes and 25% (HR 0.75, 95% CI, 0.61–0.92) reduced risk among those without (P interaction = 0.03). Similar results were observed for use of any aspirin, however, differences were statistically nonsignificant. No interaction was observed with smoking, alcohol, or BMI (Online Supplementary Data).

In this prospective study of 117,452 postmenopausal women, including more than 727 incident pancreatic cancer cases, we observed that use of aspirin was associated with moderate reductions in pancreatic cancer risk. Individual uses of ibuprofen, naproxen, or other NSAIDs, as well as the analgesic acetaminophen, were not associated with pancreatic cancer. Observed associations were particularly strong among women with prevalent diabetes.

Results of randomized chemoprevention trials of aspirin have been mixed and reliant upon very small numbers of outcomes. Results of the WHI, a randomized, placebo-controlled trial, which administered 100 mg of aspirin every second day, reported no effect on pancreatic cancer incidence after 10 years of follow-up (HR 1.42; 95% CI, 0.81–2.49) based upon 51 cases (12). In an observational analysis after 18 years of follow-up and an additional 37 incident cases, the lack of an association persisted (HR 1.19; 95% CI, 0.78–1.81; ref. 10). Similarly, investigators of the Aspirin in Reducing Events in the Elderly randomized, placebo-controlled trial of men and women reported no short-term effect of 100 mg daily aspirin on pancreatic cancer incidence (n_cases = 66; HR 1.29; 95% CI, 0.79–2.09) or mortality (n_deaths = 50; HR 1.39; 0.80–2.45; ref. 11). In contrast, Rothwell and colleagues (13) reported a strong reduction in pancreatic cancer mortality using data pooled across six trials of aspirin in men and women (HR 0.25; 95% CI, 0.07–0.92), however, again, the numbers of pancreatic cancer events were very small (n = 45).

Results from prospective observational studies of aspirin have also been mixed, yet recent meta-analyses of observational studies suggest an inverse association of similar magnitude as we observed in the WHI. Bossetti and colleagues (15) reported a meta-RR of 0.79 (95% CI, 0.64–0.98) for aspirin use versus non-use in seven cohort studies (n = 7,759 events) of pancreatic cancer risk, although the Cancer Prevention Study II, which contributed 41% of the meta-analysis’s events and reported equivocal results (RR 0.95, 95% CI, 0.86–1.09; ref. 30), examined associations of aspirin with pancreatic cancer mortality rather than incidence. At 12%, pancreatic cancer has the lowest 5-year survival of all cancers (1), thereby lessening concerns of the inclusion of cancer death as a proxy for incidence. Another meta-analysis including five cohort studies and 720 incidence cases reported similar point-estimates for aspirin use (OR 0.78, 95% CI, 0.55–1.12; ref. 18). In contrast, recent data from the Prostate, Lung, Colorectal, and Ovarian Cancer Screening Trial (n = 878 cases; ref. 14) and an individual-participant pooled analysis of the Nurses’ Health Study of women (NHS, n = 707 cases) and Health Professionals Follow-up Study of men (HPFS, n = 415 cases; ref. 16), not included in the above meta-analyses, reported no associations for aspirin use overall (and none among women specifically in the NHS).

Prospective data for NSAIDs as a class or non-aspirin NSAIDs specifically are limited to a handful of studies (16, 22, 31, 32), including a prior analysis of the WHI, originally focused on total cancer incidence (22), the Iowa Women’s Health Study (32), the pooled analysis of the NHS and HPFS cohorts (16), and a Danish population-based registry linkage study with limited capacity to adjust for confounding factors (31). Associations were found for neither non-aspirin NSAIDs nor NSAIDs as a whole. As with the current study, these prior reports did not adjust for prevalent pancreatitis. Prescription and over-the-counter analgesics, including non-aspirin NSAIDs, are indicated in pancreatitis treatment (33). The condition is additionally a pancreatic cancer risk factor (4, 7). Under the hypothesis that NSAIDs are inversely associated with pancreatic cancer risk, it may be that null results for non-aspirin NSAIDs observed here and in previous reports may reflect some level of negative confounding.

We observed stronger associations for NSAIDs and aspirin among women who reported diabetes at baseline. Diabetes is an inflammatory condition and pancreatic cancer risk factor characterized by elevated levels of cytokines, acute-phase proteins, and immune cell activation (34, 35). Given that the finding is based upon small numbers of pancreatic cancer cases with diabetes, it is possible that it is spurious. Nevertheless, it is supported by a prior study. Although the authors observed no association for aspirin overall, in the pooled analysis of the NHS and HPFS cohorts, Khalaf and colleagues (16) reported that aspirin was associated with a 29% (RR 0.71, 95% CI, 0.54–0.94) reduction in pancreatic cancer incidence among participants of both sexes with prevalent diabetes.

Aspirin and NSAIDs are thought to exert chemopreventive effects through irreversible (aspirin) or reversible (NSAIDs) bonds to COX enzymes, thereby inhibiting the inflammatory cascade. COX-2, the inducible isoform, is expressed in a high proportion of pancreatic cancers as well as adjacent “normal” tissues (36). The enzyme and its sequelae are overexpressed in chronic pancreatitis and among people with diabetes and obesity (37–39). Administration of COX-2 selective inhibitors have shown improvements for severe acute pancreatitis in recent trials (40). A biological reason for differential associations of aspirin and non-aspirin NSAIDs with pancreatic cancer risk is poorly understood but may be explained by aspirin’s various COX-independent anticancer effects (41).

This study has several important strengths, including its large sample size of postmenopausal women, prospective design, control for comprehensive confounding factors, lengthy follow-up, and its resultant large number of pancreatic cancer cases. The chief limitation was that measurement of medications in the WHI was limited to those used in a 2-week period prior to baseline or follow-up. We aimed to abrogate potential non-differential misclassification by combining baseline and year 3 medication data, as has been done previously (22). Nevertheless, we observed statistically significant associations between aspirin and pancreatic cancer even in the presence of non-differential misclassification errors that may have a bias towards the null. We were unable to examine associations restricted to more common histologies due to the absence of histology data in 43% of pancreatic cancer cases, with over a quarter of cases identified via death records. As noted above, another limitation is that prevalent pancreatitis was not measured in the WHI. The potential for confounding is likely greater for analyses of non-aspirin NSAIDs, which are more typically used to address pancreatitis-related pain, than for aspirin formulations, and may partly explain the observed null results for these formulations. Last, there were small numbers of pancreatic cancer cases in some strata of analyses, resulting in imprecise estimates of association.

This study adds to a growing literature from prospective observational studies and randomized trials that use of aspirin may provide moderate benefit for pancreatic cancer prevention. If confirmed, strong consideration of the potential tradeoff between cancer chemoprevention and the risk of gastrointestinal bleeding from long-term aspirin administration is advised (42). Additional large prospective studies with careful measurement of prevalent pancreatitis, and NSAID type, frequency, and dose are needed to further investigate the possibility of stronger benefit among individuals diagnosed with diabetes.

L.R. Jager reports grants from NIH/NHLBI during the conduct of the study. K.L. Margolis reports grants from NIH/NHLBI during the conduct of the study. No disclosures were reported by the other authors.

T.M. Brasky: Conceptualization, formal analysis, supervision, investigation, methodology, writing–original draft, writing–review and editing, AACR Membership Number: 128783. L.R. Jager: Conceptualization, formal analysis, investigation, methodology, writing–review and editing. A.M. Newton: Resources, investigation, writing–review and editing. X. Li: Conceptualization, investigation, writing–review and editing. H.A. Loomans-Kropp: Investigation, methodology, writing–review and editing. J.L. Hays: Investigation, writing–review and editing. K.L. Margolis: Conceptualization, supervision, writing–review and editing. J. Luo: Conceptualization, supervision, methodology, writing–review and editing.

The WHI program is funded by the National Heart, Lung, and Blood Institute, National Institutes of Health, U.S. Department of Health and Human Services through contracts 75N92021D00001 (G. Anderson), 75N92021D00002 (J. Wactawski-Wende), 75N92021D00003 (R. Jackson), and 75N92021D00004 (M. Stefanick), 75N92021D00005 (M. Vitolins). The funding source had no role in the collection of data, the analysis, interpretation, or writing of this report, or the decision to submit the article for publication. A short list of Women’s Health Initiative Investigators: Program Office: (National Heart, Lung, and Blood Institute, Bethesda, Maryland) Jacques Rossouw, Shari Ludlam, Joan McGowan, Leslie Ford, and Nancy Geller. Clinical Coordinating Center: (Fred Hutchinson Cancer Research Center, Seattle, WA) Garnet Anderson, Ross Prentice, Andrea LaCroix, and Charles Kooperberg. Investigators and Academic Centers: (Brigham and Women’s Hospital, Harvard Medical School, Boston, MA) JoAnn E. Manson; (MedStar Health Research Institute/Howard University, Washington, DC) Barbara V. Howard; (Stanford Prevention Research Center, Stanford, CA) Marcia L. Stefanick; (The Ohio State University, Columbus, OH) Rebecca Jackson; (University of Arizona, Tucson/Phoenix, AZ) Cynthia A.Thomson; (University at Buffalo, Buffalo, NY) Jean Wactawski-Wende; (University of Florida, Gainesville/Jacksonville, FL) Marian Limacher; (University of Iowa, Iowa City/Davenport, IA) Jennifer Robinson; (University of Pittsburgh, Pittsburgh, PA) Lewis Kuller; (Wake Forest University School of Medicine, Winston-Salem, NC) Sally Shumaker; (University of Nevada, Reno, NV) Robert Brunner. Women’s Health Initiative Memory Study: (Wake Forest University School of Medicine, Winston-Salem, NC) Mark Espeland.