Chemoprevention of gastrointestinal malignancies with nonsteroidal antiinflammatory drugs

A review


  • Nicholas J. Shaheen M.D., M.P.H.,

    Corresponding author
    1. Division of Digestive Diseases and Nutrition and the Center For Gastrointestinal Biology and Disease, the University of North Carolina at Chapel Hill, Chapel Hill, North Carolina
    • Division of Digestive Diseases and Nutrition and the Center For Gastrointestinal Biology and Disease, CB#7080, University of North Carolina at Chapel Hill, Chapel Hill, NC 27599-7080
    Search for more papers by this author
    • Fax: (919) 966-6842

    • Drs. Shaheen and Sandler have served as consultants for Merck and Company.

    • Dr. Straw currently is employed by Merck and Company.

  • Walter L. Straus M.D., M.P.H.,

    1. Outcomes Research and Management, Merck and Company, West Point, Pennsylvania
    Search for more papers by this author
  • Robert S. Sandler M.D., M.P.H.

    1. Division of Digestive Diseases and Nutrition and the Center For Gastrointestinal Biology and Disease, the University of North Carolina at Chapel Hill, Chapel Hill, North Carolina
    Search for more papers by this author
    • Drs. Shaheen and Sandler have served as consultants for Merck and Company.


In multiple studies, the chronic use of nonsteroidal antiinflammatory drugs (NSAIDs) has been associated with a decreased incidence of several types of gastrointestinal (GI) neoplasia. This effect may be mediated by one of several intracellular mechanisms, some of which involve the inhibition of the cyclooxygenase-2 (COX-2) isoenzyme. In multiple studies of colorectal carcinoma, chronic NSAID use has shown a protective effect, with the majority of studies demonstrating a 30–70% risk reduction associated with NSAID use. The effect of NSAIDs on other types of GI neoplasia is less clear, but evidence suggests that chronic NSAID use may diminish the risk of esophageal and gastric carcinomas. Data assessing the effects of NSAIDs on the incidence of pancreatic and hepatic malignancies currently are too sparse and inconsistent to draw any conclusions. The newer COX-2 specific agents may provide a less GI-toxic alternative to nonselective NSAIDs as chemoprotective agents. Cancer 2002;94:950–63. © 2002 American Cancer Society.

DOI 10.1002/cncr.10333

In the U.S., gastrointestinal (GI) carcinoma accounts for approximately one in five malignancies. In addition, one in four cancer deaths will be secondary to tumors arising from the GI tract. Five of the top 10 fatal malignancies among men, and 3 of the top 10 among women, are GI malignancies.1

Given the terrible toll of these diseases, a primary focus has been placed on prevention. One area that has gathered considerable interest is chemoprevention of gastrointestinal malignancies using nonsteroidal antiinflammatory drugs (NSAIDs). These agents, including both older nonselective inhibitors of cyclooxygenase (COX)-1 and COX-2, as well as newer specific inhibitors of COX-2, have shown promise in averting GI malignancy in several different organs. The evidence is strongest for colorectal carcinoma, but to our knowledge there are data to support a preventive role for these drugs in other GI malignancies.

The current review will assess the evidence supporting a chemopreventive effect of NSAIDs and aspirin on GI neoplasia. After a brief discussion of the possible mechanisms of such a chemopreventive effect, we will review site-specific data regarding chemoprevention of each of the major GI malignancies. In those situations in which in vivo animal data and human data are available, we will concentrate our discussion on these studies. For malignancies for which human data are to our knowledge not yet available, we will review research that has been performed on cell culture models. Finally, we will speculate with regard to possible topics for future areas of research.

NSAIDS and Chemoprevention: Possible Mechanisms

Several potential mechanisms have been proposed to explain a possible chemopreventive effect of NSAIDs in cancer. NSAIDs inhibit COX enzymes, which are the rate-limiting step in the production of prostaglandins (PGs) from arachidonic acid (Fig. 1).2 Two isoforms of COX have been identified. Although they share approximately 60% homology, these isoforms are believed to have different cellular functions. COX-1 is expressed constitutively in the majority of tissues and is believed to be involved in cell regulation and “housekeeping” functions such as platelet aggregation and regulation of blood flow. Conversely, COX-2 is an inducible enzyme that is present in increased amounts in inflammatory conditions.3–5 COX-2 activity is induced by cell mediators of inflammation such as interleukins and cytokines, and by bacterial endotoxins. NSAIDs have been postulated to inhibit carcinogenesis by both COX-dependent and COX-independent mechanisms.6, 7

Figure 1.

Arachidonic acid pathway. Arachidonic acid may be metabolized by either of two isoforms of the enzyme cyclooxygenase (COX). When metabolized by the COX-1 isoform, the resulting prostaglandins are active in a variety of “housekeeping” functions, including the maintenance of the gastric mucosa and hemostasis. When metabolized by the COX-2 isoform, the resulting prostaglandins are active in a number of proinflammatory functions. Increased COX-2 activity is observed in a number of gastrointestinal malignancies. The COX-2 isoform is inducible and may be blocked selectively by newer nonsteroidal antiinflammatory drugs.

COX-dependent neoplastic effects

There is evidence to suggest that increased COX-2 activity may contribute to carcinogenesis. Cellular levels of COX-2 are elevated in a number of GI malignancies, including colorectal carcinoma,8, 9 gastric carcinoma,10–12 and esophageal carcinoma.13 Furthermore, transfection of colon carcinoma cells with a vector expressing COX-2 was found to enhance the metastatic potential of these cells in vivo.14 This effect was reversed by the administration of sulindac sulfide.14 In addition, increased expression of COX-2 in a transformed rat epithelial cell line was found to decrease the rate of cell apoptosis.15 COX-2 overexpression has been associated with increased levels of angiogenesis,16–19 and the inhibition of COX-2 is reported to inhibit tumor angiogenesis.19–21 Therefore, a primary effect of COX-2 inhibition on tumor growth may be to starve the tumor's blood supply.

The mechanisms by which COX-2 activity may impact carcinogenesis are unclear. Because COX-2 mediates the expression of PGs, a PG-dependent mechanism has been postulated.5 PGs are known to play a role in multiple cell-regulatory functions, including DNA synthesis and cell division.22 PG expression has been found to be increased in a number of experimental tumors, as well as in sporadically occurring large bowel neoplasms in humans.22 Certainly, the inhibition of PG production might retard DNA production and cell proliferation, thereby decreasing carcinogenesis.23, 24 The mechanism(s) by which PGs promote tumor growth is unclear, and the intracellular signaling pathways activated by PGs have not been well described to date.

Other data suggest that the inhibition of COX-2 may inhibit carcinogenesis by mechanisms not involving PGs. Because the inhibition of COX-2 leads to decreased production of PGs, cellular arachidonic acid, the metabolic precursor of PG, may be increased. This increased pool of arachidonic acid may be shunted down other metabolic pathways, such that there is increased production of other metabolites that retard cellular division.25 In addition, COX enzymes participate in other cellular reactions beyond the production of PGs. For instance, COX enzymes can generate carcinogen-DNA adducts. The production of these adducts is inhibited by NSAIDs.26

Non-COX-related mechanisms

NSAIDs also may affect carcinogenesis by mechanisms not involving the inhibition of COX. Sulindac sulfone, a sulindac metabolite that does not inhibit COX, has been shown to prevent colorectal, lung, and mammary gland carcinoma in animal models.27–29 Moreover, sulindac sulfone promoted apoptosis in a manner similar to the parent compound.30 In addition, in vitro work suggests that NSAIDs retard growth in tumor cell lines that do not produce the COX enzyme.26, 31 Although preliminary, these data suggest that NSAIDs may have an effect on carcinogenesis that is independent of COX inhibition. The cellular mechanisms behind such an effect are not known.

Chemoprevention of Esophageal Carcinoma

Esophageal carcinoma will be responsible for approximately 12,500 deaths in the U.S. this year.1 Two major histologic cell types of this malignancy occur: squamous cell carcinoma and adenocarcinoma. Although the incidence of squamous cell carcinoma has remained relatively static, the incidence of esophageal adenocarcinoma has increased dramatically for reasons that to our knowledge remain unclear.32, 33 This year, approximately equal numbers of cases of squamous cell carcinoma and adenocarcinoma of the esophagus will occur. Because the 5-year mortality rates for both these types of esophageal carcinoma exceed 85%,1 effective chemoprevention may be especially useful in these patients.

The effect of NSAIDs on esophageal neoplasia is promising but preliminary. Both animal and human data suggest that NSAIDs may provide chemoprotection against esophageal neoplasia. However, these data are somewhat conflicted. The study of chemoprevention of esophageal carcinoma is hampered by the relatively low incidence of these tumors. In addition, the effect of the chemopreventive agent may vary by the histologic type of the carcinoma.

Animal studies

Animal studies of chemoprevention in esophageal carcinoma involve the application of a cancer-inducing agent, such as N-nitrosomethylbenzylamine (NMBA) or diethylnitrosamine (DENA), with subsequent treatment with a second agent either during or after tumor induction. The degree to which the second agent moderates the disease caused by the carcinogen is considered to be a measure of its potential antineoplastic effect (Fig. 2).

Figure 2.

Schematic of rodent chemoprevention studies. This diagram depicts usual rodent cancer chemoprevention studies utilizing nonsteroidal antiinflammatory drugs (NSAIDs). Healthy animals are injected with a carcinogen. At some interval after this injection, the animals are randomized to receive either the NSAID in a suspension vehicle or the suspension vehicle alone. If the NSAID is chemopreventive, the rodents in the treatment group subsequently will develop fewer and smaller tumors than those in the control group. This schematic depicts the NSAID being given after the carcinogen. Other protocols may call for pretreatment with the NSAID or the simultaneous administration of the NSAID and carcinogen.

Using this model, several NSAIDs have been studied in experimentally-induced esophageal tumors in rodent models. Results have been conflicting. Rats with NMBA-induced esophageal carcinoma did not demonstrate a decrease in the number or size of their tumors in response to sulindac given before, during, or after disease induction.34 Similarly, rats that were administered flurbiprofen, a potent nonselective COX inhibitor, during experimentally induced carcinogenesis did not appear to experience a decrease in their tumor burden. Conversely, mice with DENA-induced esophageal tumors did demonstrate a significant reduction in the number of esophageal tumors when treated concurrently or after induction with indomethacin.35, 36

Human data

Because of the low incidence of esophageal carcinoma, a prospective randomized study to demonstrate the efficacy of NSAIDs in preventing malignancy would likely be both very time-consuming and prohibitively costly. Therefore, all attempts to elucidate a possible chemopreventive effect for NSAIDs have involved either a cohort or case–control design. The data are somewhat conflicting, but on balance appear to support a possible protective effect of NSAIDs. A challenge in these types of studies is avoiding confounding by indication, a situation that occurs when the outcome event (cancer) actually provokes the exposure to the putative etiologic agent (NSAIDs). For example, exposure to NSAIDs may prevent esophageal carcinoma. However, NSAIDs may well appear to cause cancer if those patients with early esophageal carcinoma begin receiving NSAIDs to treat the pain associated with the disease. Table 1 displays the case–control and cohort studies assessing the effect of NSAID usage on esophageal carcinoma.

Table 1. Effect of Aspirin and NSAIDs on Esophageal Carcinoma: Case–Control and Cohort Studiesa
ReferenceStudy typeSubjectsDrugResultsComments
OR or RR (95% CI)Group
  • NSAIDs: nonsteroidal antiinflammatory agents; OR: odds ratio; RR: relative risk; 95% CI: 95% confidence interval; ACS: American Cancer Society; NHANES I: National Health and Nutrition Examination Survey I; ca: carcinoma; adeno: adenocarcinoma.

  • a

    See text for further details.

Gridley et al., 199340Cohort11,683 rheumatoid arthritis patientsAspirin and NSAID (presumed)1.32 (0.7–2.4)All participantsNo explicit quantification of NSAID exposure; it was presumed because of the patient population
Thun et al., 199341Cohort635,031 U.S. residents participating in an ACS studyAspirin0.69 (0.39–1.19) 0.83 (0.51–1.36) 0.78 (0.42–1.44)Occasional 1–15 times/mo > 16 times/moUsed deaths, as opposed to incident cases, to define disease status
Funkhouser and Sharp, 199537Cohort13,300 U.S. residents followed in NHANES IAspirin0.21 (0.04–0.96) 0.00 (−2.89)Occasional users Regular usersOnly 15 esophageal carcinoma cases total in the study
Garidou et al., 199642Case–control97 carcinoma cases from Tumor Registry 388 hospital controlsAnalgesics0.64 (0.21–1.93) 0.52 (0.17–1.62)Squamous cell ca AdenoAmount of usage not quantified; may include non-NSAID users
Farrow et al., 199838Case–control486 carcinoma cases (279 adenocarcinomas and 207 squamous cell carcinomas) and 687 controls; population-basedAspirin and NSAIDs0.37 (0.24–0.58)   0.49 (0.28–0.87)   0.57 (0.32–1.02)   0.33 (0.13–0.87)Current aspirin user, adeno Current aspirin user, squamous cell ca Current NSAID user, adeno Current NSAID user, squamous cell caGrouped tumors by histologic type; less impressive ORs for occasional users
Coogan et al., 200039Case–control215 esophageal carcinoma cases 5,952 controlsAspirin and NSAID0.8 (0.5–1.4) 0.9 (0.6–1.2)At least 4 days/week. < 4 days/weekNo effect for discontinued users

Funkhouser and Sharp performed a cohort analysis of 13,300 U.S. residents followed for 12–16 years as part of the National Health and Nutrition Examination Survey I (NHANES I).37 They found that occasional aspirin use was associated with a 90% decreased risk of developing esophageal carcinoma, and that no regular user of aspirin developed the disease. However, despite the large number of patient-years analyzed in their study, only 15 cases of esophageal carcinoma were reported to occur in this cohort during follow-up. In addition, the investigators were unable to quantitate the usage of aspirin, and no data were available concerning other NSAIDs. The histologic tumor type was not reported.

Farrow et al. performed a population-based case–control study of the effect of NSAIDs on the risk of esophageal carcinoma.38 Using cases reported by 3 different tumor registries compared with controls recruited using random-digit dialing and Health Care Financing Administration (HCFA) registries, they found that current users of aspirin were at a decreased risk for both adenocarcinoma (odds ratio [OR] = 0.37; 95% confidence interval [95% CI], 0.24–0.58) as well as squamous cell carcinoma of the esophagus (OR = 0.49; 95% CI, 0.28–0.87). The use of other NSAIDs also was associated with a reduced risk of esophageal carcinoma, but this association did not reach statistical significance. A dose-response relation was noted for squamous cell carcinoma, with those patients who were receiving the most aspirin found to be at the lowest risk for developing the carcinoma; no such relation was determined for adenocarcinoma.

In contrast, other data have failed to demonstrate a protective effect of NSAIDs in esophageal carcinoma. In a hospital-based case–control study of 215 subjects with esophageal carcinoma, no significant relation was found between either discontinued or continuing NSAID use and malignancy.39 A cohort study of Swedish rheumatoid arthritis patients presumably using NSAIDs actually demonstrated a statistically insignificant increase in the risk of esophageal carcinoma.40 A second large observational study by Thun et al.41 demonstrated a protective trend with increasing aspirin use in patients with esophageal carcinoma, but these data did not reach statistical significance (P = 0.54 for trend). Finally, a case–control study from Greece demonstrated an insignificant protective effect for chronic analgesic use for both squamous cell carcinoma and adenocarcinoma of the esophagus.42 It should be noted that several of these studies, because of the small number of cases occurring, may have had inadequate power with which to detect a statistically significant protective effect.

A current area of active debate is the potential role of NSAIDs as therapy for Barrett esophagus. Because Barrett esophagus is believed to be a precursor lesion to esophageal adenocarcinoma, it is plausible that intervening with a chemoprotective agent in patients with Barrett esophagus may retard progression to esophageal adenocarcinoma.43 Studies assessing the effects of NSAIDs in subjects with Barrett esophagus currently are ongoing. Another area of ongoing investigation is the degree to which COX-2 specific agents may inhibit carcinogenesis. Because elevated COX-2 levels have been demonstrated in esophageal carcinoma by both enzyme-linked immunoassay as well as Western blot analysis,13 it has been postulated that the inhibition of COX-2 specifically may account for any antineoplastic effect of NSAIDs.

Gastric Carcinoma

Approximately 22,000 new cases of gastric carcinoma will occur in the U.S. this year.1 For reasons that are unclear, the incidence of this malignancy is declining markedly in the U.S. Although rates are declining, the clinical course of the disease remains grim, with a 5-year mortality risk of 79%.1 The primary histologic type of gastric carcinoma is adenocarcinoma. Unless otherwise specified, all data presented pertain to this histologic type.

Animal studies

Similar to studies involving the esophagus, animal data regarding chemoprevention of gastric carcinoma with NSAIDs involve rodent models with chemically induced neoplasms. Again, results with respect to the effect of NSAIDs on these experimentally induced tumors are mixed. In one study of mice induced to develop gastric neoplasia by the tobacco specific carcinogen NNK, the administration of sulindac and ibuprofen led to a decrease in tumor size and number, whereas the administration of piroxicam did not.44 In a second study, experimentally induced gastric tumors in a rat model actually were greater in number in those animals treated concurrently with flurbiprofen than in controls.45

Human observational studies

Observational studies assessing the effect of NSAIDs on the incidence of gastric carcinoma have been hampered by the low and declining incidence of this malignancy in Western populations. For this reason, data again were limited to case–control or cohort studies (Table 2).

Table 2. Effect of Aspirin and NSAIDs on Gastric Carcinoma: Case–Control and Cohort Studiesa
ReferenceStudy typeSubjectsDrugResultsComments
OR or RR (95% CI)Group
  • NSAIDs: nonsteroidal antiinflammatory drugs; OR: odds ratio; RR: relative risk; 95% CI: 95% confidence interval; ACS: American Cancer Society.

  • a

    See text for further details.

Gridley et al., 199340Cohort11,683 rheumatoid arthritis patientsAspirin and NSAIDs (presumed)0.63 (0.5–0.9)All participantsNo explicit quantification of NSAID exposure; it was presumed because of the patient population
Thun et al., 199341Cohort635,031 U.S. residents participating in an ACS studyAspirin0.74 (0.51–1.07)OccasionalUsed deaths, as opposed to incident cases, to define disease status
0.58 (0.32–1.03)1–15 times/mo
0.49 (0.22–1.12)> 16 times/mo
Farrow et al., 199838Case–control612 carcinoma cases (255 adenocarcinomas of the cardia and 357 adenocarcinomas elsewhere in the stomach) and 687 controls; population-basedAspirin and NSAIDs0.80 (0.54–1.19)Current aspirin user, cardiaClassification of tumors at gastric cardia as being of esophageal versus gastric origin is difficult
0.46 (0.31–0.68)Current aspirin user, elsewhere
0.69 (0.39–1.22)Current NSAID user, cardia
0.23 (0.11–0.34)Current NSAID user, elsewhere
Coogan et al., 200039Case–control254 gastric carcinoma cases 5,952 controlsAspirin and NSAIDs0.3 (0.1–0.6) 0.7 (0.5–0.9)At least 4 days/week < 4 days/weekInsignificant effect for discontinued users

Several case–control and cohort studies of NSAID use in gastric carcinoma have demonstrated a chemoprotective effect of NSAIDs. Coogan et al.39 found that regular NSAID use (at least 4 days a week for > 3 months) reduced the risk of gastric carcinoma in a hospital-based case–control study of 254 patients (OR = 0.3; 95% CI, 0.1–0.6). The protective effect was more pronounced among those patients using NSAIDs continually for >5 years (OR = 0.2; 95% CI, 0.1–0.7) than for those using NSAIDs for < 5 years (OR = 0.4; 95% CI, 0.1–0.9). However, these data were based on eight users of NSAIDs among the case group, seven of whom used aspirin. In a large cohort study of 635,031 participants followed over 6 years, the American Cancer Society demonstrated that regular exposure to aspirin exerted a protective effect against gastric carcinoma.41 In this study, those patients reporting aspirin use ≥ 16 times a month were found to have approximately 50% the risk of gastric carcinoma compared with nonusers (OR = 0.53; 95% CI, 0.34–0.81). The previously mentioned study of Scandinavian patients with rheumatoid arthritis and a presumable increase in NSAID use demonstrated that gastric carcinoma risk was decreased compared with similar patients without arthritis in the same geographic area (standardized incidence ratio [SIR] = 0.63; 95% CI, 0.5–0.9).40

Conclusions: NSAIDS, Malignancy, and the Upper GI Tract

Although the data regarding a potential chemoprotective effect of NSAIDs in malignancies of the upper GI tract are less convincing and less numerous than those with respect to colorectal carcinoma, a growing body of evidence suggests that NSAIDs may have some beneficial effect. Although the results from animal studies are conflicting, questions regarding the applicability of these models to humans limit their utility. Human studies, although again not universally supportive of a beneficial effect, are for the most part suggestive of chemoprevention. In contrast to colonic neoplasia, in which adenocarcinoma is the only histologic type of significance, the study of esophageal carcinoma is complicated by the presence of two important histologic types: squamous cell carcinoma and adenocarcinoma. Because these histologic types do not share the same epidemiology and may not share a similar pathogenesis, future studies should consider these entities separately when possible.

Liver Carcinoma and NSAID Chemoprevention

Tumors of the liver and intrahepatic bile duct will occur in approximately 16,000 Americans this year. These tumors usually are unresectable and incurable, and approximately 14,000 deaths can be expected.1 Although malignant neoplasms of multiple cell types are found in the liver, hepatocellular carcinomas (HCCs) comprise the majority of primary liver malignancies. Data pertaining to the effect of NSAIDs on the incidence of liver carcinoma are few and conflicting.

Ex vivo data

Ex vivo studies assessing the effect of sulindac on human HCC cell lines have been performed. These studies have demonstrated a marked time and dose-dependent negative effect on cell division and viability.46 A study assessing the effect of salicylic acid on rats with DENA-induced hepatoma demonstrated no change in the frequency of metastases.47 In addition, data are beginning to appear suggesting a role for COX isoenzymes in HCC. Using histologic material obtained from 29 patients with HCC, Shiota et al.48 measured COX-2 expression in the tumor. They then compared the expression of COX-2 in tumor tissue with adjacent nonmalignant tissue. The authors found that specimens obtained from 7 of 10 patients with well differentiated HCC had more intense immunochemical staining for COX-2 in the tumor compared with normal tissue. However, this finding was not consistent, and patients with moderately differentiated (3 of 10 patients) or poorly differentiated (3 of 9 patients) tumors demonstrated no greater COX-2 expression in tumor compared with normal tissue. Koga et al.49 studied HCC tissues obtained from 44 patients with HCC and compared them with 7 surgically resected normal tissue samples as controls. They measured the expression of COX-1 and COX-2 by both anti-COX polyclonal antibody and by Western blot analysis. The authors observed that well differentiated HCC had significantly higher COX-2 expression compared with poorly differentiated or moderately differentiated HCC (P < 0.001). The expression of COX-1 was greater in well differentiated HCC compared with normal control tissue; however, there were no significant differences with regard to expression between other grades of HCC and normal tissue. The authors suggest that COX-2 up-regulation may be involved in HCC carcinogenesis.

Human observational studies

In an interview-based case–control study, 51 patients with liver carcinoma were compared with hospital control patients (persons treated for trauma or an infectious disease) with regard to prior NSAID exposure. NSAID exposure was not found to be associated with a reduced risk of liver carcinoma (OR = 0.9; 95% CI, 0.3–2.9), even after stratifying based on the intensity of NSAID exposure.39 Conversely, in the Scandinavian study of rheumatoid arthritis patients, Gridley et al. noted that the number of observed cases of liver carcinoma (4 cases reported to have occurred in a group of 11,683 subjects who were followed for an average of 8.6 years) was less than that expected based on adjusted population rates (SIR = 0.35; 95% CI, 0.1–0.9, adjusted for gender, age, site, and calendar year).40 Clearly, these data are limited and do not provide us with the ability to identify and adjust for numerous potential confounding variables.

Carcinoma of the Pancreas

Pancreatic carcinoma most commonly arises from the ductal epithelium of the pancreas. Over 29,000 incident cases of this malignancy will occur in the U.S. this year, and approximately 29,000 individuals will die. To our knowledge little progress has been made in improving the mortality rate of this disease, and 5-year survival rates are reported to be < 5%.1 Like hepatic malignancies, little is known with respect to the chemoprevention of pancreatic carcinoma.

Animal and ex vivo studies

Pancreatic carcinoma may be induced in rodent models by administering N-nitrosobis(2-oxopropyl)amine. Takahashi et al.50 treated hamsters with this agent and then subsequently treated them with indomethacin, phenylbutazone, or aspirin. Animals in the indomethacin and phenylbutazone groups displayed fewer tumors than the control group. Aspirin also showed a tendency to decrease the incidence of this malignancy, but this effect did not reach statistical significance. The authors concluded that these agents reduce the development of pancreatic carcinoma in this model in the postinitiation period. A second group of researchers assessed the effect of salicylate on human pancreatic carcinoma cell lines.51 Cells treated with salicylate demonstrated significantly inhibited cell proliferation, and less progression from the G1 phase to the S-phase.

As is the case with liver carcinoma, there are histologic and tissue culture data to suggest a potential role for COX-2 in the pathogenesis of pancreatic carcinoma. Compared with normal adjacent tissue, pancreatic tumors exhibit increased COX-2 activity. Furthermore, COX-2 expression is restricted to tumor tissue, and was not present in surrounding stromal tissue or infiltrating inflammatory cells.52 In an elegant series of experiments, Yip-Schneider et al.53 demonstrated that COX-2 is up-regulated in primary pancreatic adenocarcinoma and that several NSAIDs inhibit growth in multiple pancreatic carcinoma cell lines. These investigators treated two experimental pancreatic carcinoma cell lines, one positive for COX-2 expression (BxPC-3) and one negative for COX-2 (PaCa-2), with indomethacin and sulindac (both nonselective inhibitors of cyclooxygenase) and with NS-398, a selective COX-2 inhibitor. After 3 days of therapy, each drug inhibited the growth of both cell lines in a dose-dependent manner; however, two of the drugs (indomethacin and NS-398) demonstrated greater inhibition of cell growth in the COX-2-positive cell line BxPC-3 compared with the line that did not express COX-2 (PaCa-2). Finally, Molina et al.54 employed similar techniques to assess the activity of COX-2 in pancreatic tissue and to explore the effects of NSAIDs. In 14 of 21 specimens of pancreatic adenocarcinomas of ductal origin, COX-2 was detectable and was observed in the cytoplasm of the tumor cells. Three of five experimental pancreatic cell lines also demonstrated COX-2 expression, with levels found to correlate with the degree of tumor differentiation. The highest levels were found with the BxPC-3 (moderately well) and Capan1 (well differentiated) cell lines. They then determined whether sulindac sulfide and NS398 affected the tumor cell proliferation of these lines over 4 days of exposure. Sulindac sulfide exhibited a dose-dependent inhibition of growth of all five carcinoma cell lines, with a marked reduction of viable cells noted at the highest dosages administered. The authors found their results to be consistent with an apoptotic effect of NSAIDs.

These data suggest that COX-2 may be an essential component in the pathogenesis of pancreatic carcinoma. In addition, the inhibition of COX-2 may decrease the rate of neoplastic proliferation.

Human observational data

There are few available epidemiologic studies published to date that examine the relation between NSAID exposure and the risk of pancreatic carcinoma. In the study of Swedish rheumatoid arthritis patients noted earlier,40 no significant decreased rate of pancreatic tumors was noted among the rheumatoid arthritis patients compared with the number that would be expected (SIR = 0.83; 95% CI, 0.6–1.2). Similarly, Coogan et al.39 did not find NSAID exposure to be protective against pancreatic carcinoma, but it is worth noting that this case–control study also demonstrated a trend toward a protective effect (OR = 0.8; 95% CI, 0.5–1.1). In contrast, Langman et al.55 compared NSAID exposure among British general practice patients diagnosed with various malignancies with that of control patients who were matched for age, gender, and practice. These investigators found that pancreatic carcinoma patients tended to have had greater exposure to NSAIDs during a period of up to 36 months compared with control patients (P = 0.08), and that there was a dose-response relation favoring NSAIDs as a risk factor for pancreatic carcinoma. The authors suggest that this apparent association may be spurious, because pancreatic carcinoma may present insidiously, and patients may receive NSAID therapy for vague abdominal symptoms prior to the diagnosis of carcinoma.

Conclusions: NSAIDS, Carcinoma, and the Liver and Pancreas

Data with respect to the chemoprevention of carcinomas of the liver and pancreas currently are sparse and unconvincing. Although data from cell culture and animal models provide both a rationale for chemoprevention and some support for a potential antineoplastic effect, the applicability and clinical relevance of these data are unproven. Human studies are hampered by small numbers, methodologic concerns, and inconsistent results.

Chemoprevention of Large Bowel Neoplasia

There is a considerable body of evidence to suggest that aspirin and the nonaspirin NSAIDs decrease the risk of colorectal neoplasia. The data from animal studies, from human observational studies, and from trials in patients with polyposis appear consistent in demonstrating this effect. In fact, the effect is stronger than that for any of the other factors that have been linked with risk of colorectal neoplasia.

Animal studies

Animal studies generally involve the administration of chemical carcinogens such as azoxymethane or dimethylhyrazine to rodents. The animals predictably develop colonic adenomas and tumors. Investigators then can test whether candidate chemopreventive agents decrease the number or the size of tumors compared with placebo. Another model involves genetically altered rodents. The Min (Multiple Intestinal Neoplasm) mouse, which has a mutation of the murine homologue of the human APC mutation, has a phenotype similar to familial adenomatous polyposis (FAP) in humans. These mice spontaneously develop intestinal tumors, thereby providing an excellent experimental model system for chemopreventive drugs. It also is possible to develop mice that have mutations in both the APC and COX genes to examine the role of COX in the development of intestinal tumors. From studies of tumor development in these rodents, one then can make inferences regarding the effects of drugs that block COX selectively or nonselectively.

Animal studies over the past 20 years consistently have shown that NSAIDs decrease the number and size of intestinal tumors. In these studies, a number of different NSAIDs have been shown to decrease tumor development in rodents.56–62 It is interesting to note that, in animal studies, the protective effect was present as late as 9–13 weeks after the administration of the carcinogen.57, 63 This finding suggests that the drugs were active during both tumor initiation and tumor promotion. However, when the drug was discontinued, the occurrence of tumors increased to equal that in the untreated animals, indicating that initiated cells persisted.59

Aberrant crypts are aggregates of single to multiple colonic crypts that demonstrate dysplastic features. Aberrant crypts, which can be induced by chemical carcinogens, are believed to be the earliest detectable pathologic lesions in the development of colon carcinoma development.64 Both piroxicam and aspirin have been shown to decrease the development of aberrant crypt foci.64

Studies in Min mice have shown that aspirin can prevent the development of spontaneous intestinal adenomas.65 High-dose aspirin was found to reduce tumor multiplicity and size significantly. Other NSAIDs, including piroxicam and sulindac, have demonstrated a similar effect.66, 67 Oshima et al.68 demonstrated that treating the Apc delta716 knockout mouse model of adenomatous polyposis with COX-2 inhibitors markedly decreased the number of tumors, thereby providing direct evidence of the importance of COX-2 in intestinal tumorigenesis.

Sulindac sulfone, a metabolite of the prodrug sulindac, also was reported to reduce tumors in chemical carcinogenesis models.27 Sulindac sulfone does not block COX, suggesting that non-COX mechanisms also may play a role.

Human observational studies

Case–control studies.

Case–control studies compare the use of aspirin or NSAIDs in patients with colorectal carcinoma with the use of similar drugs in carefully selected controls. Case–control studies are susceptible to various forms of bias because they are conducted retrospectively. The choice of controls also may distort the results. Nonetheless, the rapid speed and the lower cost make case–control studies attractive. There have been at least 12 case–control studies of NSAIDs and colorectal carcinoma, all of which demonstrated a protective effect (Table 3).69–80

Table 3. Effect of Aspirin and NSAIDs on Colon Carcinoma and Adenoma: Case–Control Studiesa
OR (95% CI)Group
  • NSAIDs: nonsteroidal antiinflammatory drugs; OR: odds ratio; 95% CI: 95% confidence interval; ca: carcinoma; FOBT: fecal occult blood test.

  • a

    See text for further details.

Collet et al., 1999693844 cases and 15,373 controlsNSAIDs0.57 (0.36–0.89) 0.26 (0.11–0.61)Colon ca Rectal caBased on dispensing rates using beneficiaries of the Saskatchewan Prescription Drug Plan
Garcia-Rodriguez and Huerta-Alvarez, 2001712005 cases and 10,000 controls drawn from U.K. General Practice Research DatabaseAspirin and NSAIDs0.60 (0.4–0.9),300 mg aspirinExposure based on prescriptions for aspirin or NSAIDs
0.50 (0.4–0.7) NSAIDs
Greenberg et al., 199372793 patients enrolled in multicenter trial of vitamin supplementsAspirin0.52 (0.31–0.89)Consistent usersSimilar effects in men and women
Kune et al., 198873715 ca cases and 727 controls; population-basedAspirin and aspirin-containing drugs0.53 (0.40–0.71)Colon and rectum; men and women; decreased risk with arthritis; adjusted for diet
NSAIDs0.77 (0.60–1.01)
La Vecchia et al., 1997701357 ca cases and 1891 controlsAspirin0.70 (0.5–1.0)Controlled for education, physical activity, and selected dietary factors
Logan et al., 199374147 adenoma casesAspirin, any use0.57 (0.3–1.0)No effect for acetaminophen; controls had negative FOBT
153 controlsNSAID, any use0.50 (0.3–1.0)
Martinez et al., 199578157 adenoma casesAspirin and NSAID0.77 (0.39–1.55)WeeklyNo effect observed for duration of use
0.36 (0.20–0.63)Daily
Muscat et al., 199475511 colon caNSAID > 3×/week0.64 (0.42–0.97)MenNo effect for acetaminophen
500 controls0.32 (0.18–0.57)Women
Peleg et al., 19947697 ca cases from Tumor Registry 388 hospital controlsAspirin prescribed in past 4 years0.32 (0.12–0.84) 0.08 (0.01–0.59)208–624 aspirin days > 624 aspirin daysBased on computerized pharmacy records pharmacy records that were not validated; no effect for acetaminophen:
no effect for acetaminophen
Rosenberg et al., 1991771326 hospital ca cases 4891 hospital controlsNSAID0.50 (0.4–0.8)NSAID ≥ 4 times/weekColon and rectum; men and women; lower risk with longer use; no effect in former users
Sandler et al., 199880210 adenoma casesAspirin and NSAIDs0.56 (0.34–0.92)Similar effect aspirin and NSAIDs
169 controls
Suh et al., 199379830 colorectal ca casesAspirin daily0.55 (02.5–1.20)Trend toward decreasing risk with greater use
524 hospital controls

The first report of a protective effect of aspirin against colorectal carcinoma in humans came from a study by Kune et al. in Melbourne, Australia.73 Surprisingly, and somewhat unexpectedly, individuals who used aspirin or aspirin-containing medications regularly were approximately 50% as likely to develop colorectal carcinoma (OR = 0.53; 95% CI, 0.40–0.71). The protective effect was observed in men and women, and in patients with carcinomas of the colon and the rectum.

The aspirin hypothesis began to receive much more attention after the publication of a large hospital-based study in which regular NSAID users (mostly aspirin users) were reported to be approximately 50% as likely to develop colorectal carcinoma as nonusers.77 The effect was observed for men and women, for both colon and rectal carcinoma, and for all age groups. It is interesting to note that subjects who had stopped taking NSAIDs > 1 year prior to the time of the interview did not appear to have a decreased risk, suggesting that the drug may have delayed rather than prevented tumor development (i.e., chemoprocrastination rather than chemoprevention).

Subsequent studies by different investigators, in different countries, using different types of cases and controls have found the same decreased risk of large bowel neoplasia. The effect of the NSAIDs is specific; acetaminophen, a drug used for similar indications, demonstrated no effect.74 Some of the better studies adjust for possibly confounding effects such as diet with similar results.80

Cohort studies.

Because of their prospective direction, cohort studies are less susceptible to bias than case–control studies. There are at least seven cohort studies published to date that have evaluated the effect of aspirin or NSAIDs on colorectal carcinoma (Table 4).40, 81–86 All but one of these studies86 has shown a strong protective effect of NSAIDs.

Table 4. Effect of Aspirin and NSAIDs on Colon Carcinoma and Polyps: Cohort Studiesa
OR (95% CI)Group
  • NSAIDs: nonsteroidal antiinflammatory drugs; RR: relative risk; 95% CI: 95% confidence interval; ca: carcinoma; NR: no relative risk.

  • a

    See text for further details.

Giovannucci et al., 19948347,900 male health professionalsAspirin ≥ 2×/week0.68 (0.52–0.92)Total colorectal caConsistent users had lower risks
Giovannucci et al., 199581121,701 female registered nursesAspirin > 2×/week0.56 (0.36–0.90)20 years of consistent useDecreasing risk with longer duration of use
Gridley et al., 19934011,683 patients with rheumatoid arthritisNot known0.63 (0.5–0.9) 0.72 (0.5–1.1)Colon ca Rectal caAverage follow-up of 8.6 years; similar effects for men and women; no information on drug dose
Paganini-Hill et al., 19899013,723 men and women without colon ca at baselineAspirin daily1.46 (NR)MenFollow-up of 8.5 years
1.01 (NR)Women
Schreinamachers and Everson, 19948512,668 subjects followed an average of 12.4 yrsAspirin past 30 days0.35 (0.17–0.73) 0.78 (0.38–1.59)Men age < 65 yrs Women age < 65No effect after age 65 yrs; 59% of subjects were classified as exposed greatest effect seen at 7–10 years follow-up;
Smalley et al., 199982104,217 subjects enrolled in Tennessee Medicaid ProgramNSAIDS > 12 months0.61 (0.48–0.77)Recent usersProtective effect shared by most NSAIDs
Thun et al., 19918466,424 adult volunteers followed for 6 yrsAspirin0.60 (0.40–0.89)Men ≥ 16 times/moRectal excluded; whites only; acetominophen did not decrease risk
0.58 (0.37–0.90)Women, ≥ 16 times/mo

The first cohort study to demonstrate a protective effect of aspirin was reported by Thun et al.84 More than 600,000 men and women were followed for 6 years. There was a substantially lower risk of colon carcinoma for both men (relative risk [RR] = 0.60; 95% CI, 0.40–0.89) and women (RR = 0.58; 95% CI, 0.37–0.90) who took aspirin > 15 times per month. Acetaminophen did not appear to decrease the risk. These findings persisted when the authors controlled for body mass index, physical activity, family history, and diet. Rectal carcinoma was not studied.

A large prospective study of nurses demonstrated that women who consistently took > 2 aspirin tablets per week for 20 years had a substantial reduction in risk of colorectal carcinoma (RR = 0.56; 95% CI, 0.36–0.90).81 There was slight reduction in risk after aspirin use for 10 years, but this finding did not reach statistical significance. Controlling for risk factors for colorectal carcinoma did not appear to affect the results. A similar study in 47,900 male health professionals demonstrated that regular aspirin use decreased the incidence of colorectal carcinoma by > 30%.83

Human experimental studies.

Although the observational studies provide consistent evidence of a protective effect of aspirin and NSAIDs, there remains concern that some other aspect in the patients who use aspirin might be responsible for the observed effect. Assigning subjects to receive either drug or placebo at random reduces concerns regarding possible bias.

A randomized controlled trial of low-dose aspirin in 22,000 male physicians in the U.S. that was designed to evaluate the effect of aspirin on cardiovascular disease provided an opportunity to evaluate possible colorectal carcinoma chemoprevention. Follow-up over 12 years demonstrated no difference with regard to the incidence of colorectal carcinoma between the aspirin and placebo groups.87 The negative results could be due to the low dose of aspirin used in the trial (300 mg every other day) or to the short duration of the study. Alternatively, other characteristics associated with the use of aspirin in the observational studies might explain the discrepancy.

Another experimental strategy to test the effect of drugs on the development of colorectal carcinoma would be to conduct a randomized controlled trial in patients with FAP. In patients with FAP, the colonic mucosa is covered by adenomatous polyps. These polyps are premalignant. Indeed, unless patients with FAP undergo a colectomy, they virtually all develop colorectal carcinoma. A drug that decreases the number and size of adenomas would have the potential to prevent carcinoma. Giardiello et al.88 administered sulindac, 150 mg twice a day, or placebo in a randomized controlled trial of 22 FAP patients. The treated group had a 44% decrease in the number of polyps and a 35% decrease in the greatest dimension of the polyps. No patient achieved complete resolution of polyps, and the number and the size of the polyps increased in sulindac-treated patients within 3 months after treatment was withdrawn.

Steinbach et al.89 studied the effect of celecoxib, a selective COX-2 inhibitor, on colorectal polyps in 77 patients with FAP. After 6 months, the patients who received 400 mg of celecoxib twice a day had a 28.0% reduction in the mean number of polyps (P = 0.003) and a 30.7% reduction in polyp size (P = 0.001) compared with reductions of 4.5% and 4.9%, respectively, in the placebo group. The reductions in the group who received 100 mg of celecoxib twice a day were not significantly different from placebo. The drug recently was approved by the U.S. Food and Drug Administration as an adjunct treatment for FAP patients.

Conclusions: NSAIDS, Carcinoma, and the Colon

There is a large body of literature that supports a protective effect of NSAIDs against colorectal neoplasia. Animal and mechanistic studies indicate that this effect is biologically plausible. In observational studies, the regular use of NSAIDs appears to decrease the development of neoplasia by 50%. Randomized controlled trials have reduced concerns regarding biases that could distort the conclusions of observational studies, although the conclusions from studies of FAP patients cannot necessarily be generalized to sporadic tumors occurring in low-risk populations.

Although the protective effect is presumed to be due to the inhibition of PGs, studies of sulindac sulfone27 have suggested that non-PG mechanisms may be at work as well. The studies regarding celecoxib raise the question of whether the effect is due to the COX-2 isoenzyme.

Although the data are persuasive, they should not lead to public health recommendations for the use of these drugs in average-risk populations.80 First, we are not certain of the appropriate dose and duration. Second, we are not completely certain of the mechanism of action, although this is a less important concern. Finally, NSAIDs have well known adverse effects that might exceed their benefits in a low-risk population. Perhaps the selective COX-2 inhibitors will change the risk-benefit equation. Studies of these drugs currently are ongoing. Until the results become available, physicians should encourage healthy lifestyle choices and screening for the prevention of colorectal carcinoma.

Future Directions

Although a potential chemoprotective role for NSAIDs in GI malignancies has been postulated for many years, much remains unknown regarding the effect of these agents. The mechanism(s) of the effect are obscure, and may vary from agent to agent and tumor to tumor. For the majority of the neoplasms reviewed in the current study, the magnitude of the chemoprotective effect, if one exists, is unclear. Animal models, although helpful in discerning a possible effect of NSAIDs, often are performed with pharmacologic doses that are much higher than those achievable clinically. Human observational data often suffer from inadequate sample size to avoid a Type II statistical error, especially for rarer malignancies.

Given the potential side effects of NSAID administration, it is unclear whether subjects truly would derive benefit from the administration of NSAIDs even if a chemoprotective effect was proven to be present and strong. In addition, because these malignancies usually occur in the elderly, the effect of competing causes of mortality may be such that the incremental gain in life expectancy in those individuals who have a malignancy averted still may be modest.

Because of reduced GI toxicity, COX-2 inhibitors may provide a more attractive chemopreventive option compared with nonselective NSAIDs. Clinical trials with well developed, prespecified endpoints will be necessary to assess the value of these agents in averting malignancy. It is important to note that, because of the high cost of these new agents, cost-effectiveness analyses must be undertaken to optimize the allocation of resources.

Given the incidence and mortality associated with the disease, we believe that chemoprevention of sporadic colon carcinoma is a worthy goal. Controlled data assessing the effects of both nonspecific NSAIDs as well as COX-2 specific NSAIDs are essential. Another area of future research is the application of chemoprevention to those subjects known to be at an increased risk of GI malignancies. In addition to the data regarding FAP noted earlier, individuals with Barrett esophagus and intestinal metaplasia of the stomach also may be candidates for chemoprevention. It is less likely that the wide-scale use of any chemopreventive agent to avert sporadic GI malignancies other than colon carcinoma will be cost-effective because of the low incidence of these tumors. Data regarding the chemoprevention of these tumors will likely remain observational, but should be considered as important secondary outcomes for studies assessing other effects of NSAIDs.