The rise and decline of nonsteroidal antiinflammatory drug–associated gastropathy in rheumatoid arthritis




Nonsteroidal antiinflammatory drug (NSAID)–associated gastropathy is a major cause of hospitalization and death. This study was undertaken to examine whether recent preventive approaches have been associated with a declining incidence of NSAID gastropathy, and, if so, what measures may have caused the decline.


We studied 5,598 patients with rheumatoid arthritis (RA) over 31,262 patient-years at 8 sites. We obtained standardized longitudinal information on the patients that had been previously used to establish the incidence of NSAID gastropathy, and also information on patient risk factors and differences in toxicity between NSAIDs. Consecutive patients were followed up with biannual Health Assessment Questionnaires and medical record audits between 1981 and 2000. The major outcome measure was the annual rate of hospitalization involving bleeding, obstruction, or perforation of the gastrointestinal (GI) tract and related conditions.


Rates of GI-related hospitalizations rose from 0.6% in 1981 to 1.5% in 1992 (P < 0.001), and then declined to 0.5% in 2000 (P < 0.001). The fitted spline curve fit the data well (R2 = 0.70). The period of rise was mainly associated with increasing patient age and the GI risk propensity score. The period of decline was associated with lower doses of ibuprofen and aspirin, a decline in the use of “more toxic” NSAIDs from 52% to 42% of patients, a rise in the use of “safer” NSAIDs from 19% to 48% of patients, and increasing use of proton-pump inhibitors, but not with change in age, NSAID exposure, or GI risk propensity score.


The risk of serious NSAID gastropathy has declined by 67% in these cohorts since 1992. We estimate that 24% of this decline was the result of lower doses of NSAIDs, while 18% was associated with the use of proton-pump inhibitors and 14% with the use of less toxic NSAIDs. These declines in the incidence of NSAID gastropathy are likely to continue.

Nonsteroidal antiinflammatory drug (NSAID)–associated gastropathy, which is associated with gastrointestinal (GI) pathology involving hemorrhage, perforation, or obstruction of the GI tract and related conditions, is the most prevalent of serious drug side effects, resulting in an estimated 100,000 hospitalizations annually in the US (1). Although it was originally considered to be rare, the incidence of NSAID gastropathy is recognized to be ∼1.2–1.6% per year in patients with rheumatoid arthritis (RA), a relatively high-risk population. The relative risk of developing serious GI complications in patients exposed to NSAIDs is 5–6 times that in those not exposed (2). The present status of the epidemic (whether the incidence is rising, falling, or constant) is not known. We hypothesized that recent preventive approaches to NSAID gastropathy should have led to a reduction in its incidence.

ARAMIS (Arthritis, Rheumatism, and Aging Medical Information System) investigators have been following up consecutively enrolled RA patients from multiple settings since 1980, using systematic outcome assessment protocols (see the ARAMIS Web site at We follow up patients longitudinally, audit hospitalizations, and collect detailed data on drug use and adverse reactions (3, 4). In the present study, we sought to examine trends in the incidence of NSAID gastropathy over time in this cohort. We also sought to examine associations of NSAID gastropathy with changes in potentially explanatory variables, which were prespecified as age, sex, GI risk propensity score, NSAID dose, use of specific NSAIDs, use of misoprostol, proton-pump inhibitors, and H2 receptor antagonists, percentage of patients at each data bank center, and frequency of NSAID exposure.



We studied 5,598 RA patients whose data were drawn from 8 ARAMIS data banks in the US and Canada (Stanford, California [2 data banks], Santa Clara County, California, Wichita, Kansas, Saskatoon, Saskatchewan, Canada, Cincinnati, Ohio, Baltimore, Maryland, and Pittsburgh, Pennsylvania), comprising 31,262 patient-years of observation. These centers are population-based (Saskatoon and Santa Clara County), university clinics with quite different characteristics (Stanford, Cincinnati, Baltimore, and Pittsburgh), and private practice–based (Wichita), and thus are believed to be reasonably representative of RA patients in the US and Canada. ARAMIS is a prospective, observational data bank system in which patients are enrolled consecutively under institutional review board approval, followed up for life, and assessed prospectively by standard protocol for multiple variables, including demographics, clinical and hospital outcomes, mortality, and the frequency, dose, effectiveness, and toxicity of specific treatments (5). The RA cohort reported herein was the subject of early reports establishing incidence rates, risk factor models, and the relative toxicity of NSAIDs in gastropathy, and thus provides a unique perspective from which to evaluate trends over time (1, 2).


Patients in the ARAMIS are asked to complete the Health Assessment Questionnaire (HAQ) every 6 months. The HAQ is a widely used, comprehensive, patient self-reported outcome assessment instrument (6, 7). Patients report on medication use, dosage, duration of use, and side effects. Information on both prescription and nonprescription aspirin and NSAID use is included. The protocol requires followup of nonresponders, patient contact for missing information, acquisition of patient records for hospitalizations, and stringent quality control. Data from 1981 to 2000, during which time prospective protocols were in place and remained constant, were analyzed.

Ascertainment of serious GI events.

The primary outcomes were the number and incidence of serious GI events that required hospitalization. Patients with bleeding of the upper or lower GI tract, clinically symptomatic gastritis, ulcers, gastric outlet obstructions, and GI symptoms serious enough to require hospitalization (abdominal pain, nausea, vomiting, or diarrhea) were identified from medical records. GI-related hospitalizations were first determined by patient self-report of hospitalization. Hospital records were then audited by trained reviewers for identification of International Classification of Diseases, Ninth Revision, Clinical Modification diagnostic codes 531–535, 787, and 789. Attributions were validated by a physician who was blinded to each patient's medication use.

Statistical analysis.

The percentage of patients with GI-related hospitalizations (number of hospitalizations per 100 patient-years) was calculated as the total number of incident GI-associated hospitalizations in a given year, divided by the number of RA patients by year, and allowing only 1 hospitalization per patient per year. Since there was a period of rising incidence followed by a period of declining incidence, we used second-degree spline regression to determine the transition inflection point.

We identified patients receiving each NSAID for every year at every site, including both prescription and over-the-counter use, and computed the percentage of patients receiving each particular NSAID among all NSAID users. We prespecified NSAIDs as relatively more toxic or relatively safer based on our own data (8), on the meta-analysis by Henry et al (9), and on the CLASS and VIGOR (Celebrex Long-Term Arthritis Safety and Vioxx Gastrointestinal Outcomes Research) studies (10, 11). Time trends also allowed classification of NSAIDs into those with increased use over time and those with decreased use over time. The percentage of patients receiving gastroprotective agents was calculated as the proportion of patients being treated with H2 receptor antagonists by prescription or over-the-counter, misoprostol, sucralfate, and/or proton-pump inhibitors, divided by the number of patients per year.

We also examined trends in average age, disease duration, disability, the percentage of patients receiving NSAIDs, the percentage of patients with increased use or decreased use of NSAIDs, the percentage of patients receiving prednisone, and the GI risk propensity score. GI risk propensity scores were computed using the method of Singh et al (12), based on age, disability, global health, prior GI-related hospitalization, prednisone use, and prior NSAID-related GI side effect. This score represents the likelihood of a patient having a serious GI complication in the following year. Statistics were computed using SAS for Windows, version 8.2 (Chicago, IL).


The major finding on the rates of hospitalization for NSAID gastropathy is apparent in Figure 1. By conservatively using the fitted spline regression data, rather than the raw values for each year, we found that the rate of GI-related hospitalizations first increased from 0.6% per year in 1981 to a peak of 1.5% in 1992, and then decreased to 0.5% in 2000. The rate initially nearly tripled, then declined by 67% from its high point in 1992. The spline regression indicated a concave function that peaked in 1992. We therefore defined the period of rise as 1981–1992 (the first period) and the period of decline as 1993–2000 (the second period).

Figure 1.

Percentage of rheumatoid arthritis patients with serious gastrointestinal (GI) events that required hospitalization, over time. The dotted line is a second-degree spline-fitted curve for these data.

Trends over the period of rise and over the period of decline each were highly statistically significant (P < 0.001), as was the fitted spline regression for these trends (P < 0.001, R2 = 0.70). GI-related hospitalizations as a percentage of all hospitalizations declined from 7.3% in 1981, to 5.7% in 1992, and to 4.8% in 2000. When considering only those patients who were exposed to NSAIDs, a curve that was higher than, but of similar shape as, that in Figure 1 was found, with a beginning rate of GI-related hospitalizations of 0.5%, a peak of 2.1% in 1992, and the final value of 0.5%. Results were consistent across the 8 data bank centers.

Table 1 summarizes the trends in some potentially explanatory variables. The average age of the patients in the cohort increased by nearly 6 years from 1981 to 1992, from 56.7 years to 62.2 years, and then remained relatively constant thereafter. Composite GI risk propensity scores followed a pattern of rise through 1992 similar to that of the GI-related hospitalization rates, followed by a plateau. The percentage of patients receiving NSAIDs declined from 87% to 76% by 1992, and thereafter remained fairly constant. Average HAQ disability index scores, on a 0–3 scale (7), rose slightly in the first period and improved considerably in the second period, which is consistent with recent findings of declining disability in RA associated with more aggressive use of disease-modifying medications over the past 2 decades (13). Average pain scores, on a 0–3 visual analog scale, and average patient global health assessment scores, on a 0–100 scale, similarly rose and then fell, but differences were slight. Average disease duration rose during the first period and subsequently declined slightly. Prednisone use (average dosage ∼7 mg/day throughout) increased in the second period; this would have been expected to increase the rates of serious GI events during the period in which they were observed to decline.

Table 1. Characteristics of the study population by year*
YearNo.Mean age, years% femaleMean HAQ DI scoreMean duration of RA, years% taking NSAIDs% taking NSAIDs with less toxicity% taking NSAIDs with greater toxicity% taking prednisoneMean GI risk propensity score
  • *

    HAQ DI = Health Assessment Questionnaire disability index; RA = rheumatoid arthritis; NSAIDs = nonsteroidal antiinflammatory drugs; GI = gastrointestinal.

  • See Figure 2.

  • See Figure 3.

Figure 2.

Percentage of use of those drugs with increased usage over time (“less toxic” drugs), from among the total nonsteroidal antiinflammatory drug (NSAID) use.

Figure 3.

Percentage of use of drugs with decreased usage over time (“more toxic” drugs), from among the total nonsteroidal antiinflammatory drug (NSAID) use. The aspirin data exclude use of cardioprotective doses of aspirin.

We also examined changes in drug dosage over time. Ibuprofen and noncardiac aspirin dosages each declined by ∼40% in the second period and thus correlated with the decline in GI complications; other NSAID dosages did not appreciably change. The percentage of patients taking low-dose (≤325 mg per day) aspirin increased from 3% in 1992 to 8% in 2000, while the average low-dose cardiac aspirin dosage declined from ∼200 mg per day to ∼150 mg per day. We did not find increasing doses for any of the drugs in either period, and thus “dose creep” was eliminated as a possible cause of the rise in NSAID gastropathy.

The proportion of use of each specific NSAID in these cohorts was also calculated for each year. Figure 2 shows trends for use of the drugs that were prespecified as less toxic. The largest increase was in the use of ibuprofen (which showed, at the same time, a decline in dose) and in the use of the more specific cyclooxygenase 1–sparing drugs rofecoxib and celecoxib over the last 2 years of the study. The 5 NSAIDs that showed an increase in use are all drugs that have been reported to be among the least toxic of the NSAIDs (8–11). During our entire observation period, patients taking these drugs had a rate of serious GI events of 0.54% per year.

Figure 3 shows the time trends in market share for the 8 drugs that were prespecified as more toxic. All of these had reduced usage over the entire period. Aspirin lost the largest share, although the greatest declines were in the usage of piroxicam, sulindac, and indomethacin. Each of these 8 drugs has been reported to have a higher-than-average toxic effect on the GI tract (9, 10). Over our observation period, patients taking these drugs had a rate of serious GI events of 0.78% per year.

Time trends in the use of gastroprotective agents are shown in Figure 4. Two striking trends were evident: a rapid rise in the use of H2 receptor antagonists in the first period, and a rapid rise in the use of proton-pump inhibitors in the second period, which ended with a 16% use of proton-pump inihitors. Misoprostol (∼5%) and sucralfate were not widely used. Data on identification or treatment of Helicobacter pylori infection were not reliably reported in the charts reviewed, but treatment for H pylori appeared to be rare.

Figure 4.

Percentage of rheumatoid arthritis (RA) patients taking gastroprotective agents over time. H2 antagonists (H2) include cimetidine, ranitidine, and famotipine. Proton-pump inhibitors (PPI) include omeprazole and lansoprozole. Misoprostol (Miso) is a synthetic prostaglandin, including misoprostol alone and in combination with diclofenac.


The prevalence of serious GI events associated with NSAID use in this large, longitudinal, multicenter study of RA patients rose markedly from 0.6% of patients per year in 1981 to 1.5% per year in 1992, and then fell to 0.5% per year in 2000. Both the rise and the decline were steady, were consistent, and have plausible explanations. The decline in NSAID-related GI problems, if generalizable, represents a substantial improvement in the health of the public. These improvements are likely to be the result of the joint efforts of epidemiologists, gastroenterologists, rheumatologists, regulatory authorities, and industry.

A number of relevant events over the past decades may have directly influenced these trends (14). We have taken note of a few of the earliest reports and those that may have had some of the greatest impacts. The earliest hints of NSAID-related GI problems are found in the observations of Douthwaite and Lintott in 1938 (15), and these observations were repeated in a more modern era by Sun et al in 1974 (16). Levy noted the association between aspirin use and major GI bleeding in 1974 (17). The high prevalence of NSAID-associated endoscopic erosions and ulcers led Roth and Bennett to coin the term NSAID gastropathy in 1987 and to suggest that it might be quite common (18). The potentially life-threatening nature of these problems was suggested by Armstrong and Blower in 1987 (19). In 1988, Griffin and colleagues noted the association of NSAID use and mortality linked to peptic ulcer disease in the elderly (20). In 1989, our group, using the patient cohorts of this current report, began to quantitate the epidemiology of NSAID gastropathy, finding a GI-related hospitalization incidence of 1.6% per year in RA patients and a relative risk of 5.2 (21); these data are approximated in the present study by the average values in the period of rising incidence.

In 1991, our group extended these observations and identified age and other variables that were associated with increased risk of NSAID gastropathy (2). Griffin and colleagues also observed the strong association with age in the Tennessee Medicaid data set in 1991 (22). We also analyzed differences in toxicity among the NSAIDs in 1991, finding up to 4-fold differences (8) between each NSAID despite a prevailing prior view that there were no such differences (23). By 1996, Henry et al were able to identify 12 sets of studies consistently showing differential toxicity among the NSAIDs, with a similar magnitude of differences and with similar ordering (9).

These findings, even before they were published and disseminated, began to affect the development of new treatments. Roth and colleagues began, in 1987, to explore the uses of H2 antagonists in the prevention of NSAID gastropathy (24). Results were disappointing when regular doses were used, but prevention was subsequently documented with double doses of H2 antagonists (25). Misoprostol, a synthetic prostaglandin analog, was studied by Graham et al in 1988, and those authors showed that endoscopic ulcers decreased following this therapy (26). Decreases in GI complications of ∼40% with the use of misoprostol were shown by Silverstein et al in 1995 (27). A meta-analysis of prevention trials by Koch et al in 1996 suggested that the newer proton-pump inhibitors would be more effective preventive agents (28), and this was confirmed by Yeomans et al in 1998 (29). Treatment of H pylori infection has been found to be as effective as proton-pump inhibitors in some studies (30–32), although the incidence of this infection appears to be decreasing.

New NSAIDs were approved and came into broad use during the period of declining incidence of NSAID gastropathy in this study, and all of these agents were promoted as being safer than their predecessors. They included nabumetone and etodolac in 1991, Arthrotec (which combined diclofenac and misoprostol) in 1997, celecoxib in 1998, and rofecoxib in 1999 (30). Studies have suggested that the cyclooxygenase 1–sparing agents celecoxib and rofecoxib have toxic effects on the GI tract that are only half that of the traditional NSAIDs (10, 11, 33, 34). Even newer agents have since entered the market but were not available during the present study period. A change in treatment strategy for RA patients from one based on NSAIDs to one based on disease-modifying antirheumatic drugs (DMARDs) occurred over this period of observation (35–37) and likely explains the decline in NSAID usage during the first period.

The period of rising frequency of NSAID gastropathy was associated with a strong trend toward increasing age with NSAID use in these cohorts. Age has consistently been identified as the most important risk factor for NSAID gastropathy, other than prior GI-related hospitalization, and acts exponentially as a risk factor (2, 22). The average age of RA patients in these cohorts rose by 5.5 years from 1981 to 1992, which is sufficient by itself to account for a near doubling of the incidence of serious GI events. The percentage of NSAID-receiving patients older than age 75 years rose from 4% to 14% over this same period. There also was a smaller trend toward more use of NSAIDs in women, and an increase in average disability levels, in prednisone use, and in the overall GI risk propensity score.

Nevertheless, the percentage of RA patients taking NSAIDs in these cohorts was steadily decreasing, which should have acted to lower the risks; this decline in NSAID use was likely to be related to the increasing emphasis on use of DMARDs to control disease activity in RA (35–37). Although use of H2 antagonists increased during the period of rise, these agents have been found to be of little value, in regular doses, when used as prophylaxis against serious NSAID gastropathy (28). The frequency of prednisone use, another risk factor, increased during the period of rising incidence. Therefore, some originally plausible explanatory factors, such as sex or increasing NSAID exposure, do not appear to have played a role in the development of the epidemic. Increases in patient age, prednisone use, and the GI risk propensity score are the most likely explanations for the period of rising incidence.

In the period of declining frequency of serious GI complications, the frequency of NSAID use, average age, and GI risk propensity score all reached a plateau, and therefore could not have made major contributions to the decline. Disability levels declined somewhat, possibly suggesting that the patients were sturdier. Impressively, there was a striking shift in the use of specific NSAIDs, with a decrease in use of the more toxic drugs and an increase in use of those with lesser toxicity. There was a 40% decrease in the dosage of both aspirin and ibuprofen, both of which have a high frequency of use. Moreover, a steady stream of new drugs entered the market, including nabumetone and etodolac, and the cyclooxygenase 1–sparing drugs rofecoxib and celecoxib achieved large market shares toward the end of our observation period.

Which of these various trends may have contributed to the decline in serious GI events? The answers are intrinsically speculative, and therefore we discuss them here, rather than presenting them as results. There are 3 particularly suggestive associations. First, we observed a shift away from the use of NSAIDs with an incidence of serious toxicity in the GI tract of 0.78% per year toward NSAIDs with an average incidence of 0.54%, yielding a blended toxicity incidence rate of 0.72% in 1992 and of 0.65% in 2000, which is sufficient to explain a reduction of ∼14% of the overall 67% decline in the incidence of serious GI events (a decline of 9% of total events).

Second, use of proton-pump inhibitors rose from none in 1992 to 16% in 2000. Assuming a preventive benefit of 50% from the use of proton-pump inhibitors, their use would reduce the incidence of serious GI complications by 8%. Assuming a 40% protection rate from the use of misoprostol, with its lower frequency of use, the incidence rates would be reduced by an additional 2%. After accounting for the increased GI risk propensity scores, from a mean of 14 to a mean of 16, among those patients taking proton-pump inhibitors (8), the estimate of the effects of gastroprotection sums to ∼12% overall, and 18% of the observed decline.

Third, conservatively assuming a linear dose-response curve for NSAID complications, the 40% reduction in dose for aspirin and ibuprofen, which made up 40% of the NSAID market share, would account for a reduction of 16%, or 24% of the observed reduction. Thus, we estimate that 24% of the observed decline was due to lower doses of some NSAIDs, 18% was due to use of proton-pump inhibitors, and 14% was due to use of safer NSAIDs. We therefore have suggestive evidence to explain about two-thirds (66%) of the observed decline.

What factors may have contributed to the remainder of the decline? These are more speculative because of limitations in our data. It could be explained, in part, by uncaptured aspects of the 3 dominant forces. For example, within the “more toxic NSAID” category, were there greater decreases in use of the most toxic drugs? Were proton-pump inhibitors used almost exclusively in the highest-risk patients, which would increase effects beyond what we projected? Were dose reductions in the most toxic, but less frequently used, agents, for which we did not have sufficient data, important? Interaction terms, which go beyond our data, could have made material contributions. Moreover, our explanations are not entirely mutually exclusive, so that we might have overestimated some contributions. What were the effects, if any, of increasing use of endoscopy or generally declining hospitalization rates? As disability levels in RA decline as a result of increasing DMARD use, do RA patients become more resistant to GI complications? There are many other potential contributions to the decline that we could not measure.

For example, we do not have data on H pylori prevalence over time. Treatment of H pylori infection or reduction in prevalence of H pylori could have had an effect. We tend to discount a major effect of H pylori, however, since its prevalence in our cohort is likely to have been low (US and Canadian patients with access to care) and the relationship of H pylori to serious NSAID-related GI events itself remains controversial. Overall, it appears to be reasonable to assume that most of the decline in serious NSAID effects came from lower doses of safer drugs in conjunction with greater use of gastroprotection.

National effects upon the epidemic of NSAID gastropathy are less encouraging. The national market for NSAIDs (in millions of tablets/capsules) rose from 2,836 in 1981 to 4,242 in 1993 and to 6,305 in 2000 (Fort J: personal communication). Thus, even if the results in our RA cohorts can be generalized, much of the decline in incidence will have been countered by the increases in exposure, and the estimate of more than 100,000 hospitalizations annually in the US (2) may still be relatively current. Continued introduction and marketing of newer NSAIDs, even though they are safer, may increase the frequency of exposures to NSAIDs in the future. Migration of products from prescription to over-the-counter use may also increase exposures, although this may be offset by lower doses. New guidelines for use of low-dose aspirin for cardioprotection in essentially the same population as that which uses the most NSAIDs will act to limit declines in the incidence of serious GI effects (38); low-dose aspirin use itself, even while protecting the heart, increases the relative risk of these events by a factor of ∼2 (39).

How, then, may we continue and even accelerate these declines? At least 5 complementary approaches may be recommended, building on the immediate past (40, 41). First, we recommend even more selective use of NSAIDs, particularly in high-risk patients. Much of the NSAID use is still for simple analgesia in noninflammatory conditions. Second, introduction of newer and safer agents should be continued, although, for the reasons mentioned above, a limit may soon be reached. Third, we recommend aggressive, continued migration of the NSAID market share from more toxic to less toxic NSAIDs. This migration so far appears relatively slight. Fourth, use of proton-pump inhibitors for prophylaxis should be substantially increased, particularly in high-risk patients. These drugs appear to be greatly underutilized and are becoming available in generic versions. However, if H2 antagonists are used for prevention, the dose must be twice the standard doses, to avoid promulgation of a false sense of security. Systematic diagnosis and treatment of H pylori infection may have a role, although these findings remain controversial. Fifth, the lowest effective dose should be used for any NSAID.

Developments in reducing the incidence of NSAID gastropathy to date are gratifying. As long as trends continue toward the use of safer NSAIDs, more frequent use of proton-pump inhibitors, and de-escalation of dosages, particularly in high-risk patients, the trend toward reduction in the incidence of NSAID gastropathy is likely to continue.