Gut 60:1372-1378 doi:10.1136/gut.2010.229906
  • Colon

Non-steroidal anti-inflammatory drugs and risk of lower gastrointestinal adverse events: a nationwide study in Taiwan

  1. Mei-Shu Lai1
  1. 1Institute of Preventive Medicine, College of Public Health, National Taiwan University, Taipei, Taiwan
  2. 2Department of Internal Medicine, National Taiwan University Hospital, Taipei, Taiwan
  3. 3Cardiovascular Center, National Taiwan University Hospital Yun-Lin Branch, Dou-Liou City, Yun-Lin, Taiwan
  4. 4National Health Insurance Mediation Committee, Department of Health, Executive Yuan, Taipei, Taiwan
  5. 5Division of Health Technology Assessment, Center for Drug Evaluation, Taipei, Taiwan
  1. Correspondence to Professor Mei-Shu Lai, Institute of Preventive Medicine, College of Public Health, National Taiwan University, Taipei, Taiwan; mslai{at}
  • Revised 22 January 2011
  • Accepted 11 February 2011
  • Published Online First 17 March 2011


Objective Only limited studies have evaluated the risk of non-selective non-steroidal anti-inflammatory drugs (nsNSAIDs) and coxibs for lower gastrointestinal (GI) adverse outcomes. The objective of this study was to evaluate risks of lower GI adverse events associated with use of celecoxib, oral and parenteral nsNSAIDs.

Design Retrospective case–crossover study.

Setting Records of all patients aged ≥20 years hospitalised for lower GI adverse events (bleeding from small or large intestine, perforation, and complicated diverticular disease) in 2006 were retrieved using ICD-9-CM diagnosis codes from inpatient claims from the Taiwan National Health Insurance database.

Interventions Case periods were defined for each patient as 1–30 days prior to hospital admission date and control period as 91–120 days prior to hospital admission date. The pharmacy prescription database was searched for NSAID use during case and control periods.

Main outcome measures We calculated adjusted self-matched ORs and 95% CIs with a conditional logistic regression model to determine the associations between NSAID use and lower GI adverse outcomes.

Results A total of 1297 patients hospitalised for lower GI adverse events were included. Celecoxib was associated with an adjusted OR of 2.33 (95% CI 0.97 to 5.59); the association became statistically significant (OR: 3.26, 95% CI 1.07 to 9.91) when a different control period (31–60 days) was applied. Both oral and parenteral nsNSAIDs significantly increased risk for lower GI adverse events (OR: 2.26, 95% CI 1.78 to 2.85 and OR: 5.64, 95% CI 3.24 to 9.82, respectively).

Conclusions Oral and parenteral NSAIDs were associated with significantly increased risk for lower GI adverse events. Celecoxib also increased risk to a comparable extent, despite risk estimates being affected slightly by the control period chosen for comparison. The association of NSAIDs with specific lower GI adverse events and long-term complications requires further investigation.

Significance of this study

What is already known about this subject?

  • The use of non-steroidal anti-inflammatory drugs (NSAIDs) is an important cause of upper gastrointestinal (GI) adverse events.

  • NSAID-induced GI toxicity also involves the lower GI tract.

  • Celecoxib, a cyclooxygenase-2 enzyme selective NSAID, is also associated with an increased risk of upper GI adverse outcomes.

  • Few studies have evaluated the lower GI safety of cyclooxygenase-2 inhibitors compared to non-selective NSAIDs.

What are the new findings?

  • Non-selective NSAIDs are associated with an increased risk for lower GI toxicity.

  • Parenteral NSAIDs are associated with a significantly increased risk of lower GI toxicity.

  • Although the current analysis did not perform a head-to-head comparison between celecoxib and other non-selective NSAIDs, the risk for lower GI adverse events appeared to be comparable.

How might it impact clinical practice in the foreseeable future?

  • Cyclooxygenase-2 inhibitors, oral and parenteral NSAIDs all increase risk of lower GI adverse events. Physicians should treat patients with NSAIDs at the appropriate dosage, for the required durations, be vigilant with regard to the GI toxicity of NSAIDs, and not assume that the superior safety of cyclooxygenase-2 inhibitors reduces the need for close monitoring.


Use of non-steroidal anti-inflammatory drugs (NSAIDs) is an important cause of upper gastrointestinal (GI) adverse events.1 2 Meanwhile, increasing evidence indicates that NSAID-induced GI toxicity also involves the lower GI tract.3 Lesions induced de novo, or pre-existing conditions aggravated by NSAIDs, may manifest clinically as bleeding of the small or large intestine, perforation, or complicated diverticular disease.4 Previous reports suggest that among NSAID users, lower GI events may account for 13% to 40% of all GI events.5 During the past decade, studies have reported a clear decreasing trend in upper GI events and a significant increase in lower GI events in Western countries.6 7 However; lower GI adverse events associated with NSAID use are less clearly characterised than upper GI effects in the Asian population.1 8–10

Compared to studies of upper GI adverse events, fewer studies have evaluated the lower GI safety profiles of both cyclooxygenase-2 enzyme non-selective and selective NSAIDs (cyclooxygenase-2 inhibitors and nsNSAIDs, respectively).11–14 Controversies exist concerning the frequency and severity of a variety of adverse effects with the use of nsNSAIDs and cyclooxygenase-2 inhibitors.15–21 Additionally, although evidence suggests a marked variability in terms of upper GI safety among traditional NSAIDs, information about the effects of individual NSAIDs on lower GI adverse events is not available.14 22 Therefore, the aim of this nationwide study was to examine the risks of lower GI adverse events associated with use of individual oral and parenteral NSAIDs, including celecoxib, among the general population in Taiwan.


The protocol of this study was approved by the National Taiwan University Hospital Research Ethics Committee. There were concerns that observational studies using healthcare claims data may not provide valid estimates of the effects associated with NSAIDs use as most of the GI risk factors such as cigarette smoking, alcohol consumption, and prior GI distress symptoms were not recorded in the computerised database. These risk factors may have influenced the physician's choice of an NSAID and affected patient outcomes, and therefore failure to adjust for these factors could lead to confounding variables.23 24 In this study we used a case–crossover design for better control of unmeasured factors contributing to ‘confounding by indication’. Instead of using matched controls, past experience of the same individual served as his/her own control (figure 1). Therefore, stable confounders that could not be measured, were poorly measured, or were unknown, cancelled each other out.25 26

Figure 1

Comparison between case–crossover design and case–control design.

Linked data source

The Taiwan National Health Insurance database includes complete outpatient visits, hospital admissions, prescriptions, disease and vital status for 99% of the 23 million population of Taiwan. We established the longitudinal medical history of each beneficiary by linking several computerised administrative claims datasets and the National Death Registry through the civil identification number unique to each beneficiary.

Study population

We identified all patients aged ≥20 years hospitalised for lower GI adverse events in 2006 in Taiwan, including diverticulosis/diverticulitis of small intestine with haemorrhage (ICD-9-CM 562.02, 562.03); diverticulosis/diverticulitis of small intestine with haemorrhage (562.12, 562.13); haemorrhage of the rectum or anus (569.3); angiodysplasia of intestine with haemorrhage (569.85); perforation of intestine (569.83); and Dieulafoy lesion (haemorrhagic) of intestine (569.86), based on having an International Classification of Diseases, 9th Revision, Clinical Modification (ICD-9-CM) code as principal discharge diagnoses from inpatient claims (online only appendix).6 7 These diagnostic codes have been well evaluated in a previous validation study with an accuracy of approximately 95%.6 For those who had ≥2 hospitalisations for GI adverse events in 2006, only the first event was included. The date of hospital admission was defined as the index date. Patients who had concomitant diagnoses of trauma (code 800–804, 850–854, V57), and those who had any diagnosis of cancer (code 140–208), coagulopathy (code 286), inflammatory bowel disease (code 555, 556), vascular insufficiency of intestine (557), or gastroenteritis and colitis due to radiation (558.1) before hospitalisation were excluded. We also excluded patients who were admitted for any reason during 120 days before index date when the dosage or duration of NSAIDs use in hospitalisation was difficult to ascertain.

Data on drug exposure and confounding factors

The main exposures of interest in this study were the most commonly used selective and non-selective NSAIDs prescribed in Taiwan in 2006, including celecoxib (the only cyclooxygenase-2 inhibitor on the market at that time), indomethacin, sulindac, diclofenac, acemetacin, ketorolac, piroxicam, meloxicam, ibuprofen, naproxen, ketoprofen and mefenamic acid. We collected information on prescribed drug types (according to the anatomic therapeutic chemical (ATC) classification system), dosage, route of administration (oral, parenteral), date of prescription, supply days, and total number of drug pills dispensed, all from the pharmacy prescription database. We determined the mean daily dose by multiplying the number of pills dispensed by the dose prescribed divided by the recorded supply days. Data were presented as the number of defined daily doses (DDDs), which was defined by an expert panel as the typical maintenance dose required when the drug is used for its main indication in an adult.27 Other concomitant drug use that may also modify the risk of GI adverse events included histamine-2 receptor antagonists(ATC A02BA), proton pump inhibitors (A02BC), sucralfate (A02BX02), serotonin-selective receptor inhibitors (N06AB), other antidepressants (N06AX exclude N06AX01 N06AX02), nitrates (C01DA), calcium channel blockers (C08), vitamin K antagonists (B01AA), non-aspirin antiplatelet agents (B01AC except for B01AC06), low-dose aspirin (B01AC06), and systemic corticosteroids (H02A, H02B). We also collected information on age, sex, and co-morbidities, including myocardial infarction, stroke, diabetes mellitus, hypertension, ischaemic heart disease, atrial fibrillation, congestive heart failure, chronic renal disease, chronic liver disease, chronic lung disease, rheumatoid arthritis, osteoarthritis, peptic ulcer disease, and major depression based on ICD-9-CM codes (online only appendix).

Statistical analysis

We calculated the sex and age distribution for all cases in 2006. For each patient, we defined the case period as 1–30 days prior to the index date and control period as 91–120 days prior to the index date to avoid any carry-over effect. Persons were considered current users of an NSAID during the interval between the start date and end date of a prescription.

We compared NSAID use between case and control periods. A conditional logistic regression model was applied to calculate crude ORs and their 95% CIs by comparing current use of selective NSAID, nsNSAIDs overall and individual NSAID, with non-use as the reference group. Patients were further classified as having diverticulosis/diverticulitis of the small intestine with haemorrhage; diverticulosis/diverticulitis of colon with haemorrhage; haemorrhage of rectum or anus; perforation of intestine; and angiodysplasia/Dieulafoy lesion (haemorrhagic) of intestine with haemorrhage. Separate analyses were conducted for the use of oral and parenteral NSAIDs. To evaluate a possible frequency–response relationship, we estimated the mean daily dosage of NSAIDs during the case and control period among users and classified them as frequent (≥0.7 DDDs/day) or intermittent (<0.7 DDDs/day) users. Risk estimates were calculated for different frequencies of use. In the multivariable analysis, we calculated adjusted OR simultaneously controlled for the use of other individual NSAIDs and for potential time-varying confounding variables (discordant use of proton-pump inhibitors, histamine-2 receptor antagonists, systematic corticosteroids, or low-dose aspirin) between case and control periods. In the sensitivity analysis, we used different control periods of 31–60, 61–90, and 121–150 days prior to the index date to examine whether results would change substantially. In the auxiliary analysis, we conducted a case–control study by including all patients hospitalised for cataract surgery (code 366) in 2006 and applied the same exclusion criteria in the case–crossover analysis as in the control group. Unconditional logistic regression was used to calculate the adjusted OR for celecoxib, oral nsNSAIDs overall, and parenteral nsNSAIDs use after controlling for age, sex, proton pump inhibitors, histamine-2 receptor antagonists, and low-dose aspirin use between cases and controls. Figure 1 depicts the differences in choosing the comparison group between the main analysis of case–crossover design and the auxiliary analysis of case–control design.

Furthermore, stratified analysis was performed to evaluate potential effect modification. The cases were separated according to (1) age (20–49, 50–64, 65–79, ≥80 years), (2) sex (men and women), (3) previous peptic ulcer disease, and (4) osteoarthritis. We also estimated whether the risk was modified by concomitant use of aspirin, systemic corticosteroids, proton pump inhibitors, and histamine-2 receptor antagonists. To test the effect modification, a formal test of interaction was performed for each of all subgroups to examine if the difference in effect size between two subgroups was statistically significant. Two-sided p value <0.05 was considered to be statistically significant.

A nationwide study describing the NSAIDs use pattern in Denmark found that nearly 60% of all citizens claimed at least one prescription for NSAIDs.28 Assuming that a correlation coefficient for NSAID use between case period and control period is 0.5 and an OR of 2.5, a study of 170 participants would have a power ≥80% at 5% significance.


We identified 2669 patients aged ≥20 years who were hospitalised for lower GI adverse events in 2006 in Taiwan, for an annual incidence rate of 15.7 per 100 000 persons. After further excluding patients with a concomitant diagnosis of trauma or injury (n=8), those with any diagnosis of cancer (n=344), coagulopathy (n=8), inflammatory bowel disease (n=40), vascular insufficiency of intestine or gastroenteritis and colitis due to radiation (n=52) before index hospitalisation, and patients admitted for any reason in the 120 days before index date (n=920), a total of 1297 patients (55.9% male, mean age 64.8 years) for lower GI events were included in the final analysis. Table 1 summarises the proportion of patients with co-morbidities during the case period (1–30 days before index date) and control period (91–120 days before index date). Meanwhile, patients were also more likely to have additional co-morbidities, including ischaemic heart disease, chronic liver and lung disease, and peptic ulcer disease (12.0% vs 8.0%) in the case period than in the control period.

Table 1

Co-morbidities and concomitant medication during 1–30 days and 91–120 days before hospitalisation for lower gastrointestinal adverse events (N=1297)

Approximately 30% of included patients were new users of celecoxib and nsNSAIDs without having any NSAID prescription in the 180 days before the study period. Table 1 also summarises the use of selective and non-selective NSAIDs as well as concomitant medications during the case period and the control period. A higher proportion of patients used celecoxib and nsNSAIDs in the case period than in the control period. Furthermore, substantially more patients also received proton pump inhibitors, histamine-2 receptor antagonists, and systemic corticosteroids during the case period.

Table 2 presents crude and adjusted ORs and their 95% CIs for individual oral and parenteral NSAID estimated by conditional logistic regression. Risk estimates decreased to a modest extent after controlling for other individual NSAIDs and potential time-varying confounding variables in the multivariable analysis. Celecoxib was associated with an adjusted OR of 2.33 (95% CI 0.97 to 5.59) as compared with 2.26 (95% CI 1.78 to 2.85) for oral nsNSAIDs overall. Sulindac, indomethacin, acemetacin, and ketorolac seemed to be associated with higher risks as compared with other oral nsNSAIDs although the CIs were wide. A higher OR was evident for use of parenteral NSAIDs, with OR of 5.64 (95% CI 3.24 to 9.82). In the analysis for potential frequency–response relationships, risks were similar for regular (≥0.7 DDDs/day) and intermittent (<0.7 DDDs/day) users of nsNSAIDs.

Table 2

Risk of lower gastrointestinal events associated with current use of selective and non-selective NSAIDs (N=1297)

Table 3 shows the results for the association of NSAID use with individual components of the primary composite endpoints. Parenteral NSAIDs were associated with a strikingly high risk of perforation of intestine although the CI was wide. In the sensitivity analysis, results were similar for oral and parenteral nsNSAIDs using a different definition of control period prior to the index date (table 4). The risks for celecoxib were higher when using 31–60 days and 121–150 days before hospitalisation as the control period but lower when using 91–120 days, probably due to the random variation of the effect estimates resulting from relatively few patients taking celecoxib. Results from the auxiliary case–control analysis using patients hospitalised for cataract surgery as reference were similar to the findings from the case–crossover analysis. The adjusted ORs were 1.17 (95% CI 0.72 to 1.91) for celecoxib, 1.69 (95% CI 1.47 to 1.94) for oral nsNSAIDs overall, and 4.13 (95% CI: 3.09 to 5.52) for parenteral nsNSAIDs.

Table 3

Risk of individual lower gastrointestinal events associated with current use of selective and non-selective of non-steroidal anti-inflammatory drugs (NSAIDs)

Table 4

Risks of lower gastrointestinal events associated with the use of non-steroidal anti-inflammatory drugs (NSAIDs) on different definitions of the case and control periods (N=1297)

We did not find any significant effect modification by age, sex, prior history of peptic ulcer disease or osteoarthritis, as well as by other concomitant medication use for both celecoxib and oral nsNSAIDs. Oral non-selective NSAIDs posed a higher risk for lower GI events in patients without previous peptic ulcer disease (OR: 2.76, 95% CI 2.12 to 3.60) than in those with previous peptic ulcer disease (OR: 1.54, 95% CI 0.93 to 2.53). Reciprocally, celecoxib seemed to have a higher risk for patients with peptic ulcer disease with an OR of 4.22 (95% CI 0.92 to 19.38) compared to 1.93 (95% CI 0.86 to 5.51) for those without peptic ulcer disease. Relatively few patients had lower GI events in each subgroup and the CIs were wide.


Analysis of the nationwide health insurance claims database showed that lower GI safety varied remarkably among individual NSAIDs. Risk was substantially elevated when NSAIDs were used parenterally compared to use by the oral route. Celecoxib also increased the risk of lower GI events to a comparable extent despite the fact that the time gap between case and control periods slightly affected risk estimates.

A systematic review has summarised all published randomised trials, cohort studies, and case–control studies on the relationship between NSAIDs and lower GI effects.15 Among the seven studies included, six reported a lower rate of low GI injury for cyclooxygenase-2 inhibitors as compared with nsNSAIDs. A post hoc analysis of randomised trial found that serious lower GI events were significantly lower with rofecoxib than with naproxen.16 In contrast, in a randomised controlled trial with lower GI clinical events as the primary endpoint, etoricoxib was not associated with a significant decrease risk as compared with diclofenac.17 In a recently published large randomised trial, Chan and colleagues reported that celecoxib was associated with a 1% risk of GI adverse outcome, significantly lower than that of 4% for diclofenac plus omeprazole.18

Our study from Taiwan presents the most updated evaluation of hospitalisation for lower GI adverse events associated with NSAIDs use. The national hospitalisation rate for lower GI events was approximately 16 per 100 000 population, which is much lower than that reported in the US in 2006 (44 per 100 000).7 This number was very likely an under-estimate because most of the lower GI events in CONDOR were asymptomatic, presenting as anaemia due to occult GI blood loss, or bleeding from obscure sites that required sophisticated examinations such as colonoscopy or capsule endoscopy.

In our study, more than 10% of patients with lower GI adverse events had been diagnosed with peptic ulcer disease and 20% had received gastro-protective agents before their hospitalisation. These patients may have developed lower GI adverse events after first developing a peptic ulcer, or were later found to have lower GI complications after initially being treated as having upper GI problems. Surprisingly, the modifying effect of peptic ulcer disease on the association between oral non-selective NSAIDs and lower GI events was different from that between celecoxib and lower GI events. The differential modification could possibly be attributed to the reminding effect of previous peptic ulcer on physicians' prescription behaviour. On the other hand, there were more co-morbid diagnoses for heart, liver and lung disease before the lower GI event. This finding was consistent with previous reports that an increased number of co-morbidities was a risk factor for lower GI complications.6 This may be due to diseases per se, stress from hospitalisation, or more likely increased use of antiplatelet agents, anticoagulants, or systemic corticosteroids.

Our findings in general support results from prior studies suggesting oral traditional nsNSAIDs were associated with an increased risk for lower GI toxicity. Furthermore, a marked variability in terms of lower GI safety was also evident among individual NSAIDs, in particular for sulindac, indomethacin, acemetacin, and ketorolac, all with adjusted ORs above 2. This risk was consistent across all subgroups of patients with different characteristics and was not further reduced by concomitant use of proton pump inhibitors or histamine-2 receptor antagonists, although our study had insufficient power to detect potential effect modification.

Although the design of the current case–crossover study prevented us from drawing a conclusion from the head-to-head comparison between celecoxib and nsNSAIDs, the results still implied that use of celecoxib carried a risk for lower GI adverse events similar to use of other oral nsNSAIDs. This finding suggests that different pathogenic processes may be responsible for upper and lower GI adverse events. Potential mechanisms responsible for NSAIDs-associated lower GI toxicity include the changes in local eicosanoid metabolism coupled with a greater topical toxic effect increased by enterohepatic circulation of some NSAIDs. These effects lead to compromised mucosal cell integrity; increased epithelial permeability; and easier exposure to toxins produced by bacteria, bile acids, and pancreatic secretion with inflammation, later producing erosions and ulcers.3 Since cyclooxygenase-2 inhibitors have become increasingly used in high risk patients such as the elderly, patients with multiple co-morbidities, or those with prior GI events, we suggest that physicians be more aware of the potentially lower GI toxicity of cyclooxygenase-2 inhibitors.

Parenteral use of NSAIDs was associated with a substantially higher risk for lower GI adverse events. Ketorolac is a potent analgesic that is frequently used parenterally for the relief of acute pain among patients with injuries, renal colic, surgical pain, or those with intractable cancer pain.29 Several reports have suggested an unfavourable risk–benefit ratio for the use of ketorolac, due to high risks of upper gastrointestinal tract bleeding and acute renal failure.14 30–32 Diclofenac, ketoprofen and ketorolac were ranked among the NSAIDs highest in increasing risks of lower GI events. When these specific NSAIDs were used parentally, close monitoring for any possible adverse reaction is warranted, especially in high-risk populations. Our results imply that physicians should be vigilant about the GI toxicity of NSAIDs, identify risk factors for GI complications, screen for any decrease in haemoglobin level or positive faecal occult blood, and stop the offending drug as soon as a patient develops warning symptoms or signs.33

Limitations of our study

Our study has several limitations. First, we identified patients with lower GI adverse events by ICD-9-CM codes. We could not exclude the possibility that some patients with upper GI events may be misdiagnosed or incorrectly coded as having lower GI events. Second, we did not take over-the-counter NSAIDs into account, and therefore, misclassification might ensue and bias study results towards the null. Third, we did not have GI risk factor data such as body mass index, cigarette smoking, alcohol consumption, and prior subtle GI symptoms. However, these unmeasured factors probably did not change substantially during the relatively short study period and thus could be largely controlled by the case–crossover design. Fourth, we could not exclude the possibility that the multivariable analysis may over-control some of the variables such as concomitant use of proton pump inhibitors or histamine-2 receptor antagonists. The use of these medications might be an indicator of pre-existing NSAIDs-related GI toxicity. Fifth, due to the fact that haemoglobin levels were not available in the claims database, we could not evaluate the extent of haemoglobin decreases in our patient population: it was the main endpoint in favouring celecoxib over diclofenac plus omeperazole in the recently published CONDOR trial.19 However, among the patients included, approximately 15% had a diagnosis of anaemia and 7% received treatment with oral iron or vitamin B12 supplement before their hospitalisation. Therefore, this missing piece of information (haemoglobin) probably would not affect basic conclusions. Finally, this study only assessed the risk associated with short-term use of NSAIDs. Our findings may not be generalisable to risks associated with long-term use.


This study suggests that parenteral NSAIDs are associated with a significantly increased risk of lower GI toxicity. Celecoxib posed a similar risk as oral nsNSAIDs. The association of NSAIDs with specific lower GI adverse events and long-term complications requires further investigation.


The authors are indebted to Dr Sengwee Darren Toh and Professor Paul Chang for their critical suggestions in revising the manuscript and Yi-Ting Chou for helping with data management.


  • Chia-Hsuin Chang and Jou-Wei Lin contributed equally to this work.

  • Funding This study was supported by Taiwan Department of Health grant DOH098-TD-D-113-098016. H-C Chen had full access to all study data and takes responsibility for the integrity of the data and the accuracy of the data analysis.

  • Competing interests None declared.

  • Provenance and peer review Not commissioned; externally peer reviewed.


Relevant Article