Variability among nonsteroidal antiinflammatory drugs in risk of upper gastrointestinal bleeding

Authors

  • Elvira L. Massó González,

    1. Spanish Centre for Pharmacoepidemiological Research, Madrid, Spain
    Current affiliation:
    1. AstraZeneca, Madrid, Spain
    Search for more papers by this author
    • Ms Massó González and Dr. Patrignani contributed equally to this work.

  • Paola Patrignani,

    1. G. d'Annunzio University, School of Medicine, Centro Studi dell'Invecchiamento, Chieti, Italy
    Search for more papers by this author
    • Ms Massó González and Dr. Patrignani contributed equally to this work.

  • Stefania Tacconelli,

    1. G. d'Annunzio University, School of Medicine, Centro Studi dell'Invecchiamento, Chieti, Italy
    Search for more papers by this author
  • Luis A. García Rodríguez

    Corresponding author
    1. Spanish Centre for Pharmacoepidemiological Research, Madrid, Spain
    • Spanish Centre for Pharmacoepidemiological Research, Almirante 28, 2, 28004 Madrid, Spain
    Search for more papers by this author
    • Dr. García Rodríguez has received consulting fees from AstraZeneca (less than $10,000) and research grants from Novartis and Bayer.


Abstract

Objective

Traditional nonsteroidal antiinflammatory drugs (NSAIDs) increase the risk of upper gastrointestinal (GI) bleeding/perforation, but the magnitude of this effect for coxibs in the general population and the degree of variability between individual NSAIDs is still under debate. This study was undertaken to assess the risk of upper GI bleeding/perforation among users of individual NSAIDs and to analyze the correlation between this risk and the degree of inhibition of whole blood cyclooxygenase 1 (COX-1) and COX-2 in vitro.

Methods

We conducted a systematic review of observational studies on NSAIDs and upper GI bleeding/perforation published between 2000 and 2008. We calculated pooled relative risk (RR) estimates of upper GI bleeding/perforation for individual NSAIDs. Additionally, we verified whether the degree of inhibition of whole blood COX-1 and COX-2 in vitro by average circulating concentrations predicted the RR of upper GI bleeding/perforation.

Results

The RR of upper GI bleeding/perforation was 4.50 (95% confidence interval [95% CI] 3.82–5.31) for traditional NSAIDs and 1.88 (95% CI 0.96–3.71) for coxibs. RRs lower than that for NSAIDs overall were observed for ibuprofen (2.69 [95% CI 2.17–3.33]), rofecoxib (2.12 [95% CI 1.59–2.84]), aceclofenac (1.44 [95% CI 0.65–3.2]), and celecoxib (1.42 [95% CI 0.85–2.37]), while higher RRs were observed for ketorolac (14.54 [95% CI 5.87–36.04]) and piroxicam (9.94 [95% CI 5.99–16.50). Estimated RRs were 5.63 (95% CI 3.83–8.28) for naproxen, 5.57 (95% CI 3.94–7.87) for ketoprofen, 5.40 (95% CI 4.16–7.00) for indomethacin, 4.15 (95% CI 2.59–6.64) for meloxicam, and 3.98 (95% CI 3.36–4.72) for diclofenac. The degree of inhibition of whole blood COX-1 did not significantly correlate with RR of upper GI bleeding/perforation associated with individual NSAIDs (r2 = 0.34, P = 0.058), but a profound and coincident inhibition (>80%) of both COX isozymes was associated with higher risk. NSAIDs with a long plasma half-life and with a slow-release formulation were associated with a greater risk than NSAIDs with a short half-life.

Conclusion

The results of our analysis demonstrate that risk of upper GI bleeding/perforation varies between individual NSAIDs at the doses commonly used in the general population. Drugs that have a long half-life or slow-release formulation and/or are associated with profound and coincident inhibition of both COX isozymes are associated with a greater risk of upper GI bleeding/perforation.

Traditional nonaspirin nonsteroidal antiinflammatory drugs (NSAIDs) have been shown to increase the risk of upper gastrointestinal (GI) bleeding/perforation (1). Because of the widespread use of NSAIDs as analgesic, antiinflammatory, and antipyretic drugs, their serious upper GI complications are a major public health concern. To reduce the morbidity associated with NSAIDs it is necessary to establish specific estimates for individual drugs and individual groups of patients with different risk profiles.

In the late 1990s some NSAIDs were introduced in the market with claims of better GI safety than traditional NSAIDs (2, 3). These new NSAIDs, known as coxibs, are selective inhibitors of cyclooxygenase 2 (COX-2). Indeed, a reduced incidence of serious GI adverse effects compared with traditional NSAIDs has been demonstrated for 2 highly selective COX-2 inhibitors, rofecoxib and lumiracoxib, in large randomized clinical trials (4, 5). This was a proof of concept that sparing of COX-1 in the GI tract and possibly in platelets translates into a safer GI profile (3). In fact, COX-1 is constitutively expressed in the stomach and platelets (6), whereas COX-2 does not appear to be expressed, or is expressed at very low levels. Experimental results using selective pharmacologic inhibition or genetic deletion of COX-1 and COX-2 in mice have shown that COX-2 plays an important role in the healing of preexisting ulcers (7, 8). In fact, COX-2 is rapidly up-regulated in response to growth factors and cytokines, and it has been demonstrated that both COX-2 messenger RNA and COX-2 protein were strongly expressed in mouse stomachs in which ulcers had been induced (9, 10).

In this report, we present a summary of the main results of observational epidemiologic studies published between 2000 and 2008 addressing the risk of upper GI bleeding/perforation among NSAID users (traditional NSAIDs and coxibs). Additionally, we verified whether the degree of inhibition of whole blood COX-1 and COX-2 in vitro by average circulating concentrations of individual traditional NSAIDs and coxibs predicted the relative risk (RR) of upper GI bleeding/perforation observed in this meta-analysis of observational studies.

METHODS

We conducted a Medline search using the keywords “non-steroidal anti-inflammatory drug” or “cyclooxygenase inhibitor” combined with “peptic ulcer perforation” or “peptic ulcer” or “peptic ulcer hemorrhage” or “stomach ulcer.” The search was restricted to human studies on adults published in English between January 1, 2000 and October 20, 2008. We identified 401 entries and examined their abstracts. We reviewed the abstracts of all 401 of the entries retrieved by this strategy. When no clear reason for exclusion from the meta-analysis was revealed in the study abstract, the full article was obtained. We also reviewed the references of selected articles and previous reviews related to NSAIDs and GI outcomes, in order to capture any eligible articles that might have escaped our initial screening.

Predefined inclusion criteria were applied, and data were extracted independently by 2 of the authors (ELMG and LAGR), using a standardized data extraction form. The list of characteristics was based on literature on epidemiologic study methods in general and on previous meta-analyses of NSAID and upper GI bleeding/perforation (11). In summary, to be included in the analysis, articles had to 1) be reporting case–control or cohort studies evaluating traditional NSAID or coxib use and upper GI bleeding/perforation in the general population, and 2) provide either an estimate or enough data for us to estimate a relative risk comparing NSAID users with nonusers. Estimates published solely in letter, commentary, or abstract forms were not considered. When 2 articles reported results from the same study population, the more recent version was chosen. However, if the earlier version provided additional subanalyses, they were considered. Decisions regarding inclusion of studies were reached by consensus.

We found 18 original studies that initially were all included, but in a second review, we excluded 9 of them for the following reasons: the exposure window was loosely defined as “use in the prior year” (12), no information was provided about the reference group for the exposure (13), the study population was a cohort of myocardial infarction survivors (14), the outcome was defined as upper GI event including peptic ulcer and/or bleeding (2 articles) (15, 16), the control group was not population based (2 articles) (17, 18), or the article only provided estimators for interaction with NSAIDs and other exposures (19). Finally, there were 2 articles published using the same observational study (20, 21), and we collected data from both articles but only one is considered as part of the meta-analysis (21). Nine studies were selected for the meta-analysis (1,21–28). We did not contact study authors for additional data, except in the case of one article from our research group that did not provide data on the 95% confidence interval (95% CI) for the relative risk of some individual NSAIDs; this information was only described in a figure in the original article (1). We also included fully adjusted estimates of risk with individual NSAIDs, from another article in which we had participated (28).

Adjusted RRs estimated in the original studies were collected, together with information on study methodology and objective quality-related characteristics. Variables assessed included authors, year of publication, journal, geographic region, study years, source population, inclusion and exclusion criteria, source of cases, source of controls, outcome definition, source of exposure information, exposure definition, sample size, validation of cases obtained from computerized data by medical chart review, and attempt to control for confounding.

We calculated a summary RR and 95% CI, weighing study estimates by the inverse of the variance (29). In addition to these fixed-effects estimates, we calculated the corresponding estimates using a random-effects regression model: unless otherwise stated, results presented are from the random-effects model. We included in the analysis only those estimators with >5 controls exposed, and we calculated a summary effect for individual NSAIDs addressed in at least 2 studies. The odds ratio from case–control studies was assumed to provide a valid estimate of the RR. To determine whether it was appropriate to pool estimates of RR into one common summary measure, the heterogeneity in effects between studies was analyzed using the DerSimonian and Laird's test statistic (Q). We explored potential publication bias qualitatively, using a “funnel plot” (30).

When possible, RRs for subgroups of interest were extracted or computed from raw data provided in the publication. Definitions of categories for these purposes reflected those used in the original studies (31).

Finally, we computed the summary estimate for individual NSAIDs, also including data from a previous meta- analysis conducted with studies published between 1990 and 1999 (11), in order to derive a cumulative estimate of effect with all information available since 1990. This analysis is presented in Figure 1.

Figure 1.

Pooled estimates of the relative risk (RR) and 95% confidence interval (95% CI) of upper gastrointestinal bleeding/perforation associated with the use of individual nonsteroidal antiinflammatory drugs (NSAIDs) and total NSAID use from published studies since 1990. P values are for the heterogeneity test results; n values are the number of studies. ∗ = reported in studies published after 2000.

The inhibitory effects on platelet COX-1 and monocyte COX-2 were studied using the maximum or peak plasma concentration of a drug (Cmax) obtained after dosing that corresponded to the daily dose of different NSAIDs most commonly used (32–35), i.e., naproxen 750 mg (253 μM), ibuprofen 1,200 mg (39 μM), meloxicam 7.5 mg (3 μM), celecoxib 200 mg (1.8 μM), rofecoxib 25 mg (1 μM), indomethacin 75 mg (3 μM), diclofenac 100 mg (0.8 μM), piroxicam 20 mg (16.6 μM), ketoprofen 150 mg (9.4 μM), ketorolac 10 mg (2 μM), aceclofenac 100 mg (0.39 μM of diclofenac; aceclofenac is a prodrug biotransformed in vivo to diclofenac [35]). COX inhibition was determined in vitro using human whole blood assays (36, 37). Peripheral venous blood samples were obtained, with written informed consent, from 10 healthy subjects ages 24–35 years. The in vitro study was approved by the local Ethics Committee. Institutional review board review was not required.

RESULTS

The 9 studies included in this review are described in Supplementary Table 1 (available in the online version of this article at http://www3.interscience.wiley.com/journal/76509746/home). Two were cohort studies (23, 24), 3 were nested case–control studies (1, 22, 26), and 4 were case–control studies (21, 25, 27, 28). All nested case–control and cohort studies used computerized records as the source of exposure and outcome information, including 2 studies based on hospital registers linked to the Pharmacoepidemiologic Database in Denmark (24, 26), while the 4 case–control studies were field-based with patient interviews.

All patients had been hospitalized or referred to a specialist for upper GI bleeding and/or perforation. Many studies had the following exclusion criteria: cancer (22, 24, 25, 27–29) (n = 6), esophageal varices (1, 22, 24, 25, 27, 28) (n = 6), Mallory-Weiss disease (1, 22, 24, 27, 28) (n = 5), alcoholism (1, 22, 24) (n = 3), chronic liver disease (1, 22, 24, 27, 28) (n = 5), or/and coagulopathies (1, 22, 25, 27, 28) (n = 5). There were 3 studies that did not apply any of the exclusion criteria mentioned above (21,23,26).

Differences in RRs associated with different study designs were found. Cohort studies and nested case–control studies showed lower estimates in comparison with field-based case–control studies.

All studies had specific definitions of exposure and outcome, and similar ascertainment of comparison groups. The most frequent confounders considered were age (1, 21–23, 25–28), sex (1, 21–23, 25–28), prior ulcer history (1, 21, 22, 25–28), and concomitant medication (1, 21, 22, 25, 26, 28).

Estimates of the RR of upper GI bleeding/perforation associated with traditional NSAID use (1,21–24,26–28) are shown in Table 1. The pooled RR was 4.50 (95% CI 3.82–5.31). Statistically significant heterogeneity was found between effect measures obtained in different studies. Table 1 shows the RRs of upper GI bleeding/perforation for traditional NSAIDs compared with no use, including dose- and duration-response relationships. Estimates of risk from cohort and nested case–control studies were relatively similar, resulting in a traditional NSAID summary estimate of 4.15 (95% CI 3.33–5.18) while, as noted above, risk estimates from field-based case–control studies tended to be higher, with a summary estimate of 5.44 (95% CI 4.01–7.37).

Table 1. Relative risk (95% confidence interval) of upper gastrointestinal bleeding/perforation associated with nonsteroidal antiinflammatory drug (NSAID) use and dose and duration of the treatment: results from published studies and pooled estimates
Author, year (ref.)No. of casesNo. of controlsNSAIDsDosageDuration
CurrentPastLowHigh1–30 days31–180 days180–365 days>365 days
Cohort and nested case–control studies          
 Garcia-Rodriguez and Hernandez-Diaz, 2001 (22)2,10511,5004.1 (3.6–4.8)1.0 (0.7–1.3)2.4 (1.9–3.1)4.9 (4.1–5.8)4.3 (3.3–5.6)5.2 (4.0–6.9)4.8 (3.3–7.0)3.5 (2.8–4.3)
 Mamdani et al, 2002 (23)1874.0 (2.3–6.9)
 Mellemkjaer et al, 2002 (24)5155.6 (5.0–6.2)1.4 (1.3–1.5)
 Norgard et al, 2004 (26)7802,9063.3 (2.4–4.4)
 Garcia-Rodriguez and Barreales Tolosa, 2007 (1)1,56110,0003.7 (3.1–4.3)1.3 (1.1–1.5)2.5 (2.0–3.2)4.9 (4.0–6.1)4.6 (3.4–6.2)4.1 (2.7–6.2)3.4 (2.5–4.8)3.1 (2.4–4.0)
Pooled cohort and nested case–control studies          
 Fixed effects  4.58 (4.26–4.93)1.36 (1.28–1.45)2.45 (2.07–2.90)4.90 (4.29–5.60)4.43 (3.63–5.40)4.84 (3.85–6.08)3.94 (3.08–5.04)3.33 (2.82–3.92)
 Random effects  4.15 (3.33–5.18)1.30 (1.13–1.49)2.45 (2.07–2.90)4.90 (4.29–5.60)4.43 (3.63–5.40)4.84 (3.85–6.08)3.99 (2.85–5.58)3.33 (2.82–3.92)
 Heterogeneity (P)  26.96 (0.000)4.742 (0.093)0.056 (0.814)0.000 (1.00)0.109 (0.741)0.878 (0.349)1.844 (0.175)0.509 (0.476)
Case–control studies          
 Lanas et al, 2003 (21)1,1222,2317.4 (4.5–12.0)
 Laporte et al, 2004 (25)2,8137,193
 Sakamoto et al, 2006 (27)1753476.1 (2.7–13.4)7.4 (2.3–23.9)
 Lanas et al, 2006 (28)2,7775,5324.7 (4.0–5.4)0.9 (0.7–1.1)3.5 (2.9–4.2)6.5 (5.2–8.1)6.9 (5.7–8.4)5.5 (3.3–9.3)2.5 (1.6–3.7)2.2 (1.7–2.9)
Pooled case–control studies          
 Fixed effects  4.92 (4.27–5.67)6.53 (5.25–8.12)
 Random effects  5.44 (4.01–7.37)6.53 (5.25–8.12)
 Heterogeneity (P)  3.295 (0.193)0.046 (0.831)
All studies pooled          
 Fixed effects  4.65 (4.36–4.96)1.32 (1.24–1.41)2.88 (2.54–3.27)5.30 (4.73–5.94)5.56 (4.84–6.38)4.94 (4.01–6.09)3.51 (2.84–4.34)2.97 (2.58–3.42)
 Random effects  4.50 (3.82–5.31)1.17 (0.96–1.42)2.79 (2.17–3.58)5.36 (4.57–6.29)5.22 (3.78–7.20)4.94 (4.01–6.09)3.47 (2.45–4.91)2.90 (2.22–3.80)
 Q heterogeneity (P)  31.03 (0.000)16.76 (0.001)7.78 (0.020)4.89 (0.180)9.92 (0.007)1.07 (0.585)5.22 (0.074)7.21 (0.027)

In the analysis according to daily dosage, use of traditional NSAIDs at low or medium doses was associated with an RR of 2.79 (95% CI 2.17–3.58), and the summary RR among those receiving higher doses was 5.36 (95% CI 4.57–6.29). The increased risk was present throughout the duration of treatment; however, most notably among field-based studies, the data indicated a slightly greater hazard near the time of initiation of therapy.

Table 2 shows the estimates of upper GI bleeding/perforation according to the plasma half-life formulation of NSAIDs, as well as with stratification by daily dosage. NSAIDs with a long plasma half-life, as well as NSAIDs administered in a slow-release formulation, were associated with a greater RR than NSAIDs with a short half-life, even after accounting for the daily dosage administered.

Table 2. Relative risk (95% confidence interval) of upper gastrointestinal bleeding/perforation according to plasma half-life formulation of nonsteroidal antiinflammatory drugs (NSAIDs), compared with nonuse and stratified according to daily dosage: results from published studies and pooled estimates
Author, year (ref.)No. of casesNo. of controlsTraditional NSAIDs overallLow-to-medium–dose NSAIDsHigh-dose NSAIDs
Half-life <12 hoursHalf-life ≥12 hoursSlow- releaseHalf-life <12 hoursHalf-life ≥12 hoursSlow- releaseHalf-life <12 hoursHalf-life ≥12 hoursSlow- release
Garcia-Rodriguez and Hernandez- Diaz, 2001 (22)2,10511,5003.1 (2.5–3.8)4.5 (3.5–5.9)5.4 (4.0–7.1)2.4 (1.8–3.2)2.5 (1.3–4.9)3.6 (1.6–8.0)4.1 (3.1–5.3)5.2 (3.9–7.0)5.7 (4.3–7.8)
Mamdani et al, 2002 (23)187
Mellemkjaer et al, 2002 (24)515
Norgard et al, 2004 (26)7802,906
Garcia-Rodriguez and Barreales Tolosa, 2007 (1)1,56110,0002.4 (1.9–3.1)4.5 (3.3–6.2)6.5 (4.7–8.9)2.0 (1.5–2.7)3.0 (1.7–5.4)4.3 (2.6–7.2)3.2 (2.3–4.6)5.4 (3.7–7.8)8.4 (5.7–12.6)
Lanas et al, 2003 (21)1,1222,231
Laporte et al, 2004 (25)2,8137,193
Sakamoto et al, 2006 (27)175347
Lanas et al, 2006 (28)2,7775,5324.0 (3.3–4.9)9.4 (6.9–12.7)3.5 (2.8–4.5)6.4 (3.9–10.6)5.0 (3.6–6.9)12.4 (8.1–18.8)
All studies pooled           
Fixed effects  3.21 (2.83–3.63)5.62 (4.76–6.65)5.87 (4.74–7.26)2.68 (2.30–3.13)3.98 (2.86–5.53)4.09 (2.66–6.28)4.08 (3.41–4.87)6.42 (5.25–7.86)6.55 (5.17–8.32)
Random effects  3.12 (2.35–4.14)5.74 (3.58–9.21)5.87 (4.74–7.26)2.58 (1.84–3.62)3.74 (2.07–6.75)4.09 (2.66–6.28)4.06 (3.21–5.13)6.92 (4.15–11.57)6.76 (4.64–9.86)
Q heterogeneity (P)  10.289 (0.006)15.615 (0.000)0.717 (0.397)9.239 (0.010)6.274 (0.043)0.134 (0.715)3.390 (0.184)12.214 (0.002)2.348 (0.125)

Table 3 shows the estimates of RRs of upper GI bleeding/perforation for individual traditional NSAIDs, compared with no use. The pooled estimates of RR for the 3 most widely used traditional NSAIDs were 5.63 (95% CI 3.83–8.28) for naproxen, 2.69 (95% CI 2.17–3.33) for ibuprofen, and 3.98 (95% CI 3.36–4.72) for diclofenac. Piroxicam and ketorolac were the traditional NSAIDs with the greatest RR (9.94 [95% CI 5.99–16.50] and 14.54 [95% CI 5.87–36.04], respectively).

Table 3. Relative risk (95% confidence interval) of upper gastrointestinal bleeding/perforation associated with use of individual nonsteroidal antiinflammatory drugs (NSAIDs): results from published studies and pooled estimates
Author, year (ref.)No. of casesNo. of controlsNSAID
MeloxicamNaproxenIbuprofenDiclofenacIndomethacinKetoprofenPiroxicanKetorolacAceclofenac
Cohort and nested case–control studies           
 García-Rodríguez and Hernandez-Diaz, 2001 (22)2,10511,5003.8 (0.8–17.2)4.0 (2.8–5.8)2.5 (1.9–3.4)4.6 (3.6–5.8)5.2 (3.2–8.3)3.3 (1.9–5.9)6.2 (3.7–10.1)
 Mamdani et al, 2002 (23)187
 Mellemkjaer et al, 2002 (24)5153.03 (2.1–4.2)2.41 (2.0–2.9)4.87 (3.5–6.6)4.28 (2.9–6.0)6.3 (4.5–8.5)5.00 (3.3–7.2)
 Norgard et al, 2004 (26)7802,906
 García-Rodríguez and Barreales Tolosa, 2007 (1)1,56110,0002.7 (1.5–4.8)8.1 (5.2–12.6)2.0 (1.4–2.9)3.7 (2.9–4.6)7.2 (3.8–13.7)5.4 (1.7–16.5)
Pooled cohort and nested case–control studies           
 Fixed effects  2.82 (1.64–4.85)4.25 (3.42–5.29)2.36 (2.04–2.73)4.26 (3.68–4.94)4.96 (3.81–6.45)5.40 (4.12–7.06)5.42 (3.98–7.38)
 Random effects  2.82 (1.64–4.85)4.54 (2.65–7.79)2.36 (2.04–2.73)4.28 (3.62–5.05)4.96 (3.81–6.45)4.95 (3.12–7.85)5.42 (3.98–7.38)
 Heterogeneity (P)  0.167 (0.683)11.928 (0.003)0.99 (0.609)2.516 (0.284)1.967 (0.374)3.805 (0.149)0.44 (0.507)
Case–control studies           
 Lanas et al, 2003 (21)1,1222,2315.3 (2.6–10.8)5.1 (3.1–8.4)18.5 (9.2–36.9)0.7 (0.3–1.6)
 Laporte et al, 2004 (25)2,8137,1935.7 (2.2–15.0)10.0 (5.7–17.6)3.1 (2.0–4.9)3.7 (2.6–5.4)10.0 (4.4–22.6)10.0 (3.9–25.8)15.5 (10.0–24.2)24.7 (8.0–77.0)1.4 (0.6–3.3)
 Sakamoto et al, 2006 (27)175347
 Lanas et al, 2006 (28)2,7775,5326.4 (3.0–13.7)7.0 (4.7–10.4)3.7 (2.9–4.7)2.9 (2.2–3.8)5.3 (2.5–11.1)6.2 (2.0–19.0)12.3 (8.0–18.9)9.7 (3.9–24.2)2.6 (1.6–4.3)
Pooled case–control studies           
 Fixed effects  6.12 (3.37–11.10)8.14 (0.23–290.99)3.56 (2.87–4.40)3.42 (2.80–4.18)7.07 (4.07–12.26)8.21 (3.98–16.92)14.45 (10.90–19.15)14.02 (6.89–28.53)1.75 (1.20–2.56)
 Random effects  6.12 (3.37–11.10)8.14 (0.23–290.99)3.56 (2.87–4.40)3.59 (2.66–4.86)7.11 (3.82–13.23)8.21 (3.98–16.92)14.45 (10.90–19.15)14.54 (5.87–36.04)1.44 (0.65–3.20)
 Heterogeneity (P)  0.034 (0.853)0.016 (0.992)0.464 (0.496)4.05 (0.132)1.26 (0.261)0.406 (0.524)1.123 (0.57)1.59 (0.208)7.33 (0.026)
All studies pooled           
 Fixed effects  4.01 (2.68–5.99)5.16 (4.33–6.15)2.69 (2.38–3.03)3.95 (3.50–4.44)5.30 (4.18–6.72)5.68 (4.41–7.31)9.25 (7.51–11.38)14.02 (6.89–28.53)1.75 (1.20–2.56)
 Random effects  4.15 (2.59–6.64)5.63 (3.83–8.28)2.69 (2.17–3.33)3.98 (3.36–4.72)5.40 (4.16–7.00)5.57 (3.94–7.87)9.94 (5.99–16.50)14.54 (5.87–36.04)1.44 (0.65–3.20)
 Q heterogeneity (P)  3.75 (0.289)22.49 (0.000)11.22 (0.024)9.59 (0.088)4.52 (0.340)5.34 (0.254)22.75 (0.000)1.59 (0.208)7.33 (0.026)

Only individual traditional NSAIDs addressed in 2 or more studies were included in the dose-response analyses (1, 22, 24, 25, 28). NSAIDs for which data were obtained according to this criterion were naproxen and ibuprofen. A clear dose-response was found for both NSAIDs (data not shown).

For coxibs, the pooled RR of upper GI bleeding/perforation was 1.88 (95% CI 0.96–3.71), with RRs of 1.42 (95% CI 0.85–2.37) for celecoxib and 2.12 (95% CI 1.59–2.84) for rofecoxib, compared with no NSAID use (Table 4). One report included information on an effect of etoricoxib.

Table 4. Relative risk (95% confidence interval) of upper gastrointestinal bleeding/perforation associated with coxib use overall and use of the individual coxibs celecoxib and rofecoxib, compared with no nonsteroidal antiinflammatory drug use: results from published studies and pooled estimates
Author, year (ref.)No. of casesNo. of controlsCoxibs, current useCoxibs, past useCelecoxibRofecoxib
Cohort and nested case–control  studies      
 García-Rodríguez and Hernandez-Diaz, 2001 (22)2,10511,500
 Mamdani et al, 2002 (23)187 1.0 (0.7–1.6)1.9 (1.3–2.8)
 Mellemkjaer et al, 2002 (24)365 
 Norgard et al, 2004 (26)7802,9061.3 (0.7–2.8)2.1 (1.2–3.5)
 García-Rodríguez and Barreales Tolosa, 2007 (1)1,56110,0002.6 (1.9–3.6)0.7 (0.5–1.0)2.7 (1.7–4.5)2.3 (1.4–3.8)
 Lanas et al, 2003 (21)1,1222,231
 Laporte et al, 2004 (25)2,8137,1937.2 (2.3–23.0)
 Sakamoto et al, 2006 (27)175347
 Lanas et al, 2006 (28)2,7775,5321.5 (0.9–2.4)1.0 (0.4–2.3)1.0 (0.4–2.1)2.1 (1.1–4.0)
All studies pooled      
 Fixed effects  2.10 (1.61–2.75)0.71 (0.51–0.98)1.42 (1.09–1.86)2.08 (1.64–2.63)
 Random effects  1.88 (0.96–3.71)0.71 (0.51–0.98)1.42 (0.85–2.37)2.12 (1.59–2.84)
 Heterogeneity (P)  5.51 (0.019)0.070 (0.791)9.98 (0.019)5.61 (0.230)

Calculation of summary estimates of RR associated with upper GI bleeding/perforation for individual NSAIDs with inclusion of all published studies starting in 1990 and fulfilling our inclusion criteria generally yielded results that were only slightly different from those obtained using only studies published after 2000, but resulted in slightly more precise estimates (Figure 1). A statistically significant correlation (r2 = 0.98, P < 0.0001) between the 2 sets of estimates of RR was found (data available from the author upon request).

The Begg's funnel plot of standard error (log) as a measure of the size of the study versus the RR (log) was shaped like a pyramid (data not shown). Therefore, publication bias was unlikely in this meta-analysis.

The concentration-response curves for inhibition of whole blood COX-1 and COX-2 activities in vitro by ibuprofen, naproxen, meloxicam, celecoxib, rofecoxib, indomethacin, piroxicam, and diclofenac used in this meta-analysis were previously reported (38). Those for ketorolac, aceclofenac, and ketoprofen are available from the author upon request. For individual NSAIDs, the risk of upper GI bleeding/perforation was 4.15 (95% CI 2.59–6.64) for meloxicam, 3.98 (95% CI 3.36–4.72) for diclofenac (as noted above), 2.69 (95% CI 2.17–3.33) for ibuprofen (as noted above), 2.12 (95% CI 1.59–2.84) for rofecoxib, 1.42 (95% CI 0.85–2.37) for celecoxib, and 1.44 (95% CI 0.65–3.2) for aceclofenac. RR estimates of >5 were observed for ketorolac (14.54 [95% CI 5.87–36.04]) (as noted above), piroxicam (9.94 [95% CI 5.99–16.50]) (as noted above), naproxen (5.63 [95% CI 3.83–8.28]) (as noted above), ketoprofen (5.57 [95% CI 3.94–7.87]), and indomethacin (5.40 [95% CI 4.16–7.00]).

Interestingly, all of these latter individual traditional NSAIDs were characterized by a profound and coincident inhibition (>80%) of both COX isozymes. We ascertained the relationship between the degree of inhibition of whole blood COX-1 in vitro produced by average circulating therapeutic concentrations of different NSAIDs and the corresponding RRs of upper GI bleeding/perforation from our meta-analysis and found a trend toward correlation between these 2 sets of data (r2 = 0.34, P = 0.058) (Figure 2A). As shown in Figure 2B, no correlation between the RR of upper GI bleeding/perforation and the degree of inhibition of whole blood COX-2 in vitro produced by average circulating therapeutic concentrations of different NSAIDs was found (r2 = 0.06, P = 0.46). Similarly, we did not observe any significant correlation between the RR of upper GI bleeding/perforation and the ratio of COX-2:COX-1 inhibition at Cmax (an index of achieved COX-2 selectivity), the COX-1:COX-2 50% inhibition concentration (IC50) ratio (an index of experimental COX-2 selectivity in vitro, i.e., a chemical feature of the drugs), or the IC50 values for COX-1 or COX-2 (data available from the author upon request).

Figure 2.

Relationship between the degree of inhibition of whole blood cyclooxygenase 1 (COX-1) (A) and COX-2 (B) at the maximum plasma concentration (Cmax) of individual nonsteroidal antiinflammatory drugs and relative risk (RR) of upper gastrointestinal bleeding/perforation. Linear regression analysis (performed using GraphPad InStat) yielded a nonsignificant correlation between the 2 variables. Rofe = rofecoxib; cele = celecoxib; melo = meloxicam; aceclof = aceclofenac; diclo = diclofenac; indo = indomethacin; piro = piroxicam; ibu = ibuprofen; ketoro = ketorolac; nap = naproxen; keto = ketoprofen.

DISCUSSION

Previous overviews of epidemiologic studies published in the 1990s consistently showed that use of traditional NSAIDs was associated with an ∼4-fold increased risk of upper GI bleeding/perforation and a clear dose-dependent elevation in risk (39–41). This meta-analysis has added the cumulative evidence published since 2000, confirming this magnitude of increased risk with traditional NSAIDs as well as the dose-response relationship. In addition, we report here for the first time the summary estimate of upper GI bleeding/perforation with coxibs, which are associated with an ∼2-fold increased risk. In addition to this difference in upper GI bleeding/perforation risk between traditional NSAIDs and coxibs, we saw marked variability between individual NSAIDs, beyond the difference between traditional NSAIDs and coxibs. On the one hand, some traditional NSAIDs, such as ketorolac and piroxicam, were associated with a greatly elevated risk of upper GI bleeding/perforation. On the other hand, another traditional NSAID, ibuprofen, and a coxib, rofecoxib, shared a very similar risk of upper GI bleeding/perforation. Presumably, this latter result is partly determined by the fact that ibuprofen is mostly administered at low doses. Taken together these results suggest that NSAID-associated upper GI toxicity is the result of 2 pharmacologic features: drug exposure (lower for ibuprofen versus rofecoxib) and sparing of COX-1 activity (higher for rofecoxib versus ibuprofen).

These findings support the notion that there are multifactorial determinants in the risk of upper GI bleeding/perforation among NSAID users. The clinical background, such as a previous history of peptic ulcer or its complications, use of concomitant medications (e.g., aspirin), or a possible genetic susceptibility still to be elucidated, all play a role in determining the final absolute risk in an individual. In the present study, we aimed to shed some light on the contribution of pharmacokinetic and pharmacodynamic features of the different NSAIDs to the risk of upper GI bleeding/perforation. Importantly, we showed that persistent exposure to the drug is an important independent determinant; in fact, drugs with a long half-life or slow-release formulation were associated overall with a greater risk than NSAIDs with a short half-life. This characteristic was illustrated by the findings for piroxicam, an NSAID associated with an elevated risk of upper GI bleeding/perforation, which has the longest plasma half-life of all of the studied NSAIDs (i.e., 50 hours) due to its enterohepatic circulation property (32).

As expected, we observed the lowest GI toxicity with coxibs, i.e., celecoxib and rofecoxib, which supports the notion that sparing of COX-1 in the GI tract and possibly in platelets (3) translates clinically to a lower risk of upper GI bleeding/perforation. Accordingly, we found a relationship between the degree of inhibition of whole blood COX-1 in vitro produced by average circulating therapeutic concentrations of different NSAIDs and the RR of upper GI bleeding/perforation for the same drugs determined in our meta-analysis. However, this relationship was not as strong as could have been expected, suggesting a more complex mechanistic explanation of a prostanoid-dependent role in GI physiology and pathology.

Several lines of evidence obtained from studies of experimental animals suggest a cytoprotective role of prostanoids in the GI tract, derived from both COX isozymes (8). Thus, COX-1–derived prostanoids (mainly prostaglandin E2) are involved in the constitutive defense mechanism operating under physiologic conditions (3); in contrast, endogenous prostaglandin E2 derived from COX-2 plays an important part in the spontaneous healing of gastric ulcers (7, 8, 42, 43). This is supported by clinical data showing that coxibs are still associated with a small risk of upper GI bleeding/perforation, though smaller than that caused by traditional NSAIDs (4, 5). The mechanism could include the expression of COX-2 in previously silent ulcers, as a defense mechanism operating for the resolution of the GI damage. Thus, inhibition of COX-2 activity might translate into delayed healing. Furthermore, Schmassmann et al (8) showed that inhibition of both COX-1 and COX-2 delayed healing more markedly than inhibition of COX-2 alone and suggested that COX-1–derived prostanoids may be important in the healing mechanism when COX-2–derived prostanoids are deficient.

Interestingly, we found that individual traditional NSAIDs associated with high risk of upper GI bleeding/perforation (ketorolac, piroxicam, naproxen, ketoprofen, and indomethacin) were characterized by a profound and coincident inhibition (>80%) of both COX isozymes. These results may suggest that both COX isozymes are a source of cytoprotective prostanoids; thus, simultaneous inhibition, translating into a profound suppression of prostanoids, might be a hazard for the GI system. This is mechanistically consistent with results obtained in mice, showing that inhibition of both COX-1 and COX-2 is required for the formation of gastric lesions (44).

In conclusion, the risk of upper GI bleeding/perforation varied between individual NSAIDs at the doses commonly used in the general population. Despite the fact that upper GI bleeding/perforation risk is multifactorial, we showed that drugs with a long half- life or slow-release formulation and/or associated with profound and coincident inhibition of both COX isozymes—which translates into deficiency of prostanoid generation in the GI tract—were associated with a greater risk of upper GI bleeding/perforation. The findings from all of these epidemiologic studies, though still open to some residual confounding, have robust mechanistic support from results obtained in experimental animals.

AUTHOR CONTRIBUTIONS

All authors were involved in drafting the article or revising it critically for important intellectual content, and all authors approved the final version to be published. Dr. García Rodríguez had full access to all of the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Study conception and design. Patrignani, García Rodríguez.

Acquisition of data. Massó González, Patrignani, Tacconelli, García Rodríguez.

Analysis and interpretation of data. Massó González, Patrignani, García Rodríguez.

Ancillary