Celecoxib

Two systematic reviews included in the original CER evaluated the efficacy of celecoxib versus nonselective NSAIDs.^{50, 51} We identified two fair-quality head-to-head trials of celecoxib versus diclofenac (n=925 and n=249) published since the original CER (Appendix H).^{52, 53}

A good-quality systematic review (published in 2002) funded by the makers of celecoxib found similar effects on Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC) scores associated with celecoxib and nonselective NSAIDs based on data from published and unpublished randomized trials of at least 12 weeks’ duration in patients with either osteoarthritis (OA) or rheumatoid arthritis (RA).⁵⁰ A more recent systematic review (published in 2005) with access to all unpublished manufacturer-held clinical trial reports found celecoxib at doses of 200–400 mg associated with slightly higher rates of withdrawals due to lack of efficacy compared to nonselective NSAIDs (RR 1.1; 95% CI 1.0, 1.2), based on data from 31 primarily short-term (≤12 weeks) trials.⁵¹

The two largest head-to-head trials of celecoxib versus nonselective NSAIDs are the Celecoxib Long-term Arthritis Safety Study (CLASS)⁵⁴ and the Successive Celecoxib Efficacy and Safety Study-1 (SUCCESS-1).⁵⁵ Both systematic reviews included CLASS (n=7,968), a pivotal, long-term (6 to 13 months) trial of celecoxib versus the nonselective NSAIDs ibuprofen or diclofenac for rheumatoid and osteoarthritis.⁵⁴ The nonselective NSAIDs were associated with a slightly higher (but statistically significant) likelihood of withdrawal due to lack of efficacy compared to celecoxib (15% vs. 13%, p=0.005). CLASS focused on assessment of adverse events rather than efficacy, and other efficacy results were not reported. The Moore et al. systematic review⁵¹ included the large (n=13,274), Successive Celecoxib Efficacy and Safety Study (SUCCESS-1), which found no clinically meaningful (and mostly statistically nonsignificant) differences after 12 weeks in efficacy (pain, global assessment of arthritis, or WOMAC total score) between celecoxib 100 mg or 200 mg twice daily and the nonselective NSAIDs diclofenac and naproxen in patients with osteoarthritis.⁵⁵ Withdrawals due to lack of efficacy were not reported.

A new, fair-quality trial (high loss to followup) found no differences between celecoxib 200 mg once daily and diclofenac 50 mg twice daily in pain scores, global assessment of arthritis, or patient satisfaction through 52 weeks of followup in older (≥60 years) patients (n=925) with osteoarthritis.⁵² Withdrawals due to adverse events were slightly less frequent with celecoxib compared with diclofenac, but the difference was not statistically significant (27% vs. 31%, RR 0.87, 95% CI 0.71–1.1). Another new, fair-quality trial (high loss to followup, allocation concealment method not described, failure to report intention-to-treat analysis) reported results inconsistent with other trials.⁵³ It failed to demonstrate noninferiority of celecoxib 200 mg compared to diclofenac 50 mg three times daily on pain at 6 weeks (mean difference between drugs in change from baseline 12 mm on a 0 to 100 visual analogue scale (VAS), 95% CI 5.8 to 18) or 12 weeks (10 mm, 95% CI 2.8 to 17) in patients with hip osteoarthritis requiring joint replacement surgery. Withdrawals due to lack of efficacy were similar (13 percent vs. 11 percent).

Partially Selective NSAIDs

Three systematic reviews included in the original CER evaluated the efficacy of the partially selective NSAIDs etodolac or nabumetone versus nonselective NSAIDs.^{9, 56, 57} One new, good-quality systematic review evaluated comparative efficacy of the partially selective NSAIDs for osteoarthritis or rheumatoid arthritis (Appendix H).⁵⁸ We identified no new head-to-head trials of partially selective NSAIDs versus nonselective NSAIDs published since the original CER.

Eleven randomized, double-blinded trials of meloxicam 7.5 mg, 15 mg, or 25 mg versus other NSAIDs for osteoarthritis found no clear or consistent differences in efficacy.^59–69 In two of the trials, meloxicam was associated with a greater likelihood of withdrawal due to lack of efficacy than nonselective NSAIDs.^{63, 68} The new systematic review, which included trials of patients with osteoarthritis or rheumatoid arthritis, found meloxicam associated with lower efficacy compared to nonselective NSAIDs for pain (difference 1.7 points on a 10-point VAS pain scale, 95% CI 0.8 to 2.7) and withdrawals due to lack of efficacy (RR 1.5, 95% CI 1.2 to 1.7).⁵⁸

The original CER included several good-quality Cochrane systematic reviews of randomized trials that found no difference between etodolac and various nonselective NSAIDs for OA of the hip (trials published through 1994),⁷⁰ back (through 1998),⁹ or knee (through 1997).⁵⁷ In seven trials published after or not included in the Cochrane reviews, there were also no differences between sustained-release etodolac and diclofenac⁷¹ or tenoxicam;⁷² or between standard-formulation etodolac and piroxicam (2 trials^{73, 74}), naproxen (2 trials^{75, 76}), or nimesulide⁷⁷ for OA of the knee, hip, or foot. The new systematic review found no differences between etodolac and various nonselective NSAIDs for pain (mean difference 2.1, 95% CI −2.1 to 6.2) or withdrawals due to lack of efficacy (RR 1.0, 95% CI 0.85 to 1.2) in patients with osteoarthritis or rheumatoid arthritis.⁵⁸

The Cochrane review of NSAIDs for knee osteoarthritis found nabumetone similar in efficacy to the nonselective NSAIDs diclofenac SR⁷⁸ and etodolac⁷⁹ in two 4-week trials.⁵⁷

Nonselective NSAIDs

The original CER included several good-quality systematic reviews by the Cochrane Collaboration of trials that compared various nonselective NSAIDs for OA of the hip (trials published through 1994),⁷⁰ back (through 1998),⁹ or knee (through 1997).⁵⁷ These reviews found no clear differences in efficacy between non-aspirin, primarily nonselective NSAIDs. We identified no new head-to-head trials comparing efficacy of one non-aspirin, nonselective NSAID versus another. The large SUCCESS-1 trial included diclofenac and naproxen arms, but only reported combined efficacy results for these two nonselective NSAIDs.⁵⁵

Aspirin or Salsalate

We identified no new head-to-head trials comparing efficacy of aspirin or salsalate versus other NSAIDs. A head-to-head trial included in the original CER found salsalate 3 g once daily and aspirin 3.6 g once daily associated with similar efficacy in patients with OA after 2 weeks of treatment.⁸⁰

Randomized Controlled Trials

Celecoxib: GI Harms

One systematic review of randomized trials of serious GI harms associated with celecoxib versus nonselective NSAIDs was included in the original CER (Appendix H).⁵¹ We included another fair-quality systematic review that only had preliminary results available at the time of the original CER (Appendix H).⁸¹ We identified one new pooled analysis of three similarly designed, 12-week trials of celecoxib versus diclofenac⁸² and one other new head-to-head trial of celecoxib versus diclofenac,⁵² but they either did not report serious GI events⁸² or reported too few events (two GI ulcers in nearly 1,000 patients)⁵² to affect the conclusions of the systematic reviews.

The systematic reviews both included the pivotal CLASS trials (n=7,968),⁵⁴ which compared the risk of serious GI harms associated with celecoxib versus nonselective NSAIDs for osteoarthritis or rheumatoid arthritis. CLASS was designed as two trials with separate patient recruitment and randomization procedures: one compared celecoxib 400 mg twice a day with ibuprofen 800 mg three times a day and the other compared celecoxib 400 mg twice a day with diclofenac 75 mg twice a day. The prespecified primary outcome was ulcer-related complications, defined as gastric or duodenal perforation, gastric outlet obstruction, or upper GI bleeding (POBs).⁸³ Another prespecified outcome was ulcer related complications plus symptomatic ulcers (PUBs). The planned maximum duration of the trials were 15 and 12 months, respectively, or until at least 20 ulcer-related complications occurred in each trial, or 45 in both trials combined.⁸⁴ The prespecified criteria to conclude superiority of celecoxib was statistically significant differences between celecoxib and each of the comparators, as well as between celecoxib versus the comparator groups combined.

CLASS was stopped early after reaching a predefined threshold of ulcer complications. The main publication in the Journal of the American Medical Association (JAMA) reported 6-month results even though the median duration of followup was 9 months (the rationale for reporting truncated data was high attrition), and combined the ibuprofen and diclofenac results without reporting the results of the two trials separately.⁵⁴ Additional details of the study were subsequently made public on the Food and Drug Administration (FDA) Web site.⁸⁴

CLASS randomized 3,987 subjects to celecoxib and 3,981 subjects to nonselective NSAIDs. The JAMA article reported celecoxib associated with fewer PUBs (a secondary outcome) compared to the combined nonselective NSAIDs (32/3,987 vs. 51/3,981, annualized incidence rates 2.1% vs. 3.5%, p=0.02),⁵⁴ while the rates of POBs (the primary outcome) were not significantly different (13/3,987 vs. 22/3,981, annualized incidence rates 0.76% vs. 1.4%, p=0.09). By 12 months, according to FDA documents (see Table 13, FDA Medical Officer Review)⁸⁴ there was no longer a trend favoring celecoxib for POBs (17/3987 [0.43%] events with celecoxib vs. 21/3,981 [0.53%] with the nonselective NSAIDs,⁸⁴ relative risk 1.1, 95% CI 0.47 to 2.6^{85, 86}, also see Figure 4, Scheiman review⁸⁷). For the individual comparisons between celecoxib and ibuprofen or diclofenac, which were not reported in the JAMA article, there was no difference in the rate of ulcer complications at either 6 months or the end of followup.⁸⁵ For the secondary outcome of PUBs, celecoxib was superior to ibuprofen, but not to diclofenac at 6 months and the end of followup.⁸⁵ Celecoxib was also associated with a lower risk of hemoglobin (>2 g/dL) and/or hematocrit drops (≥0.10), among all patients (2.4% vs. 4.4% and 5.7% for celecoxib, diclofenac, and ibuprofen, respectively.⁸⁴

Table 13

Incidence of hypertension in the Nurses’ Health Study and Physicians’ Health Study according to use of acetaminophen or NSAIDs.

About 20 percent of the patients in the CLASS trial took aspirin in addition to their study NSAID. When patients taking aspirin were excluded from the analysis, there were fewer confirmed serious ulcer complications in the celecoxib group than in the ibuprofen group (p=0.03).^{84, 85} However, serious ulcer complications were equivalent for celecoxib and diclofenac after exclusions of patients taking aspirin.

The new, fair-quality, nonmanufacturer-funded systematic review found celecoxib associated with a lower risk of POBs compared to nonselective NSAIDs (3 trials, RR 0.23, 95% CI 0.07 to 0.76) as well as a lower risk of PUBs (4 trials, RR 0.39, 95% CI 0.21–0.73).⁸¹ Use of 12-month instead of 6-month CLASS data did not significantly alter the pooled estimates. The systematic review also found selective COX-2 inhibitors as a class associated with lower risk of GI adverse events and withdrawal due to GI adverse events compared to nonselective NSAIDs, but did not report separate analyses for celecoxib.

The largest study in the Rostom et al. review was a manufacturer-funded combined analysis by Goldstein et al. of 14 randomized controlled trials (RCTs) of celecoxib (not including CLASS) versus placebo or nonselective NSAIDs (usually naproxen).⁸⁸ The trials ranged in duration from 2 to 24 weeks, with most lasting 6 or 12 weeks. The definition of ulcer complications (POBs) was similar to the one used in CLASS, and in all trials a blinded Safety Committee adjudicated potential ulcer complications. Not all of the included trials have been published, and their quality was not assessed by Goldstein et al. In addition, data were pooled across trials without regard to randomization, duration of therapy, or which comparator NSAID was evaluated. In the 14 trials, there were 2 POBs among 6,376 patients in the celecoxib group (3 per 10,000) and 9 among 2,768 in the NSAIDs group (33 per 10,000). This corresponded to annual rates of 2 per 1,000 patient-years for celecoxib and about 17 per 1,000 patient-years for NSAIDs (p=0.002). Rostom et al. found that excluding this study eliminated heterogeneity from the pooled analyses, but celecoxib was still associated with a lower risk of POBs (RR 0.42, 95% CI 0.22 to 0.80) and PUBs (RR 0.34, 95% CI 0.22 to 0.80) compared to nonselective NSAIDs.⁸¹

A systematic review by Moore et al. included in the original CER was funded by Pfizer and the Oxford Pain Relief Trust.⁵¹ The authors obtained a declaration from Pfizer that they had received information on all completed clinical trials of celecoxib and could publish whatever results they found, but much of the data on which this meta-analysis was based is not publicly accessible. Thus, although the meta-analysis methods appeared appropriate, it is impossible to verify the reproducibility of the meta-analysis. Rather than including the pooled analysis by Goldstein et al., ⁸⁸ Moore et al. appeared to have access to the individual trial methods and data.

All 18 trials of celecoxib versus nonselective NSAIDs included in the systematic review were rated 5 out of 5 on the Jadad quality scale, and 16 out of 16 on an 8-item validity scale.⁵¹ Only 2 of the 31 trials were longer than 12 weeks in duration. Although POBs was not evaluated as an outcome, celecoxib was associated with a lower risk of clinical ulcers and bleeds than nonselective NSAIDs in 18 trials (RR 0.61, 95% CI 0.46 to 0.81). When the analysis was limited to trials evaluating doses of 200 or 400 mg daily of celecoxib (excluding CLASS), the benefit was more pronounced (RR 0.35, 95% CI 0.22 to 0.56). The meta-analysis also found celecoxib associated with a lower risk of hemoglobin fall of 20 g/L or more (RR 0.72, 95% CI 0.56 to 0.92) and hematocrit fall of 5% or more (RR 0.78, 95% CI 0.69 to 0.89) compared with nonselective NSAIDs.⁵¹

In addition to having access to the individual trials included in Goldstein et al., another difference between the systematic review by Moore et al. and the one by Rostom et al. is that the latter did not include results of SUCCESS-1, the largest (N=13,274) randomized controlled trial of celecoxib.⁵⁵ SUCCESS-1 found celecoxib associated with a lower risk of POBs than naproxen or diclofenac after 12 weeks in patients with osteoarthritis (0.1% vs. 0.8%, odds ratio [OR] 0.14, 95% CI 0.03 to 0.68). Post hoc analysis of nonaspirin users found nonselective NSAIDs associated with a significantly higher risk of ulcer complications compared to celecoxib, though the estimate was very imprecise (OR 12, 95% CI 1.4 to 100).⁵⁵

There are several possible reasons why the results of the systematic reviews differed from those of CLASS, which did not clearly show a decreased risk of POBs for celecoxib compared to nonselective NSAIDs. First, the incidence of POBs in CLASS was relatively high.⁵⁴ In the CLASS trials, the annualized rate of POBs was 0.8/100 patient-years for celecoxib and 1.4 per 100 patient-years for nonselective NSAIDs,⁵⁴ compared to 0.1/100 patient-years and 0.8/100 patient-years, respectively, in SUCCESS-1.⁵⁵ The high rate of POBs in the CLASS trials could be due in part to enrollment of a higher-risk population, the use of concomitant medications, or other factors. In CLASS, 20 percent of patients randomized to celecoxib were on aspirin and 31 percent on corticosteroids,⁵⁴ whereas in SUCCESS-1, 7 percent were on aspirin and corticosteroid use was not permitted.⁵⁵ In addition, antiulcer medications (except for occasional antacids) were prohibited in CLASS, but used in 16 percent of celecoxib patients in the Goldstein et al. combined analysis.⁸⁸ Another potential explanatory factor is that the high dose of celecoxib used in CLASS—400 mg twice daily—was evaluated in few other trials, and could be associated with an increased risk of bleeding compared to lower doses. Finally, different comparator NSAIDs could be associated with different risks of GI complications. Pooling data from trials evaluating different comparator NSAIDs could obscure differential effects on GI safety if they were present.

Observational Studies

One new systematic review⁹⁷ and five systematic reviews^{10, 98–101} included in the original CER evaluated serious GI harms associated with various NSAIDs.

The new, fair-quality (did not assess quality of included studies) systematic review (by Massó González et al.) found celecoxib associated with an increased risk of upper GI bleeding or perforation compared to nonuse (four studies, RR 1.4, 95% CI 0.85 to 2.4), but the risk was lower than for nonselective NSAIDs as a group (eight studies, RR 4.5, 95% CI 3.8 to 5.3) as well as for individual nonselective NSAIDs, though confidence interval estimates overlapped in some cases (Table 3, Appendix H).⁹⁷

Table 3

Risk of upper gastrointestinal bleeding or perforation with use of an NSAID compared with nonuse of NSAIDs, systematic review of observational studies.

Meta-analyses of observational studies included in the original CER reported similar findings. In a collaborative meta-analysis of cohort and case-control studies published between 1985 and 1994, use of all nonselective NSAIDs were associated with significantly increased risks of peptic ulcer complication hospitalizations relative to nonuse.¹⁰⁰ As in the Massó González et al. review, ibuprofen was associated with the lowest risk of peptic ulcer complication-related hospitalizations compared to other nonselective NSAIDs.⁹⁵ In two other meta-analyses of cohort and case-control studies published between 1990 and 1999, however, risk of upper GI bleeds was no lower for ibuprofen compared to any other non-aspirin, nonselective NSAID when results were stratified by low to medium (RR 2.1 vs. nonuse, 95% CI 1.6 to 2.7) or high dose (RR 5.5 vs. nonuse, 95% CI 3.0 to 10) (Table 4).^{98, 101} A systematic review of observational studies published through 2002 also found GI bleeding risk increased for all nonselective NSAIDs, with risk appearing related more to dose than to the specific drug evaluated.¹⁰

Table 4

Serious gastrointestinal events in observational studies.

Eight large case-control (>1,000 cases) or cohort (n>100,000) studies reported risks of serious upper GI complications associated with various NSAIDs (Table 4, Appendix H).^{98, 102–108} Two of the studies were published after the original CER,^{102, 108} and all but two were included in the new systematic review.^{103, 108} Three studies used a cohort design^106–108 and the remainder used a case-control (or nested case-control) design. Two case-control studies were rated good quality^{102, 104} and the remainder of the observational studies rated fair quality (Appendix G). The most common methodological shortcomings in the fair-quality case-control studies were failure to report the proportion of patients who met inclusion criteria who were excluded from the study and unclear accuracy of methods used to ascertain exposures and potential confounders. The most common methodological shortcomings in the fair-quality cohort studies were noncomparability of groups at baseline, unclear blinding status of outcomes assessors and data analysts, and failure to report attrition from a defined inception cohort. Four of the observational studies found celecoxib associated with an no increased risk of upper GI complications compared to nonuse^{103, 104, 106} or acetaminophen use.¹⁰⁸ A fifth study found celecoxib associated with an increased risk of upper GI perforation or bleeding compared to nonuse, but risk estimates were similar or lower than those for nonselective NSAIDs.¹⁰²

The partially selective NSAID meloxicam was evaluated in four of the large observational studies.^{98, 102, 104, 105} Meloxicam was associated with a risk of upper GI bleeding relative to nonuse of NSAIDs that was generally in the midrange of risks reported for various nonselective NSAIDs. Only one study reported risks associated with other partially selective NSAIDs, and estimates were imprecise.⁹⁸

For various nonselective NSAIDs, the observational studies generally showed increased risk of GI bleeding relative to nonuse.^{98, 102, 103, 105–107} Naproxen was associated with a higher risk than ibuprofen in seven studies,^{98, 102–105, 107, 108} though the risk estimates were relatively close in two of them.^{103, 107} Comparative data for other nonselective NSAIDs was less consistent. For example, diclofenac was associated with similar or lower risk compared to ibuprofen in three studies, ^{103, 104, 108} but higher in four others.^{98, 102, 105, 107}

The risk of upper GI bleeding was similar with aspirin compared to non-aspirin, nonselective NSAIDs in one large nested case-control study.¹⁰³ Systematic reviews of observational studies included in the original CER found that aspirin increases risk of serious GI events relative to placebo or nonuse, at a rate similar to that of other nonselective NSAIDs.^{99, 100}

Serious GI event rates (bleeding, perforation, obstruction) associated with salsalate were reported in one smaller cohort study (n=1,198) of long-term care residents in Indiana.¹⁰⁹ The number of cases of GI-related hospitalizations associated with salsalate (1, 5.9 percent) after 14 months was similar to that of other selective and nonselective NSAIDs.

Randomized Controlled Trials

Celecoxib

Four systematic reviews or meta-analyses included in the original CER (one as an earlier version available only as an FDA briefing document¹¹⁰) evaluated risk of serious CV events in randomized controlled trials of celecoxib (Table 5, Appendix H).^{51, 111–113} Two new systematic reviews were identified for this update (Table 5, Appendix H).^{114, 115} We identified one new placebo-controlled Chinese trial of celecoxib for prevention of gastric cancer that reported serious CV events,¹¹⁶ and one head-to-head trial of celecoxib versus diclofenac for osteoarthritis.⁵²

Table 5

Meta-analyses of serious cardiovascular events in trials of celecoxib.

The systematic reviews all included CLASS.⁵⁴ Six-month data from CLASS showed no association between celecoxib and risk of myocardial infarction or any CV event (stroke, myocardial infarction, or angina) compared with the nonselective NSAIDs (myocardial infarctions 0.3% [10/3987] vs. 0.3% [11/3981]).⁵⁴ A subsequent analysis based on complete followup data also showed no differences in the rates of any significant CV event for the overall sample (0.5% [19/3987] vs. 0.3% [13/3981]) or for the subgroup who did not use aspirin.¹¹⁷ Approximately 2,770 subjects in CLASS (about one-third of the sample) had at least 9 months of followup, and 1,126 had at least 12 months of followup.

Three systematic reviews provided the best information on CV risks associated with long-term use of celecoxib.^{111, 114, 115} All included preliminary or published results from trials of celecoxib for prevention of colon polyps or Alzheimer’s disease (Adenoma Prevention with Celecoxib trial [APC], Alzheimer’s Disease Anti-Inflammatory Prevention Trial [ADAPT], Prevention of Colorectal Sporadic Adenomatous Polyps [PreSAP]). Two systematic reviews were rated fair-quality due to failure to adequately assess trial quality^{111, 114} or report statistical heterogeneity.¹¹⁴ The third systematic review was rated good quality.¹¹⁵ All of the meta-analyses excluded a number of short-term trials,^{111, 114, 115} one of the meta-analyses excluded trials that did not have at least two arms with at least 100 patient years of followup,¹¹⁵ and one of the meta-analyses¹¹¹ excluded trials without publicly available information on CV events. Although excluding short-term trials limited conclusions regarding short-term risks, data on long-term harms may be more relevant for patients using NSAIDs for chronic conditions such as osteoarthritis.

One of the two systematic reviews was a new study which limited inclusion to randomized, double-blind, placebo-controlled trials with planned followup of at least 3 years.¹¹⁴ It included 6 trials (3,664 people randomized to celecoxib), none of which evaluated patients with osteoarthritis. Three trials evaluated celecoxib for colon polyp prevention (APC, PreSAP, and the Celecoxib/Selenium trial), one for prevention of Alzheimer’s disease (ADAPT), one for prevention of recurrent breast cancer (MA27), and one for treatment of retinopathy (CDME). Relative to placebo, the overall risk of a CV event (CV death, myocardial infarction, stroke, heart failure, or a thromboembolic event) in patients randomized to celecoxib at any dose was increased (hazard ratio [HR] 1.6, 95% CI 1.1 to 2.3). The absolute difference in risk of a CV event was 3.7/1000 patient-years (11.2/1000 patient-years with celecoxib vs. 7.5/1000 patient-years with placebo), or 1 additional CV event for about every 270 patients treated with celecoxib instead of placebo for 1 year. However, the risk appeared to vary at different doses, and was lowest for celecoxib 400 mg once daily (HR 1.1, 95% CI 0.6 to 2.0), intermediate for celecoxib 200 mg twice daily (HR 1.8, 95% CI 1.1 to 3.1), and highest for celecoxib 400 mg twice daily (HR 3.1, 95% CI 1.5 to 6.1). In subgroup analyses, patients at higher baseline risk were at disproportionately increased risk of CV events compared to those at lower baseline risk (p-value for interaction 0.003).

The second systematic review, which was also a new study, limited inclusion to trials with at least 100 patient years of followup and performed a network analysis to incorporate indirect evidence into pooled estimates.¹¹⁵ It included 31 trials of various NSAIDs versus placebo or other NSAIDs, with 6 trials of celecoxib versus placebo (12,799 patient years), including ADAPT, APC, and PreSAP (these three trials accounted for 6,801 patient-years of celecoxib exposure). It found celecoxib associated with a nonstatistically significant trend toward increased risk of myocardial infarction (RR 1.4, 95% CI 0.7 to 2.7) and composite cardiovascular events (nonfatal myocardial infarction, nonfatal stroke, or cardiovascular death, RR 1.4, 95% CI 0.94 to 2.2). Results were insensitive to a variety of sensitivity analyses based on methodological factors (such as use of independent adjudication of harms) or dose. There was no difference in risk of myocardial infarction between celecoxib and naproxen, ibuprofen, or diclofenac.

The third systematic review, which was included in the original CER, limited its analysis to trials that were at least 6 weeks in duration and reported CV events in published articles or publicly available material.¹¹¹ It found the risk of myocardial infarction increased in three trials (APC, ADAPT, PreSAP; none evaluated arthritis patients) that compared celecoxib to placebo (OR 2.3, 95% CI 1.0 to 5.1) and in five trials (APC, CLASS, ADAPT, PreSAP, VACT; the latter two evaluated arthritis patients) that compared celecoxib to placebo, diclofenac, ibuprofen, or paracetamol (OR 1.9, 95% CI 1.2 to 3.1). No heterogeneity was present. There was no association between celecoxib use and either cerebrovascular events, CV death, or composite CV events. The meta-analysis did not include the large (N=13,274), 12-week SUCCESS-I Study, which reported results consistent with its findings (10 myocardial infarctions or 0.55/100 patient-years in the combined celecoxib arms versus 1 myocardial infarction or 0.11/100 patient-years in the combined nonselective NSAID arms).⁵⁵

Neither of the systematic reviews included a new, fair-quality head-to-head trial (n=916) that found no difference in risk of myocardial infarction after 1 year in 916 patients randomized to celecoxib versus diclofenac for osteoarthritis (0.9% vs. 1.3%, RR 0.67, 95% CI 0.19 to 2.35),⁵² or a new, fair-quality Chinese trial (n=1,024) that found no difference in risk of CV events (defined as fatal or nonfatal myocardial infarction, and ischemic or hemorrhagic stroke) between celecoxib 200 mg twice daily and placebo after 1.5 years in patients at high risk for gastric cancer (0.86% vs. 1.1%, OR 0.84, 95% CI 0.23 to 3.2).¹¹⁶ In both trials, the number of events was small (9 or 10 total), and it was unclear if myocardial infarctions were subject to blinded adjudication.

Three meta-analyses included in the original CER found no increased risk of serious CV events with celecoxib versus placebo.^{51, 112, 113} However, these meta-analyses did not include trials completed after 2004, including two large, long-term trials of colon polyp prevention (APC and PreSAP).^{118, 119} These two trials account for a high proportion of the myocardial infarctions in the celecoxib trials (70 events in persons randomized to celecoxib, compared with 31 in one of the meta-analyses¹¹³). The pooled relative risk from these trials for celecoxib versus placebo was 1.9 (95% CI 1.1 to 3.1, no heterogeneity) for the composite outcome of CV death, nonfatal myocardial infarction, nonfatal stroke, or heart failure.¹²⁰ Rates of fatal or nonfatal myocardial infarction were 1.6 percent (22/1356) versus 0.4 percent (3/679) in the APC trial and 9/933 (1.0 percent) versus 4/628 (0.6 percent) in PreSAP. The meta-analyses also focused almost exclusively on short-term trials, with the proportion 12 weeks or shorter in duration ranging from 87 percent to 94 percent.^{51, 112, 113} In addition, two of the meta-analyses were rated poor quality, in part due to failure to assess study quality and because they pooled raw event rates for a particular drug and dose across studies,^{112, 113} resulting in loss of randomization effects, and making it impossible to evaluate heterogeneity across studies.

A meta-analysis¹²¹ that was included in the original report was excluded from this section because it pooled risks of different COX-2 selective NSAIDs together. Based on published and unpublished data from 121 RCTs, including the polyp prevention trials previously mentioned, the relative risk for any vascular event with COX-2 selective NSAIDs as a class compared to placebo was 1.4 (95% CI 1.1 to 1.8). Much of the association appeared to be related to an increased risk of myocardial infarction (RR 1.9, 95% CI 1.3 to 2.6), with no increased risk of stroke (RR 1.0, 95% CI 0.71 to 1.5). From 41 trials, the raw event rate for myocardial infarction in patients randomized to celecoxib was 0.5 percent (44/8976 person-years) compared to 0.2 percent (9/4953 person-years) in those randomized to placebo. Based on the forest plot presented with the meta-analysis, the point estimate for celecoxib was similar to the overall pooled estimate for all COX-2 selective NSAIDs, and just met criteria for statistical significance. A trend towards increased risk of vascular events (p=0.03) with higher doses of celecoxib was observed, but nearly all of the events at the highest (800 mg daily) dose occurred in the polyp prevention trials. Analyses on the effects of duration and independent event adjudication were not stratified by specific COX-2 inhibitor, nor were estimates of CV risk with specific COX-2 inhibitors relative to naproxen or nonnaproxen NSAIDs.

In summary, celecoxib appears to be associated with an increased risk of myocardial infarctions or thromboembolic CV events compared to placebo. Much of the evidence for increased CV risk comes from two large, long-term polyp prevention studies that compared celecoxib 200 or 400 mg twice daily, or 400 mg once daily, to placebo.

Other NSAIDs

One systematic review included in the original CER evaluated risk of serious CV events associated with nonselective NSAIDs.¹²¹ We identified one new systematic review¹¹⁵ Two trials included in the original CER and not included in the systematic review also reported serious CV events in patients prescribed naproxen.^{55, 122}

A new, good-quality systematic review by Trelle and colleagues of 31 trials (with at least two arms with at least 100 patient-years of followup) compared CV risks associated with various nonselective NSAIDs, based on a network analysis.¹¹⁵ It found ibuprofen associated with increased risk of composite cardiovascular outcomes (nonfatal myocardial infarction, nonfatal stroke, or cardiovascular death) compared with placebo (RR 2.3, 95% CI 1.1 to 4.9) and diclofenac associated with a trend towards increased risk (RR 1.6, 95% CI 0.85 to 3.0). Among the nonselective NSAIDs, with respect to specific CV outcomes, diclofenac was associated with the highest risk of stroke (RR 2.9, 95% CI 1.1 to 8.4) and cardiovascular death (RR 4.0, 95% CI 1.5 to 13). Naproxen was associated with only a slight, nonsignificant trend toward increased risk (RR 1.2, 95% CI 0.78 to 1.9). There were no statistically significant differences in risk of composite CV outcomes between naproxen, ibuprofen, and diclofenac.

A fair-quality systematic review included in the original CER by Kearney and colleagues of 91 trials (mostly ranging from 4 to 13 weeks in duration) evaluated risks associated with any nonselective NSAID (33,260 person-years of exposure) compared to any COX-2 selective NSAID (23,325 person-years of exposure).¹²¹ Most of the trials evaluated naproxen (42 trials), ibuprofen (24 trials), and diclofenac (26 trials); only 7 evaluated other nonselective NSAIDs. Generalizability to usual practice could be limited because the majority of the trials evaluated higher than standard doses of NSAIDs. Much of the data regarding CV event rates were obtained by requesting unpublished data from trial sponsors.

Table 6 shows estimates of risk for different CV outcomes with COX-2 inhibitors relative to nonselective NSAIDs. Risk of myocardial infarction was similar with COX-2 inhibitors and nonnaproxen NSAIDs, but about twofold as great for COX-2 inhibitors compared with naproxen (0.6% or 99/16360 vs. 0.3% or 30/10,978, RR 2.0, 95% CI 1.4 to 3.0). This is equivalent to about 1 additional myocardial infarction for every 300 patients treated for 1 year with a COX-2 inhibitor instead of naproxen. COX-2 inhibitor use was also associated with a lower risk of stroke relative to nonnaproxen NSAIDs (RR 0.62, 95% CI 0.41 to 0.95). In subgroup analyses of specific nonselective NSAIDs (ibuprofen, diclofenac, other nonselective NSAIDs), the difference in stroke risk was only observed with diclofenac, which was usually evaluated at high doses (RR 0.48, 95% CI 0.27 to 0.83). There was insufficient data to analyze the effects of lower doses on estimates of risk.

Table 6

Rate ratios (95% CI): COX-2 inhibitor relative to nonselective NSAID.

Kearney et al. found insufficient data to directly estimate risks of nonselective NSAIDs from placebo-controlled trials. Indirect analyses (based on trials of nonselective NSAIDs versus COX-2 inhibitors and trials of COX-2 inhibitors vs. placebo) suggested an increased risk of vascular events with ibuprofen (RR 1.5, 95% CI 0.96 to 2.4) and diclofenac (RR 1.6, 95% CI 1.1 to 2.4) relative to placebo, but not with naproxen (RR 0.92, 95% CI 0.67 to 1.3). However, indirect analyses should be interpreted with caution because they can give discrepant results compared to head-to-head comparisons.¹²³

The Kearney meta-analysis did not include results of the large SUCCESS-1 trial, which reported 0.61 MIs/100 patient-years with naproxen (n=905), and no cases of MI in diclofenac users (n=3489).⁵⁵ It also didn’t include the Alzheimer’s Disease Anti-Inflammatory Prevention Trial (ADAPT), which was terminated early in December 2004 because of an “apparent increase in CV and cerebrovascular events among the participants taking naproxen when compared with those on placebo.”¹²⁴ Results from ADAPT showed a nonsignificant increased in risk of CV deaths (HR 1.5, 95% CI 0.30 to 7.3), myocardial infarction (HR 1.5, 95% CI 0.69 to 3.2), or stroke (HR 2.1, 95% CI 0.81 to 5.6).¹²² Naproxen was associated with an increased risk based on the composite outcome of CV death, myocardial infarction, stroke, congestive heart failure, or transient ischemic attack (HR 1.6, 95% CI 1.0 to 2.6). The decision to terminate ADAPT has been criticized because rigorous stopping protocols were not used, the increased risk associated with naproxen for individual and most composite CV outcomes did not reach statistical significance, the events were not adjudicated, and the number of events was small.¹²⁵

Aspirin and Salsalate

Aspirin is known to be protective against occlusive vascular events because of its irreversible antiplatelet effects. In a collaborative meta-analysis of infidel patient data from 22 randomized trials (over 110,000 participants), lower doses of aspirin (primarily less than 325 mg daily) were associated with decreased risk of serious vascular events when given for primary prevention (0.51% aspirin vs. 0.57% control per year, p=0.0001) or secondary prevention (6.7% vs. 8.2%, p<0.0001).⁹⁶ The populations evaluated in these trials probably varied substantially from trials of patients with arthritis.

Observational Studies

Three systematic reviews evaluated CV risk associated with various NSAIDs.^126–128 Two were included in the original CER and focused on risks associated with naproxen.^{126, 127} The third was a new, good-quality systematic review of CV risk (primarily myocardial infarction) from 23 observational studies that was published too late to be included in the original CER, though results were summarized in a brief addendum (Appendix H).¹²⁸ We also identified four large observational studies not included in the original CER.^{108, 129–131}

The new systematic review included a total of 23 observational studies (16 case-control and 7 cohort studies).¹²⁸ It found diclofenac associated with the highest risk, followed by indomethacin and meloxicam (Table 7). Celecoxib, naproxen, piroxicam, and ibuprofen were not associated with increased risks. For all NSAIDs, increases in risk were modest (RR <1.5), and all of the main analyses were characterized by substantial between-study heterogeneity.

Table 7

Rate ratios for cardiovascular events (95% CI)*: NSAID use compared with nonuse of NSAIDs.

Nineteen large observational studies (case-control studies with >1000 cases or cohort studies with >100,000 subjects) evaluated risk of CV events associated with various NSAIDs (Table 8, Appendix H).^{108, 129–146} All of these studies except for four^{108, 129–131} were included in the original CER. Seven studies^{108, 129–132, 141, 146} not included in the systematic review of observational studies.¹²⁸ Six studies used a cohort design^{108, 130, 131, 140, 146, 147} and the remainder a case-control (or nested case-control) design. Three studies evaluated the UK General Practice Research Database^{133, 134, 143} and three evaluated the same Canadian (Quebec) database.^{108, 130, 141} Only one study was rated good quality,¹³⁹ the remainder were rated fair quality. The most common methodological shortcoming in the fair-quality case-control studies was failure to report the proportion of patients who met inclusion criteria who were excluded from the study. The most common methodological shortcomings in the fair-quality cohort studies were unclear blinding status of outcomes assessors and data analysts, and failure to report attrition from a defined inception cohort. Interpretation of the studies was complicated by the use of different study designs, adjustment for different numbers and types of confounders, and evaluation of different populations and outcomes.

Table 8

Cardiovascular events in observational studies.

Sixteen observational studies evaluated risk of serious CV events (primarily myocardial infarction) associated with celecoxib.^{108, 129–132, 135–140, 145–149} Three studies found celecoxib associated with similar risk of CV events compared to naproxen, ibuprofen, or diclofenac.^{130, 145, 146} A fourth study found ibuprofen (OR 1.3, 95% CI 1.0 to 1.6) and naproxen (OR 1.4, 95% CI 1.1 to 1.8) associated with a higher risk of acute MI requiring admission or sudden cardiac death than celecoxib.¹³⁵ Twelve studies found no increased risk of serious CV events with celecoxib relative to nonuse of NSAIDs.^{108, 129, 131, 135, 136, 138–140, 145–148} Three studies found current (RR 1.6, 95% CI 1.2 to 2.0),¹³² new (RR 2.1, 95% CI 1.4 to 3.1),¹³⁷ or any (HR 1.5, 95% CI 0.99 to 2.2)¹⁴⁹ use of celecoxib associated with increased risk compared to nonuse of NSAIDs. In these studies, the increased MI risk was either time-¹³⁷ or dose-dependent.¹³²

The nonselective NSAID naproxen has received additional scrutiny since the VIGOR trial¹⁷ showed an increased risk of CV events with rofecoxib versus naproxen, due to the hypothesis that naproxen might be protective against myocardial infarction. In addition, a systematic review¹²¹ of randomized trials (described earlier) found that naproxen was not associated with the same increased in CV risk as other nonselective and selective NSAIDs. In addition to the new systematic review of observational studies described above (which found a neutral effect of naproxen on CV risk),¹²⁸ two systematic reviews included in the original CER specifically focused on CV risks associated with naproxen.^{126, 127} The first, a meta-analysis of 11 observational studies of naproxen (four based on the General Practice Research Database) found naproxen associated with a small cardioprotective effect (OR 0.86, 95% CI 0.75 to 0.99), with Merck-funded studies reporting larger effect sizes.¹²⁷ Nine observational studies published after this systematic review showed no cardioprotective effect associated with naproxen,^{108, 129, 132, 135–137, 139, 148, 149} though one other study showed a modest protective effect (HR 0.79, 95% CI 0.67 to 0.93).¹³¹ An FDA review included in the original CER concluded no cardioprotective effect of naproxen after taking into account various methodological issues.¹²⁶

Large observational studies found no other nonselective NSAID consistently associated with increased risk of CV events compared to nonuse of NSAIDs.^{129, 131–137, 139, 140, 142–144} For example, ibuprofen was associated with a modest increased risk (OR 1.4, 95% CI 1.3 to 1.6 and OR 1.2, 95% CI 1.1 to 1.4) of serious CV events compared to nonuse of NSAIDs in two^{129, 136} studies, but no increased risk in nine others.^{131–135, 143, 144, 147, 149}

Partially selective NSAIDs have not been well studied in large observational studies. Three studies found no increased risk of serious CV events with meloxicam compared to nonuse.^{129, 134, 139} One study found no increased risk of acute myocardial infarction with use of etodolac or nabumetone versus nonuse of NSAIDs, but estimates were imprecise.¹³³

In April 2005, after reviewing the available observational data, the FDA issued a Public Health Advisory stating, “Long-term controlled clinical trials have not been conducted with most of these (nonselective) NSAIDs. However, the available data suggest that use of these drugs may increase CV risk. It is very difficult to draw conclusions about the relative CV risk among the COX-2 selective and nonselective NSAIDs with the data available. All sponsors of nonselective NSAIDs will be asked to conduct and submit to FDA a comprehensive review and analysis of available controlled clinical trial databases pertaining to their NSAID product(s) to which they have access to further evaluate the potential for increased CV risk.”¹⁵⁰ The FDA also required labeling changes to both prescription and nonprescription nonselective NSAIDs warning about potential CV risks.

Mortality

We identified no new studies evaluating mortality associated with different NSAIDs. Large clinical trials included in the original CER did not show differences in mortality between different NSAIDs.^{54, 152} In CLASS, mortality rates were 0.47%, 0.37%, and 0.45% for celecoxib, diclofenac, and ibuprofen, respectively.⁸⁴ In SUCCESS-1, 5 deaths (0.06 percent) were observed after 12 weeks in the celecoxib group and 5 (0.11 percent) in the nonselective NSAIDs group.⁵⁵ A meta-analysis that included unpublished company clinical trial data (including CLASS and SUCCESS-1) found no significant difference in rates of death in patients randomized to celecoxib compared with nonselective NSAIDs, though there were few events (0.03% or 6/18,325 in the celecoxib arms vs. 0.11% or 14/12,685 in the NSAID arms).⁵¹

One retrospective cohort study of Saskatchewan health-services databases that followed patients from 6 months following prescription until death found nabumetone associated with significantly lower rates of all-cause mortality compared with diclofenac (OR 2.0; 95% CI 1.2 to 3.1) and naproxen (OR 3.0, 95% CI 1.9 to 4.6).¹⁵³ However, we found no other studies that replicated this finding.

Hypertension, CHF, Edema, and Renal Function

Six systematic reviews or meta-analyses included in the original CER evaluated comparative risks of hypertension, CHF, edema, and renal function associated with various NSAIDs.^{19, 51, 110, 154–156} A seventh systematic review was published too late to be fully included in the original CER, but described in an appendix.¹⁵⁷ It was rated fair quality because it did not assess the quality of included studies. Two new observational studies evaluated risk of congestive heart failure in high risk patients.¹⁵⁸

All NSAIDs appear to be associated with increases in blood pressure. However, evidence regarding differential effects of specific NSAIDs is somewhat conflicting. One meta-analysis included in the original CER found that nonselective NSAIDs raised mean blood pressure by an average of about 5.0 mm Hg (95% CI 1.2 to 8.7).¹⁵⁴ Piroxicam produced the most marked elevation in blood pressure compared to placebo. In head-to-head trials, there were no significant differences between indomethacin and sulindac (10 trials), indomethacin and salicylate (1 trial), diclofenac and sulindac (1 trial), ibuprofen and sulindac (1 trial), and naproxen and sulindac (3 trials). Another meta-analysis found that piroxicam and ibuprofen had negligible effects on blood pressure, and that indomethacin and naproxen were associated with the largest increases.¹⁵⁵ In both meta-analyses, aspirin and sulindac were associated with minimal hypertensive affect. More than half of the published NSAID trials did not report hypertension rates as an outcome.¹⁵⁵

Several meta-analyses of celecoxib included in the original CER found no increased risk of hypertension compared to nonselective NSAIDs.^{19, 51, 110} A fair-quality meta-analysis found celecoxib (dose not specified) not associated with an increased risk of hypertension compared to either placebo (RR 0.81, 95% CI 0.13 to 5.21) or nonselective NSAIDs (RR 0.82, 95% CI 0.68 to 1.00).¹⁹ A Pfizer-funded meta-analysis submitted to the FDA found an increased risk of developing hypertension with celecoxib at any dose compared to placebo (1.1% vs. 0.7%, p=0.02), though the risk was lower than for nonselective NSAIDs (1.5% vs. 2.0%, p=0.002).¹¹⁰ A third meta-analysis, funded in part by the manufacturer, reported similar findings for risk of hypertension (celecoxib vs. nonselective NSAID, RR 1.1, 95% CI 0.90 to 1.3).⁵¹ The fourth meta-analysis, which was included as an appendix in the original CER, found celecoxib associated with slightly lower risk of hypertension (RR 0.83, 95% CI 0.71 to 0.97) compared with control treatments (placebo, other NSAID, or mixed/other).¹⁵⁷ Most of the trials included in the meta-analyses were short-term and only one meta-analysis⁵¹ evaluated the quality of the trials.

Results from large trials of celecoxib are mostly consistent with the meta-analyses. In CLASS (median duration of followup 9 months), celecoxib was associated with a similar rate of hypertension (new-onset and aggravated preexisting) compared with diclofenac (2.7 percent vs. 2.6 percent), and a lower rate compared with ibuprofen (2.7 percent vs. 4.2 percent).¹¹⁷ In the shorter-term (12 weeks) SUCCESS-I trial (N=13,274), rates of hypertension were similar with celecoxib 100 or 200 mg twice a day compared with either diclofenac or naproxen (RR 0.86, 95% CI 0.62 to 1.20).⁵⁵ The APC polyp prevention trial found celecoxib associated with greater systolic blood pressure elevations compared to placebo at 1 and 3 years at either 200 mg twice daily (2.0 mm Hg at 1 year and 2.6 mm Hg at 3 years) and 400 mg twice daily (2.9 mm Hg at 1 year and 5.2 mm Hg at 3 years).¹²⁰ On the other hand, the PreSAP polyp prevention trial found no difference in systolic blood pressure increases between celecoxib 400 mg once daily and placebo.¹²⁰

With regard to renal dysfunction, it is unclear whether COX-2 selective NSAIDs as a class are associated with clinically important differences in risk compared to nonselective NSAIDs. A systematic review included in the original CER of five small (sample size range 15 to 67), short-term (28 days or less) trials found that COX-2 selective NSAIDs had similar effects on glomerular filtration rate and creatinine clearance compared to nonselective NSAIDs in three trials, and were modestly superior in two.¹⁵⁶ The clinical effects of the modest differences observed in the latter two trials were unclear. Another systematic review found no difference in risk of creatinine increase greater than 1.3 times the upper limit of normal with celecoxib at 200 to 400 mg compared with nonselective NSAIDs (RR 0.78, 95% CI 0.46 to 1.3).⁵¹ CLASS showed no differences in the risk of experiencing an increase in serum creatinine >1.0 mg/dl with celecoxib (0.2 percent), diclofenac (0.1 percent), or ibuprofen (0.2 percent), though the nonselective NSAIDs were associated with slightly greater increases in serum creatinine, particularly in patients with prerenal azotemia at baseline.¹⁵⁹ A systematic review of randomized trials included as an appendix in the original CER found celecoxib associated with lower risk of renal dysfunction (RR 0.61, 95% CI 0.40 to 0.94) compared to control treatments (placebo, other NSAID, or mixed/other), but no difference for composite renal events (RR 0.97, 95% CI 0.84 to 1.1).¹⁵⁷

Two systematic reviews of randomized controlled trials included in the original CER found no clear difference between celecoxib and nonselective NSAIDs in risk of heart failure. In one systematic review, heart failure was more frequent with celecoxib than with placebo (13 of 8,405 vs. 1 of 4,057, p=0.05), though not compared with nonselective NSAIDs (0.1% vs. 0.2%, p=0.06).¹¹⁰ A second meta-analysis also found no significant difference between celecoxib and nonselective NSAIDs in risk of heart failure (RR 0.70, 95% CI 0.43 to 1.1).⁵¹ Similar results were observed in large trials of celecoxib. In CLASS, CHF rates were similar with celecoxib versus ibuprofen or diclofenac (0.3 percent vs. 0.3 percent).¹¹⁷, and withdrawals due to heart failure rare with all three NSAIDs (0.1% vs. <0.1% vs. 0.3%).¹⁵⁹ The APC polyp prevention trial found no difference in rates of heart failure between celecoxib versus placebo, though event rates were low (five cases of heart failure among 1,356 subjects).¹¹⁹

The risks of hypertension and heart failure with celecoxib and nonselective NSAIDs were evaluated in several observational studies. A new Danish cohort study of patients who had been hospitalized for congestive heart failure found use of celecoxib, ibuprofen, diclofenac, or naproxen at any dose associated with similar risk of hospitalization due to congestive heart failure (HR estimates ranged from 1.2 to 1.4), though celecoxib and diclofenac were associated with greater risk of death (HR 1.8, 95% CI 1.6 to 1.9 and HR 2.1, 95% CI 2.0 to 2.2, respectively) compared with ibuprofen and naproxen (HR 1.3, 95% CI 1.2 to 1.4 and HR 1.2, 95% CI 1.1 to 1.4, respectively).¹⁶⁰ A new nested case-control study found indomethacin associated with increased risk of heart failure compared to celecoxib (OR 2.0, 95% CI 1.2–3.6) in patients older than 66 years recently hospitalized for heart failure.¹⁵⁸ There was no difference in risk of heart failure between other nonselective NSAIDs (diclofenac [OR 0.82, 95% CI 0.51 to 1.3] and ibuprofen [OR 1.5, 0.66 to 3.2]) or acetaminophen (OR 1.2, 95% CI 0.92 to 1.4) relative to celecoxib. A retrospective cohort study included in the original CER based on the same Canadian database found nonselective NSAIDs associated with an increased risk of death (HR 1.5, 95% CI 1.2 to 2.0), recurrent heart failure (HR 1.2, 95% CI 0.92 to 1.6), or either (HR 1.3, 95% CI 1.0 to 1.6) in similarly high risk patients.¹⁶¹ Another retrospective cohort study included in the original CER found nonselective NSAIDs (RR 1.4, 95% CI 1.0 to 1.9) but not celecoxib (RR 1.0, 95% CI 0.8 to 1.3) associated with increased risk of heart failure admission compared to nonuse.¹⁶² A case-control study based on data from the General Practice Research Database found nonselective NSAIDs associated with an increased risk of newly diagnosed heart failure compared to nonuse of NSAIDs (RR 1.6, 95% CI 1.2 to 2.1).¹⁶³

A fair-quality systematic review included as an appendix in the original CER found no difference between celecoxib and controls (placebo, other NSAIDs, or mixed/other) in risk of arrhythmia, but the number of events was small (RR 0.84, 95% CI 0.45 to 1.6) and most trials didn’t report arrhythmias.¹⁵⁷

Hepatotoxicity

One systematic review¹⁶⁴ included in the original CER and one new meta-analysis¹⁶⁵ evaluated randomized controlled trials reporting hepatotoxicity associated with various NSAIDs. Another systematic review included in the original CER evaluated observational studies.¹⁶⁶ We identified one new randomized controlled trial of celecoxib versus diclofenac that reported rates of hepatic adverse events,⁵² and a report of hepatotoxicity from the diclofenac arm of a large randomized trial.¹⁶⁷

The new meta-analysis included 41 randomized trials involving celecoxib.¹⁶⁵ It found risk of hepatobiliary abnormalities (clinical or laboratory) similar for celecoxib (276/24933 or 1.1 percent), ibuprofen (38/2484 or 1.5%, p=0.06 vs. celecoxib), and placebo (36/4057 or 0.89%, p=0.21 vs. celecoxib); slightly lower rate for naproxen (0.68%, p=0.03 vs. celecoxib); and slightly higher for diclofenac (324/2618 or 4.24%, p<0.0001 vs. celecoxib). No patient randomized to an NSAID met Hay’s rule (elevation of alanine aminotransferase ≥3 times the upper limit of normal with an elevation of bilirubin ≥2 times the upper limit of normal), and no cases of liver failure or drug-related liver transplant were reported. The rate of alanine aminotransferase (ALT) abnormalities was higher with diclofenac (78/1000 patient-years) compared with the other NSAIDs or placebo (16 to 28/1000 patient-years). Four deaths occurred (2 in patients randomized to celecoxib, 1 naproxen, and 1 diclofenac), but none were considered related to drug treatment. A systematic review included in the original CER reported similar findings.¹⁶⁴ Based on 67 published articles and 65 studies accessible from the FDA archives, it found diclofenac (3.6%, 95% CI 3.1% to 4.0%) associated with higher rates of aminotransferase elevations greater than 3 times the upper limit of normal compared with placebo (0.29%; 95% CI 0.17% to 0.51%) and other NSAIDs (all ≤ 0.43 percent), and a higher rate of liver-related discontinuations compared to placebo (2.2%, 95% CI 1.8 to 2.6). Serious complications related to liver toxicity were rare: only one liver-related hospitalization (among 37,671 patients) and death (among 51,942 patients) occurred in a patient on naproxen in a trial of rofecoxib versus naproxen. Data from the diclofenac arm (n=17,289) of a randomized trial showed similar results.¹⁶⁷ The rate of aminotransferase elevation greater than three times the upper limit of normal was 3.1 percent, with four cases of liver-related hospitalizations (0.023 percent) and no cases of liver failure, death, or transplant.

Large trials that have evaluated diclofenac also suggested an increased risk of hepatotoxicity compared to other NSAIDs. In CLASS, celecoxib was associated with a lower risk of elevation in serum ALT (0.6 percent vs. 2.2 percent), serum AST (0.5 percent vs. 1.8 percent), and withdrawals due to hepatic enzyme elevations (<0.1 percent vs. 1.2 percent) compared to diclofenac or ibuprofen.⁵⁴ In SUCCESS-1, rates of increase in ALT levels were 0.5 percent with celecoxib versus 1.3 percent with diclofenac or naproxen (p<0.001).⁵⁵ A smaller (n=916), new trial comparing celecoxib versus diclofenac also found a lower risk of hepatic function abnormalities with celecoxib compared to diclofenac (0.6 percent vs. 3.5 percent).⁵²

A systematic review of seven population-based epidemiological studies found a similarly low risk of serious hepatic toxicity associated with NSAIDs.¹⁶⁶ In those studies, the excess risk of liver injury associated with current NSAIDs ranged from 4.8 to 8.6/100,000 person-years of exposure compared with past use. There were zero deaths from liver injury associated with NSAIDs in more than 396,392 patient-years of exposure. A recent cohort study from Italy found that nimesulide, an NSAID not available in the United States, was associated with a higher incidence of serious liver injury compared with other NSAIDs.¹⁶⁸ None of the other NSAIDs, including celecoxib, were associated with an increased risk of serious liver injury. An earlier review of five population-based studies found sulindac associated with a five- to tenfold higher incidence of hepatic injury compared with other NSAIDs.¹⁶⁹ Diclofenac was associated with higher rates of aminotransferase elevations compared with users of other NSAIDs, but not with a higher incidence of serious liver disease.

Celecoxib

Two systematic reviews^{50, 51} included in the original CER and one new systematic review⁵⁸ evaluated the relative tolerability of celecoxib compared to nonselective NSAIDs (Table 9). We also identified one new pooled analysis of randomized trials from the Pfizer registry,¹⁷⁰ one randomized trial not included in the original CER,⁵² and one pooled analysis of three similarly designed trials.⁸²

Table 9

Systematic review of tolerability of COX-2s compared with NSAIDs.

The new systematic review found no differences between celecoxib and nonselective NSAIDs in the risk of any adverse event (RR 0.96, 95% CI 0.91 to 1.0), GI adverse events (RR 0.90, 95% CI 0.78 to 1.0), or withdrawals due to adverse events (RR 0.86, 95% CI 0.73 to 1.0).⁵⁸ However, celecoxib was associated with a lower likelihood of withdrawals due to GI adverse events (RR 0.45, 95% CI 0.35 to 0.56). A systematic review included in the original CER reported found celecoxib associated with decreased risk of withdrawal due to adverse events (RR 0.86, 95% CI 0.81 to 0.91), withdrawal due to GI adverse events (RR 0.75, 95% CI 0.70 to 0.80), or any GI adverse event (RR 0.85, 95% CI 0.82 to 0.88).⁵¹ The risk of serious adverse events (RR 1.0, 95% CI 0.91 to 1.2) and any adverse event (RR 0.96, 95% CI 0.94 to 0.98) were similar. An older systematic review reported results consistent with the other two systematic reviews.⁵⁰ All of the systematic reviews included the large and longer-duration CLASS trial, which reported lower risks of withdrawal due to adverse events (18 percent vs. 21 percent) and withdrawal due to GI adverse events (8.7 percent vs. 11 percent) with celecoxib compared to diclofenac or ibuprofen.⁵⁴

A meta-analysis of 21 randomized trials from the Pfizer registry reported results that were generally consistent with the systematic reviews.¹⁷⁰ It found celecoxib associated with lower risk of GI adverse events (20 percent) compared to naproxen (32 percent), ibuprofen (31 percent), or diclofenac (24 percent), as well as lower likelihood of withdrawal due to GI adverse events (4.2% vs. 5.0% to 8.5%). However, this study was rated poor quality, in part because it did not assess study quality and because raw event rates were pooled across studies, resulting in loss of randomization.

One new randomized trial (n=925) found celecoxib and diclofenac associated with no difference in risk of withdrawal due to adverse events (27% vs. 31%, respectively, p=0.22) or withdrawal due to GI adverse events (15 percent vs. 14 percent).⁵² A pooled analysis from three similarly designed trials of patients in Asia (total n=880) found no difference between celecoxib and diclofenac in risk of withdrawal due to adverse events (3.4% vs. 6.1%, p>0.05).⁸²

Partially Selective NSAIDs

Two systematic reviews^{89, 92} of randomized trials included in the original CER and one new systematic review⁵⁸ evaluated the tolerability of meloxicam, etodolac, or nabumetone compared to nonselective NSAIDs. The new systematic review found meloxicam associated with decreased risk of any adverse event (RR 0.91, 95% CI 0.84 to 0.99), any GI adverse event (RR 0.31, 95% CI 0.24 to 0.39), and withdrawals due to GI adverse events (RR 0.61, 95% CI 0.54 to 0.69), though there was no difference in the risk of withdrawal for any adverse event (RR 0.92, 95% CI 0.66 to 1.3).⁵⁸ The median Jadad quality score for the trials included in the systematic review was three (maximum five), indicating moderate overall quality. A meta-analysis of meloxicam studies included in the original CER reported similar findings, with lower risks of any GI event (OR 0.64; 95% CI 0.59 to 0.69) and withdrawals due to GI events (OR 0.59; 95% CI 0.52 to 0.67) with meloxicam compared with nonselective NSAIDs.⁹²

The new systematic review also evaluated tolerability of etodolac.⁵⁸ It found etodolac associated with a lower risk of any adverse event (RR 0.83, 95% CI 0.70 to 0.99) compared to nonselective NSAIDs, but there was no difference in risk of GI adverse events (RR 0.77, 95% 0.55 to 1.1), withdrawal due to adverse events (RR 0.93, 95% CI 0.77 to 1.1), or withdrawal due to GI adverse events (RR 0.95, 95% CI 0.54 to 1.6). Only two of 29 trials of etodolac scored 5 out of 5 on the Jadad quality scale; 7 received only 2 points.

In a meta-analysis included in the original CER, the incidence of GI adverse events was slightly but statistically significantly lower with nabumetone compared to nonselective NSAIDs (25 % vs. 28%, p=.007), corresponding to about one fewer event for every 34 patients treated with nabumetone.⁸⁹

Nonselective NSAIDs

A Cochrane review included in the original CER evaluated the tolerability of different NSAIDs.⁵⁶ The only relatively consistent finding was that indomethacin was associated with higher rates of toxicity than other NSAIDs, but it was not clear if these differences were statistically significant.

Aspirin and Salsalate

Five randomized trials (all included in the original CER) evaluated the efficacy or safety of aspirin or salsalate compared with nonaspirin NSAIDs in patients with arthritis.^{80, 171–174} All were short-term (≤ 12 weeks) and involved a total of 471 patients; of the subjects enrolled, only 4 had osteoarthritis of the hip/knee for every 100 patients with rheumatoid arthritis. Aspirin was associated with higher incidence of overall adverse events than salsalate (70% vs. 40%, p<0.05)⁸⁰ and diclofenac (61% vs. 46%; p<0.05);¹⁷³ these led to higher rates of withdrawals due to adverse events for aspirin compared with diclofenac (23% vs. 6%; p<0.05). Salsalate was associated with a higher incidence of overall adverse events compared to other nonselective NSAIDs in two^{171, 174} of three trials, but the actual rates were not reported.

The overall safety profile of salsalate has also been evaluated in the rheumatoid arthritis population using the Arthritis, Rheumatism, and Aging Medical Information System (ARAMIS) databases. These studies reported summary measures of drug toxicity based on tabulations of mean frequencies of overall adverse events per patient years, weighted by severity, and adjusted for differences in demographic factors. Numerically larger index scores indicate greater levels of toxicity. The summary index score takes into account symptoms from all body systems, laboratory abnormalities, and all-cause hospitalizations.^175–178 Symptoms were assessed every 6 months using patient self-report in response to open-ended questions. Hospitalizations and deaths were ascertained from discharge summaries and death certificates. Descriptions of study methods varied, but the ARAMIS studies were somewhat vague with regard to patient selection and ascertainment methods; adverse events were not clearly defined or prespecified; exposure duration and length of followup were unclear; and adjustments were made only for demographic factors such as age and gender. Because the results of these studies are more subject to recall bias and had other methodological shortcomings, the findings that aspirin, salsalate, and ibuprofen were the least toxic among the NSAIDs studied (Table 10) are less convincing than if they were reported in more rigorously designed observational studies.

Table 10

Toxicity index scores from ARAMIS database studies.

Acetaminophen

Four systematic reviews included in the original CER evaluated the efficacy and safety of acetaminophen compared with NSAIDs (selective or nonselective) for osteoarthritis.^179–182 We identified no new systematic reviews. One new randomized trial compared acetaminophen versus naproxen for osteoarthritis.¹⁸³ One new observational study evaluated risk of acute myocardial infarction associated with various NSAIDs compared to acetaminophen.¹⁰⁸

The systematic reviews generally met all criteria for good-quality systematic reviews, except that three^180–182 did not provide sufficient detail about trials that were excluded. The overall conclusion from the reviews was that NSAIDs were modestly superior to acetaminophen for general or rest pain (Table 11). For pain on motion and overall assessment of clinical response, NSAIDs also appeared modestly superior, though the differences were not always statistically significant.^{180, 181} Only two reviews assessed functional disability; neither found clear differences.^{180, 181}

Table 11

Pain relief in systematic reviews of acetaminophen compared with NSAID.

The risk of adverse events with acetaminophen versus NSAIDs was assessed in three systematic reviews (Table 12).^{179, 180, 182} In two reviews, there were no differences in withdrawal due to any adverse event.^{179, 180} Acetaminophen was associated with lower risk of GI adverse events compared with nonselective NSAIDs in two systematic reviews (though not compared with coxibs)^{180, 182} and lower risk of withdrawals due to GI adverse events in one systematic review.¹⁸⁰ One systematic review found no difference between NSAID and acetaminophen in serious GI, renal, or CV harms, but found few events in the primarily small, short-term trials (data not provided).¹⁸⁰

Table 12

Adverse events in systematic reviews of acetaminophen compared with NSAID.

A new, fair-quality (high loss to followup) randomized trial found no differences in withdrawal due to lack of efficacy or WOMAC scores between acetaminophen 4 g once daily and naproxen 750 mg once daily for osteoarthritis after 6 months (n-=105) or 1 year (n=476).¹⁸³ Acetaminophen and naproxen were also associated with similar rates of withdrawal due to adverse events (25% vs. 22%, NS), serious adverse events (3.5 percent vs. 2.5 percent), any adverse event (72 percent vs. 74 percent), renal adverse events (three total), or hepatic enzyme increases (three in acetaminophen group vs. zero in the naproxen group). Naproxen was associated with an increased risk of constipation (9.9% vs. 3.1%, p<0.002) and peripheral edema (3.9% vs. 1.0%, p<0.033) compared to acetaminophen.

Clinical trials of acetaminophen have not been large enough to assess serious but less common complications such as PUBs, myocardial infarction, acute renal failure, or hypertension. Several observational studies included in the original CER provide some additional information about the safety of acetaminophen relative to NSAIDs. A fair-quality nested case-control study of 1,197 cases and 10,000 controls from a population-based cohort of 458,840 people in the General Practice Research Database found current acetaminophen use associated with a lower risk for symptomatic peptic ulcer (adjusted RR 1.9, 95% CI 1.5 to 2.3) than NSAID use (adjusted RR 4.0, 95% CI 3.2 to 5.1) when each was compared with nonuse.¹⁸⁴ There was no clear relationship between higher acetaminophen dose and increased risk for symptomatic ulcers. An earlier analysis on the same database also found current acetaminophen use associated with a lower risk for upper GI bleeds or perforations (adjusted RR 1.3, 95% CI 1.1 to 1.5) than current NSAID use (adjusted OR 3.9, 95% CI 3.4 to 4.6), each compared with nonuse.⁹⁸ A retrospective cohort study of elderly patients found that patients using lower doses of acetaminophen (<2,600 mg once daily) had lower rates of GI events (defined as GI-related hospitalizations, ulcers, and dyspepsia) compared with users of NSAIDs (RR 0.73, 95% CI 0.67 to 0.80 for 1,951 to 2,600 mg once daily), but the risks were similar at higher doses (RR 0.93 to 0.98).¹⁸⁵ Although GI hospitalization rates were not reported separately, the authors noted that dyspepsia was responsible for most of the increase in GI events in the high-dose acetaminophen groups. A meta-analysis on individual patient data from three earlier retrospective case-control studies (2472 cases) was consistent with the above studies.¹⁸⁶ It found acetaminophen associated with a minimal increase in the risk for serious upper GI bleeding (OR 1.2, 95% CI 1.1 to 1.5). By contrast, nonselective NSAIDs were associated with higher risks, though estimates of risk varied considerably for different NSAIDs (OR 1.7 for ibuprofen to 35 for ketoprofen).

No randomized trial evaluated the association between acetaminophen use and myocardial infarction or other thromboembolic CV events. An analysis from the large, prospective Nurses’ Health Study found heavy use of acetaminophen (more than 22 days/month) associated with an increased risk of CV events (RR 1.4, 95% CI 1.1 to 1.6) similar to that with heavy use of NSAIDs (RR 1.4, 95% CI 1.3 to 1.6).¹⁸⁷ Dose-and frequency-dependent effects were both significant. A new retrospective cohort study found no difference in risk of acute myocardial infarction between celecoxib, ibuprofen, diclofenac, or naproxen versus acetaminophen (Table 13, Appendix H).¹⁰⁸

The association between renal failure and acetaminophen use was evaluated in several case-control studies included in the original CER. Interpretation of these studies is difficult because many had important flaws (such as failure to identify patients early enough in the course of their disease to ensure that the disease had not led to a change in the use of analgesics, failure to specify diagnostic criteria, failure to adjust for the use of other analgesics, incompleteness of data on exposure, and use of proxy respondents) in the collection or analysis of data.¹⁸⁸ The largest (926 cases) case-control study was designed to try to avoid many of these flaws, though results remain susceptible to confounding by indication.¹⁸⁹ It found regular use of acetaminophen associated with an increased risk for chronic renal failure (Cr >3.8 for men and >3.2 for women) compared with nonuse (OR 2.5, 95% CI 1.7 to 3.6). Use of NSAIDs was not associated with an increased risk (OR 1.0). A prospective cohort study of 1,697 women in the Nurses’ Health Study found increased lifetime acetaminophen exposure associated with a higher risk of decline in glomerular filtration rate of 30% or greater (p<0.001), though NSAIDs were not (p=0.88).¹⁹⁰ The absolute risk of renal function decline, however, was modest, even in women reporting high amounts of lifetime acetaminophen use. Compared with women consuming less than 100 g of cumulative acetaminophen, the odds of a decline in GFR of at least 30 mL/min per 1.73 m² for women consuming more than 3,000 g was 2.04 (95% CI, 1.28 to 3.24). By contrast, analyses of men in the Physicians’ Health Study found no association between acetaminophen or NSAIDs and change in kidney function.^{191, 192}

The risk of heart failure associated with acetaminophen has not been well studied. In a single study using the General Practice Research Database, current use of acetaminophen was associated with a higher risk of newly diagnosed heart failure compared with nonuse (RR 1.3, 95% CI 1.1 to 1.7), though the risk was lower compared with current use of NSAIDs (RR 1.6, 95% CI 1.2 to 2.0).¹⁶³

The risk of hypertension has been evaluated using data from the Nurses’ Health Studies^193–195 and the Physicians’ Health Study.¹⁹⁶ In the Nurses’ Health Studies, acetaminophen and NSAIDs were associated with similar increases in risk of incident hypertension (Table 13). In the Physicians’ Health Study, there was no association between NSAID or acetaminophen use and hypertension.

Glucosamine and Chondroitin

Five new systematic reviews (Appendix I)^197–201 and four systematic reviews ^202–205 included in the original CER evaluated benefits and harms of glucosamine and chondroitin. New trials identified for this update include one trial of glucosamine versus acetaminophen and placebo,²⁰⁶ two trials of glucosamine versus placebo,^{207, 208} four trials of chondroitin versus placebo,^209–212 and one trial of the combination of glucosamine and chondroitin versus placebo (Table 14, Appendix I).²¹³ We also identified two followup analyses from the previously included Glucosamine/chondroitin Arthritis Intervention Trial (GAIT).^{214, 215}

Table 14

Response rates in the Glucosamine/chondroitin Arthritis Intervention Trial (GAIT).

Glucosamine

The most promising results for glucosamine have been reported in trials evaluating a pharmaceutical grade glucosamine not available in the United States, and sponsored by its European manufacturer (the Rotta Corporation). Because the content and purity of over-the-counter glucosamine preparations vary substantially, the results of trials that evaluated pharmaceutical grade glucosamine may not be directly applicable to over-the-counter preparations available in the U.S.²¹⁶

The original CER included a good-quality Cochrane review (searches through November 2004) with four short-term (4 to 8 weeks) head-to-head trials of glucosamine versus an oral NSAID (ibuprofen or piroxicam).²⁰⁵ Two of the trials were rated 5 out of 5 on the Jadad scale, and the other two were rated 3 or 4 out of 5. Three of the trials were sponsored by the European manufacturer; the fourth²¹⁷ was also conducted in Europe, but funding information was not reported. One of the trials has only been published as an abstract,²¹⁸ with analyses based on data from an unpublished manuscript. Two of the four trials found glucosamine superior to oral NSAIDs for efficacy,^{217, 218} and two found no difference.^{219, 220} In pooled analyses, glucosamine was superior to an oral NSAID for improving pain (three trials, standardized mean difference −0.40, 95% CI −0.60 to −0.19), but not for improving function measured with the Lequesne Index (two trials, standardized mean difference [SMD] −0.36, 95% CI −1.07 to 0.35). Glucosamine was also associated with fewer adverse events (RR 0.29, 95% CI 0.19 to 0.44) and withdrawals due to toxicity (RR 0.06, 95% CI 0.01 to 0.25).

Three head-to-head trials^221–223 included in the original CER were not included in the Cochrane review. The large, (n=1,583), NIH-funded, good-quality GAIT trial compared glucosamine versus celecoxib, as well as placebo, chondroitin, and the combination of glucosamine plus chondroitin (Tables 14 and 15, Appendix I).²²¹ GAIT evaluated pharmaceutical-grade glucosamine hydrochloride (over-the-counter supplements commonly available in the United States are typically glucosamine sulfate) and chondroitin sulfate under an investigational new drug application. It found no differences between glucosamine and celecoxib in the proportion of responders defined by those with at least a 20 percent decrease in WOMAC pain score (70% vs. 64%, RR 0.91 [95% CI 0.82 to 1.02]), or as defined using Outcomes Measures in Arthritis Clinical Trials-Osteoarthritis Research Society International (OMERACT-OARSI) criteria (67% vs. 61%, RR 0.90 [95% CI 0.80 to 1.01]). There were also no differences in change from baseline on WOMAC scores, SF-36 Mental or Physical Component summary scores, or the Health Assessment Questionnaire. The number of withdrawals due to adverse events was similar (2.8 percent vs. 2.2 percent), with no serious GI adverse events or deaths in either group. One patient randomized to glucosamine had chest pain and one patient randomized to celecoxib had a stroke. The celecoxib group experienced a nonsignificant but higher incidence of “cardiac” events compared to patients randomized to other treatments, though these were predominantly arrhythmias (palpitations and atrial fibrillation) rather than ischemic events (data not reported). Two small (n=40 and n=45), 12-week Canadian trials (not funded by the European manufacturer of pharmaceutical grade glucosamine) found no differences between glucosamine and ibuprofen for general osteoarthritis pain²²² or temporomandibular joint osteoarthritis.²²³ Only limited details of the study design were reported for the first trial, though the second met all criteria for a good-quality study.

Table 15

Efficacy, glucosamine and chondroitin trials.

One new, fair-quality trial sponsored by the European manufacturer of pharmaceutical grade glucosamine found no difference between glucosamine and acetaminophen in improvements from baseline on the Lequesne Index (−3.1 [95% CI −3.8 to −0.8) vs. −2.7 [95% CI −3.3 to −2.1], respectively), the WOMAC total score (−13 [95% CI −16 to −10] vs. −12 [95% CI −15 to −10.0]), the WOMAC pain score (−2.7 [95% CI −3.3 to −2.1] vs. −2.4 [−3.0 vs. −1.8]), and the WOMAC function score (−9.2 [−11 vs. −7.2] vs. −8.7 [−11 vs. −6.8]).²⁰⁶ There was also no difference in the proportion of responders based on OARSI-A criteria (40 percent vs. 33 percent). Adverse events were similar.

Four systematic reviews not included in the original CER focused on evaluations of glucosamine versus placebo (Appendix I).^{197, 198} The first, fair-quality systematic review, by Bjordal et al., was based on 7 randomized trials (sample size range 10 to 126, median 46, total n=401).¹⁹⁷ It found glucosamine associated with a statistically significant but clinically nonsignificant beneficial effect on pain compared to placebo (mean difference 4.7 points on a 100-point scale, 95% CI 0.3 to 9.1). The second, good-quality systematic review, by Wandel et al., differed from the first in that in focused on larger (n>200) randomized trials (7 trials of glucosamine, 1,939 patients randomized to glucosamine vs. placebo), included more recently published trials, and conducted network analysis to incorporate indirect evidence.¹⁹⁸ It also found a statistically significant but clinically nonsignificant beneficial effect of glucosamine on pain (−0.4 cm on a 10 cm scale, 95% credible interval −0.7 to −0.1) and joint space narrowing (−0.2 mm, 95% CI −0.3 to 0.0) compared to placebo. No differences were found when trials were stratified according to whether they evaluated glucosamine hydrochloride or sulfate. There was no difference between glucosamine and placebo in withdrawals due to adverse events. A third, good-quality systematic review (by Vlad et al.) of 15 randomized trials (sample size range 24 to 630, median 155, total n=2,613) found glucosamine associated with an SMD of 0.35 (95% CI 0.14 to 0.56), based on the primary outcome reported in each study.²⁰¹ However, the overall estimate was associated with substantial statistical heterogeneity (I²=80 percent). In stratified analyses, there was no statistically significant effect and heterogeneity was absent in subgroups of trials without industry funding (four trials, SMD 0.05, 95% CI −0.32 to 0.41), those that did not evaluate a Rottapharm produce (seven trials, SMD 0.11, 95% CI −0.16 to 0.38), and those with adequate allocation concealment (five trials, SMD 0.09, 95% CI −0.24 to 0.42). A fourth, fair-quality systematic review by Lee et al. found glucosamine associated with decreased joint space narrowing (SMD 0.43, 95% CI 0.23 to 0.63) and decreased risk of >0.5 mm joint space narrowing (OR 0.36, 95% CI 0.20 to 0.64), but results were based on only two trials.²⁰⁰

Other systematic reviews^{202, 204, 205} included in the original CER are now outdated, as they excluded recent good-quality and relatively large trials. The Cochrane review included in the original CER found higher trial quality and evaluation of non-Rotta brand glucosamine associated with lower estimates of benefits.²⁰⁵ Other older systematic reviews also found important methodological flaws in the glucosamine trials that could have exaggerated estimates of effect.^{202, 204}

The previously described, good-quality GAIT trial is the largest trial of glucosamine.²²⁴ It found no difference between glucosamine and placebo in the likelihood of experiencing a >20% improvement in WOMAC pain score after 24 weeks (64% vs. 60%, RR 1.1 [95% CI 0.94 to 1.2]), or various WOMAC, SF-36, and Health Assessment Questionnaire scores. There was also no difference in joint space width narrowing,²¹⁴ likelihood of achieving a 20 percent reduction in WOMAC pain (OR 1.2, 95% CI 0.65 to 2.0), or improvement in WOMAC function after 24 months.²¹⁵

Three trials of glucosamine versus placebo have been published since the original CER.^206–208 All except one²⁰⁸ were included in the Wandel et al. systematic review.¹⁹⁸ Of the three new trials, two were rated good quality.^{207, 208} Both found no differences on outcomes related to pain, function, or (in one trial) radiographic narrowing between glucosamine and placebo for hip osteoarthritis²⁰⁷ or low back pain with degenerative osteoarthritis.²⁰⁸ The hip osteoarthritis trial also found no differences in efficacy in subgroups defined by radiographic severity, type of osteoarthritis (localized or generalized), level of pain, and other factors.²⁰⁷ The third, fair-quality trial (high attrition) found glucosamine more effective than placebo in improving the Lequesne score (difference in mean change from baseline −1.2 [95% CI −2.3 to −0.8] on a 24 point scale), WOMAC Function score (difference −3.7 [95% CI −6.9 to −0.5] on a 68-point scale), and in the proportion experiencing an OARSI response (40% vs. 21%, RR 1.9 [95% CI 1.2 to 2.9]).²⁰⁶ In all three trials, withdrawals due to adverse events and specific adverse events were similar with glucosamine and placebo.

Previous Bleeding Ulcer

Two randomized trials included in the original CER^{233, 234} and one new trial²³⁵ compared the risk of GI harms in patients with a recent bleeding ulcer randomized to celecoxib versus the combination of celecoxib plus a PPI (Table 16).

Table 16

Celecoxib in patients with bleeding ulcer history.

In two fair-quality, 24-week trials (total n=529) included in the original CER of patients with a history of a recent bleeding ulcer, rates of recurrent bleeding were similar for celecoxib (200 mg daily 3.7 percent and twice a day 4.9 percent) and the combinations of extended-release diclofenac 75 mg twice a day plus omeprazole 20 mg daily (6.3 percent)²³³ or naproxen 250 mg three times a day plus lansoprazole 30 mg a day (6.3 percent)²³⁴ (differences not statistically significant). There were also no differences between celecoxib and either combination therapy in GI, renal, and CV adverse events or in rates of withdrawal due to adverse events. One exception was that celecoxib 200 mg daily was associated with a higher rate of dyspepsia than naproxen 250 mg three times a day plus lansoprazole 30 mg daily in one trial.²³⁶

A new, fair-quality, 12-month trial (n=273) of patients with recently healed GI bleeding (following cessation of NSAID therapy and treatment with a PPI for 8 weeks) found celecoxib 200 mg twice daily plus esomeprazole 20 mg twice daily associated with significantly fewer ulcer bleeding recurrences compared with celecoxib alone.²³⁵ After a median of 13 months, zero events occurred in the combined treatment group, compared with 12 (8.9 percent) in the celecoxib alone group (p=0.0004). Similar results were found among those taking low-dose aspirin (0% vs. 19%; p=0.03). Other adverse events and rates of discontinuations were similar between groups.

Underlying Cardiovascular or Renal Risk

We found no randomized trials designed to assess whether the relative harms and benefits associated with different oral treatments for osteoarthritis vary according to underlying CV or renal risk. A new systematic review of long-term celecoxib trials found that risk of CV events doubled between patients at low and moderate baseline CV risk (HR 2.0, 95% CI 1.5 to 2.6) and doubled again in patients at high baseline risk (HR, high risk to low risk, 3.9, 95% CI 2.3 to 6.7).¹¹⁴ Most^{132, 133, 145, 237} but not all¹³⁴ large observational studies also found a history of coronary heart disease, coronary heart disease risk factors, or categorization as high CV risk associated with increased risk estimates with NSAIDs as a group. A good-quality, population-based study of a very high-risk group of 58,000 Danish patients with previous myocardial infarction found hazard ratios for death of 2.6 (95% CI 2.2 to 3.1) for celecoxib, 1.5 (95% CI 1.4 to 1.7) for ibuprofen, 2.4 (95% CI 2.1 to 2.8) for diclofenac, and 1.3 (95% CI 1.2 to 1.4) for other NSAIDs compared to nonuse of NSAIDs.²³⁸ Based on the rates of death in this population (95 per 1,000 person-years in those not using NSAIDs), the estimated number of patients needed to treat with an NSAID for one year to cause one additional death was 14 (95% CI 10 to 24) for celecoxib, 45 (95% CI 29 to 102) for ibuprofen, and 24 (95% CI 16 to 45) for diclofenac.

We found no trials evaluating comparative risks of different oral medications in patients with known congestive heart failure. A new cohort study found celecoxib, ibuprofen, diclofenac, and naproxen associated with similar risk of hospitalization due to acute myocardial infarction (HR estimates ranged from 1.3 to 1.5) or hospitalization due to congestive heart failure (HR estimates ranged from 1.2 to 1.4) following an index hospitalization for congestive heart failure, though celecoxib (HR 1.8, 95% CI 1.6 to 1.9) and diclofenac (HR 2.1, 95% C I 2.0 to 2.2) were associated with greater risk of death compared to ibuprofen (HR 1.3, 95% CI 1.2 to 1.4) or naproxen (HR 1.2, 95% CI 1.1 to 1.4).¹⁶² A new nested case-control study found indomethacin associated with increased risk of heart failure compared to celecoxib (OR 2.0, 95% CI 1.2 to 3.6) in patients older than 66 recently hospitalized for heart failure.¹⁵⁸ There was no difference in risk of heart failure between other nonselective NSAIDs (diclofenac [OR 0.82, 95% CI 0.51 to 1.3] and ibuprofen [OR 1.5, 0.66 to 3.2]) or acetaminophen (OR 1.2, 95% CI 0.92 to 1.4) relative to celecoxib. A retrospective cohort study included in the original CER based on the same Canadian database found nonselective NSAIDs associated with an increased risk of death (HR 1.5, 95% CI 1.2 to 2.0), recurrent heart failure (HR 1.2, 95% CI 0.92 to 1.6), or either (HR 1.3, 95% 1.0 to 1.6) in similarly high risk patients.¹⁶¹

One new trial (n=88) compared ibuprofen, acetaminophen, and piroxicam in hypertensive patients on lisinopril/hydrochlorothiazide or amlodipine.²³⁹ Both NSAIDs blunted the effects of the antihypertensive drugs, with the lisinopril/hydrochlorothiazide combination more affected. Acetaminophen had almost no effect on blood pressure.

Nonselective NSAIDs

Concomitant use of anticoagulants and nonselective NSAIDs increase the risk of GI bleeding three-to sixfold compared to anticoagulants alone.^{240, 241} Three observational studies included in the original CER evaluated risk of bleeding in patients on an NSAID plus anticoagulants versus an anticoagulant alone.^242–244 We identified no new studies.

A good-quality nested case-control study of elderly (>66 years old) patients on warfarin in Ontario, Canada, evaluated the association between hospitalization for upper GI bleeding (361 cases) and use of selective or nonselective NSAIDs.²⁴² It found that after adjustment for potential confounders (antiplatelet agents, hypoglycemic agents, glucocorticoids, gastroprotective agents, history of previous bleed, and comorbidities), recent use of nonselective NSAIDs (OR 1.9, 95% CI 1.4 to 3.7), and celecoxib (OR 1.7, 95% CI 1.2 to 3.6) were associated with increased and overlapping risks for upper GI bleeding, compared with nonuse. Because this study relied on pharmacy databases to identify exposures prior to hospitalization, it could not assess the confounding effects of over-the-counter use of aspirin, other NSAIDs, or acid suppressive medications. It also was unable to control for variations in INR level and the risk for bleeding.

In a fair-quality cohort study of patients enrolled in an anticoagulation clinic, 1,145 patients who were receiving warfarin (INR ≥1.4) but not aspirin, acetaminophen, or other nonselective NSAID were indentified retrospectively. ²⁴³ Eleven percent (n=123) were taking celecoxib concurrently with warfarin during the study period. The risk of major bleeding events (requiring hospitalization, transfusion or resulting death) was not significantly elevated in the celecoxib group (RR 1.0, 95% CI 0.14 to 7.8).

A smaller, fair-quality nested retrospective study of patients in the Netherlands evaluated the risk of bleeding in anticoagulated patients receiving partially selective (meloxicam or nabumetone) or nonselective NSAIDs.²⁴⁴ This study differed from the others in that it included all cases of bleeding, including minor visible bleeding, hematoma, or black tarry stools. Patients were identified as having exposure to anticoagulation by being enrolled in a pharmacy-based anticoagulation program. Bleeding events were identified through the pharmacy clinic records, and discharge diagnosis records (national database). Patient questionnaires were sent out to those identified as having a bleeding event, to assess exposure status and comorbidities. Patients were interviewed over the phone if answers were incomplete or unclear. The response rates were significantly higher in the cases (approximately 70 percent) compared with controls (approximately 31 percent). The study found that nonselective NSAIDs were associated with an increased risk of bleeding compared with partially selective NSAIDs after adjustment for duration of use and INR level (OR 3.1, 95% CI 1.2 to 8.0).

Aspirin and Anticoagulation

In the original CER, we found no studies evaluating risks and benefits of concomitant anticoagulants and aspirin in patients with arthritis. No new studies were identified for this update. Combination therapy has been studied in patients with indications for thromboembolic prophylaxis. However, the results of those studies are not directly applicable to patients with arthritis because of important differences in the populations (particularly with regard to CV risk), and because aspirin was used in lower, prophylactic doses (rather than anti-inflammatory and analgesic doses). One fair-quality meta-analysis (did not evaluate quality of included trials) found major bleeding risk increased with warfarin plus aspirin versus warfarin alone (at the same intensity) in patients with mechanical heart valves (three trials, RR 1.58, 95% CI 1.02 to 2.44).²⁴⁵ In patients with recent myocardial infarction or atrial fibrillation (one trial each), the increase in risk was not statistically significant (RR 3.07, 95% CI 0.33 to 28.38 and RR 2.13, 95% CI 0.20 to 23.03, respectively). In patients with mechanical heart valves, the increase in bleeding risk was offset by a reduction in thromboembolic events (RR 0.33, 95% CI 0.19 to 0.58), and there was no difference in all-cause mortality (RR 0.78, 95% CI 0.29 to 1.83). Other evidence on the risks and benefits of combination therapy has focused on comparing warfarin plus aspirin to aspirin alone. A good-quality meta-analysis of 10 trials found the combination of warfarin plus aspirin increased the risk of major bleeding compared with aspirin alone following myocardial infarction or the acute coronary syndrome (RR 2.5, 95% CI 1.7 to 3.7).²⁴⁶ However, the increase in bleeding risk was offset by lower risks for myocardial infarction, ischemic stroke, and revascularization. Mortality did not differ.

Other Analgesics

No study evaluated risk of bleeding in anticoagulated patients on acetaminophen compared with those on NSAIDs. A small, randomized controlled trial found acetaminophen associated with greater increases in INR levels compared with placebo.²⁴⁷ Several observational studies have also found an association between excess anticoagulation and use of acetaminophen.^{248, 249} However, changes in INR are not the only important factor for predicting increased risk of bleeding. NSAIDs, for example, also affect platelet function and disrupt the gastric mucosal lining. Studies evaluating actual bleeding complications are necessary to better assess the comparative risks from acetaminophen and other NSAIDs.

No studies evaluated risk of bleeding in anticoagulated patients on glucosamine, chondroitin, or topical agents.

Celecoxib Plus Aspirin Compared With Nonselective NSAID Plus Aspirin

Beneficial effects of COX-2 selective inhibition on GI complication rates could be attenuated or eliminated by the concomitant use of aspirin. The original CER included two large trials (CLASS⁵⁴ and SUCCESS-1⁵⁵) and a systematic review⁵¹ that reported rates of ulcer complications associated with celecoxib and nonselective in subgroups of patients also using aspirin. Two new observational studies also compared risks of serious GI harms with celecoxib and nonselective NSAIDs in aspirin users.^{250, 251}

In the 20 percent of patients in CLASS who took aspirin in addition to their study drug, there was no difference in rates of ulcer complications (2.0% vs. 2.1%, p=0.92) or ulcer complications plus symptomatic ulcers (4.7% vs. 6.0%, p=0.49) in patients randomized to celecoxib versus those randomized to diclofenac or ibuprofen.^{54, 252} There were also no differences when celecoxib was compared to diclofenac and ibuprofen separately. In SUCCESS-1, among the 7 percent of the study population on aspirin, only four ulcer complications occurred, resulting in imprecise estimates (OR 2.0, 95% CI 0.14 to 27).⁵⁵ The systematic review found that use of aspirin increased the rate of endoscopic ulcers by about 6 percent in patients randomized to celecoxib (4.2% without aspirin and 9.9% with aspirin) or those randomized to a nonselective NSAID (18% and 24%, respectively).⁵¹ Celecoxib (any dose) was associated with a lower risk of endoscopic ulcers in aspirin users (RR 0.48, 95% CI 0.28 to 0.83), but ulcer complications were not reported in this subgroup.

The two new, fair-quality retrospective cohort studies were conducted by the same authors and evaluated the same Quebec health services administrative databases (Appendix H).^{250, 251} One study found use of celecoxib plus aspirin associated with a lower risk of hospitalizations due to ulcer complications compared with use of a nonselective NSAID plus aspirin (HR 0.62, 95% CI 0.48 to 0.80) among patients 65 years or older.²⁵⁰ The second study found that in patients 50 years and older, users of celecoxib plus aspirin had a lower risk of hospitalization for GI complications (HR 1.8, 95% CI 1.5 to 2.1) than users of diclofenac plus aspirin (HR 2.8, 95% CI 2.2 to 2.8, 95% CI 2.2 to 3.5), though estimates for ibuprofen plus aspirin (HR 1.4, 95% CI 0.8 to 2.7), naproxen plus aspirin (HR 2.2, 95% CI 1.6 to 3.0), and piroxicam plus aspirin (HR 2.0, 95% CI 0.8 to 5.4) were similar (each compared with users of acetaminophen without aspirin).²⁵¹

Impact of Concomitant PPI use on GI Risk With a Celecoxib or Nonselective NSAID

Subgroup analyses from four randomized controlled trials (reported in three publications) provided evidence on the effects of adding a PPI to celecoxib or nonselective NSAIDs in aspirin users.^{235, 253, 254} Only one fair-quality trial reported ulcer complications. ²³⁵ In patients (n=273) at very high risk for rebleeding (recently healed GI bleed) enrolled in this study, low-dose aspirin was started during the trial period in 43 patients. The rate of recurrent bleeding in this subgroup was 0 percent with celecoxib 200 mg twice daily plus esomeprazole 20 mg twice daily group compared with 19 percent in the group taking celecoxib alone (p=0.03).²³⁵

In two similarly designed, fair-quality trials (reported in one publication, total n=861) the pooled rate of endoscopically proven gastric ulcer in the subgroup also taking low-dose aspirin (n= 201) was significantly lower with naproxen plus a PPI (3 percent) compared with enteric coated naproxen alone (28%; p<0.001).²⁵³ A large, fair-quality trial of celecoxib versus naproxen plus lansoprazole in low-dose aspirin users (n = 1,045) found no difference in risk of endoscopically proven gastric or duodenal ulcers (9.9% vs. 8.9%, p=0.65).²⁵⁴ Post-hoc analyses showed no effect based on aspirin dose (81 mg or 325 mg daily).

Celecoxib

The original CER included a systematic review, a randomized trial not included in the systematic review, and two large observational studies on effects of aspirin on CV harms associated with celecoxib use. We identified no new studies.

A systematic review of 84 placebo-controlled trials of celecoxib that permitted aspirin use found a very similar risk of vascular events among aspirin users (RR 1.6, 95% CI 0.90 to 2.7) and aspirin nonusers (RR 1.5, 95% CI 1.1 to 2.0), though the absolute rate of events was higher in aspirin users (1.9 percent/year vs. 1.1 percent/year), perhaps due to higher baseline risk.¹²¹ In a large celecoxib polyp prevention trial not included in the systematic review, use or nonuse of low-dose aspirin did not affect the observed increased risk of thrombotic events.²⁵⁵ Consistent with these findings, two large observational studies found no significant interaction between concurrent NSAID and aspirin use and risk of myocardial infarction.^{134, 136}

Nonselective NSAIDs

It has been suggested that some nonselective NSAIDs may reduce or eliminate the CV benefits associated with low-dose aspirin.³⁶ In particular, ibuprofen is thought to be associated with unique pharmacokinetic and pharmacologic properties that could interfere with aspirin. Six observational studies^{108, 133, 256–259} and one subgroup analysis from a randomized trial²⁶⁰ evaluated effects of concomitant NSAIDs on CV risk in aspirin users (Table 17, Appendix H). The studies used heterogeneous designs, outcome measures, and methods of analysis, making it difficult to reach firm conclusions about comparative risks.

Table 17

Observational studies of the cardiovascular risk with concomitant aspirin.

Three observational studies found no increase in risk of mortality²⁵⁷ or myocardial infarction^{133, 256} in users of a nonselective NSAID plus aspirin versus aspirin alone (Table 17). A subgroup analysis from a randomized trial also found no increased risk with short-term (<60 days) use of a nonselective NSAID plus aspirin compared to aspirin alone.²⁶⁰ The estimate for longer term use suggested increased risk, but was imprecise and not statistically significant.

For the effect of adding ibuprofen to aspirin, one fair-quality study of patients recently discharged from the hospital for a CV disease diagnosis found an increased risk of overall and CV mortality with the combination of ibuprofen and aspirin compared to aspirin alone (Table 17).²⁵⁸ The study did not report baseline characteristics of patients, although the analysis did control for potential confounders.

Two other observational studies evaluated risk of acute MI with ibuprofen plus aspirin versus aspirin alone.^{256, 259} A fair-quality case-control study found the risk of first nonfatal MI was elevated in those using ibuprofen plus aspirin compared with those using only aspirin, while a retrospective cohort study (also fair quality) found that adding ibuprofen to aspirin resulted in decreased risk of myocardial infarction (Table 17).^{256, 259} These studies used different methods, which could account for their discrepant findings. The case-control study identified controls from the community, used telephone interviews to collect exposure and covariate data, and considered the patient to be exposed to ibuprofen or aspirin if they reported using the drug(s) in the week prior to the event.²⁵⁶ Recall bias is a major concern with this study, and differences between groups suggest potentially important differences in baseline risk. The cohort study used Veterans Affairs prescription and medical records to identify regular users of ibuprofen and aspirin or aspirin alone with matching based on age, race, sex, and cholesterol levels.²⁵⁹ The study did not measure potential confounders or conduct adjusted analyses, and very limited information was provided about the patients’ comparability at baseline.

Two other observational studies found no statistically significant differences in risk between ibuprofen plus aspirin versus ibuprofen alone¹⁰⁸ or use of ibuprofen plus aspirin versus nonuse of NSAIDs (including aspirin),¹³³ but were not designed to assess risk associated with addition of ibuprofen to aspirin.