Effect of nonsteroidal antiinflammatory drugs on the C-reactive protein level in rheumatoid arthritis: A meta-analysis of randomized controlled trials

Authors


Abstract

Objective

To evaluate the effects of oral nonsteroidal antiinflammatory drugs (NSAIDs) on C-reactive protein (CRP) levels in rheumatoid arthritis (RA) patients, with a prespecified focus on the different NSAIDs.

Methods

We performed a systematic search in Medline via PubMed, the Cochrane Central Register of Controlled Trials, EMBase via OVID, the Institute for Scientific Information Web of Science, and other sources. Eligible trials were parallel-group, randomized, placebo-controlled trials of oral NSAID therapy in RA patients for which there were extractable CRP data. Standardized mean differences (SMDs) with 95% confidence intervals (95% CIs) were calculated from the differences in means of CRP levels between groups (active treatment minus placebo) divided by the pooled SDs. For the meta-analysis, a random-effects model was used to estimate the overall change in CRP level, and stratified analysis was used to examine differences among NSAIDs.

Results

We included 19 trials of 10 different NSAIDs. Overall, NSAIDs showed no effect on the CRP level (SMD 0.01 [95% CI −0.03, 0.06], P = 0.62). However, the prespecified stratified analysis indicated varying effects on the CRP level according to the different NSAIDs; lumiracoxib caused a statistically significant and consistent (I2 = 0%) increase in the CRP level (SMD 0.13 [95% CI 0.01, 0.25], P = 0.037), whereas naproxen caused a statistically significant and consistent (I2 = 0%) decrease in the CRP level (SMD −0.11 [95% CI −0.20, −0.02], P = 0.022).

Conclusion

Overall, NSAIDs have no effect on the CRP level. However, the nonselective NSAID naproxen was associated with a significant decrease in the CRP level, whereas the cyclooxygenase 2–selective NSAID lumiracoxib was associated with a significant increase in the CRP level. This finding is interesting considering the suspected influence of NSAIDs on cardiovascular complications.

Nonsteroidal antiinflammatory drugs (NSAIDs) constitute an important class of therapeutic agents in the treatment of rheumatoid arthritis (RA), where they are known to reduce inflammation, swelling, and pain (1). The beneficial actions of NSAIDs have been linked to their ability to inhibit inducible cyclooxygenase 2 (COX-2) at sites of inflammation, while many of their adverse effects (e.g., gastrointestinal [GI] damage) are associated with inhibition of constitutive COX-1 (2). Selective inhibitors of COX-2, such as lumiracoxib, rofecoxib, etoricoxib, valdecoxib, and celecoxib, which are classified as coxib NSAIDs (3), have been developed with the purpose of having antiinflammatory properties without the adverse GI effects seen with COX-1 inhibition. Clinically, analgesic and antiinflammatory efficacy of the selective COX-2 inhibitors is similar to that of the nonselective NSAIDs (e.g., naproxen and ibuprofen), but with a lower incidence of gastric and duodenal ulcers (1). However, the risk of serious side effects such as cardiovascular (CV) events has previously been reported to be associated with the use of nonselective and selective COX-2 inhibitors (4–6), particularly with rofecoxib (7), although a clear relationship between COX-2 selectivity and risk of CV events was not observed in a recent large network meta-analysis of NSAIDs (6).

Rofecoxib and valdecoxib were voluntarily withdrawn from the US market in 2004 and 2005, respectively, following the observation of an increased risk of serious adverse CV events (8). Etoricoxib and lumiracoxib were never available on the US market, as they were not approved by the Food and Drug Administration (FDA). Celecoxib is the only coxib NSAID available on the US market in 2012. Rofecoxib, valdecoxib, and lumiracoxib were withdrawn from the European Union (EU) market in 2004, 2005, and 2007, respectively. Etoricoxib and celecoxib are the only coxib NSAIDs available on the EU market in 2012.

The C-reactive protein (CRP) concentration is commonly used in RA as a biomarker of systemic inflammation and is included as a surrogate marker of disease activity (9, 10). Another aspect of CRP is its association with a risk of CV complications (11). Interventions leading to a reduction in the CRP level significantly reduce the incidence of the first major CV event and death from any cause (12). Although CRP is an independent predictor of death from CV disease in RA patients (13), an association between a reduction in the CRP level and a drop in the number of CV events has not been shown for these patients. Although NSAIDs are referred to as being antiinflammatory, their effect on CRP levels in RA patients is still poorly documented in the literature (14–18). An effect of NSAIDs on CRP would influence the rheumatologist's evaluation by lowering the Disease Activity Score (19). Based on a systematic review and meta-analysis, the objective of the present study was to explore the empirical evidence for NSAID therapy as antiinflammatory compounds in RA patients according to changes in CRP levels, and with a prespecified focus on the different NSAIDs.

MATERIALS AND METHODS

The meta-analysis was performed according to the recommendations of the Cochrane Collaboration (20), and the findings are reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses guidelines (21). The protocol of the analyses and inclusion criteria was specified in advance and registered in the International Prospective Register of Systematic Reviews, PROSPERO 2011 (registration no. CRD42011001157).

Data sources and searches.

A comprehensive search from the middle 1950s to June 2011 was performed to identify parallel-group, randomized, placebo-controlled trials, in any language, in Medline via PubMed, the Cochrane Central Register of Controlled Trials, EMBase via OVID, and the Institute for Scientific Information Web of Science, by combining the search terms “NSAIDs” (identified NSAIDs with their unique drug names as text word or key words) and “RCTs” (methodological filter for randomized controlled trials) and “rheuma*” in the title (individual bibliographic searches are shown in Supplementary Table 1, available on the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). The reference lists of relevant retrieved articles were searched by hand for other relevant trials. Ongoing and completed trials were searched via www.clinicaltrials.gov, www.clinicalstudyresults.org, US FDA, and pharmaceutical companies' web sites. When feasible, the authors and companies were contacted to obtain unpublished data.

Study selection.

Inclusion criteria were 1) parallel-group, randomized, placebo-controlled trials; 2) RA patients meeting the 1987 revised criteria of the American College of Rheumatology (ACR) (22) or equivalent treated with any orally administered NSAID; and 3) reporting of CRP level. Cluster-randomized trials and crossover trials were not included due to statistical methodological issues unrelated to the level of CRP (23). There were no restrictions regarding concomitant medication (comedication), dosage, or duration of therapy. Although several of the coxib NSAIDs are no longer available on the market, we chose to include studies of these drugs in order to give the most elaborated picture of the association. There were no language restrictions.

Two reviewers (ST and EMB) screened all the references by title and abstract. Any that did not fulfill the criteria were rejected. Reports of the remaining studies were read in full and reviewed for eligibility. Any disagreement was resolved by discussion followed by a consensus meeting. The remaining studies were included in the meta-analysis.

Data extraction and quality assessment.

Using a standardized data extraction form, 1 reviewer (ST) extracted the data from the included trials and a second reviewer (RC) checked the extracted data. Information extracted included trial sponsors, country, trial duration, patient inclusion and exclusion criteria, diagnostic criteria, proportion of females, average disease duration, concurrent treatments, and type and dose of NSAID. Outcomes were determined in the intent-to-treat (ITT) population when possible, including mean ± SD changes in group CRP level (last value minus baseline value). In cases where SD was not available, SD was estimated with algebraic recalculations or various approximation methods (24).

Two independent reviewers (ST and RC) assessed the risk of bias using the following 3 criteria, as applied by Wood et al (25): 1) reported use of an appropriate randomization and concealment technique was considered adequate if the investigators could not discern next treatment allocation; 2) use of an appropriate blinding procedure was considered adequate if participants and caregivers were blinded to treatment allocation, and if it was unlikely that the blinding had been broken (blinding of outcome assessors responsible for analyzing CRP were not part of this assessment); 3) statistical analysis was considered adequate, including appropriate handling of missing data (i.e., proper ITT analysis, including all patients randomized). A criterion was judged adequate if risk of bias was low, inadequate if risk of bias was high, and unclear if risk of bias could not be determined. In case of disagreement, the decision was adjudicated by a third reviewer (HB).

Data synthesis and statistical analysis.

As the specific analyses in a meta-analysis are integral to understanding the results, we wished our report to be completely transparent. Therefore, we have described our statistical approach in detail. The treatment effects were expressed as standardized mean differences (SMDs) by dividing the difference in mean changes at the end of followup by the pooled SD of changes (26). SMD was chosen because changes in CRP levels were reported differently in many trials, and the SMD made it possible to compare CRP levels in a uniform manner. To aid interpretation of SMDs, we translated them back to their original units by multiplying the pooled SMD (20) with the SD from an observational Danish RA cohort (SD = 28 mg/liter, mean = 22 mg/liter) (27). Negative effect sizes indicate a beneficial effect (reduction in CRP level) of the NSAID intervention relative to placebo. Trials with multiple NSAID arms were treated as individual trials, referred to as “randomized comparisons” (i.e., 3-arm trials with 2 active interventions will generate 2 randomized comparisons with placebo). In studies with multiple arms, the numbers of patients in the placebo groups were divided by the number of active treatment arms, thus increasing the standard errors and avoiding double counting of patients and yielding more correct (conservative) estimates. Realizing that all of the (multiple) comparisons were based on the same overall model, we report 95% confidence intervals (95% CIs) with the corresponding P values without giving any explicit interpretation of statistical significance.

We conducted meta-analyses in Review Manager, version 5.1 (The Nordic Cochrane Centre, The Cochrane Collaboration) and SAS software (version 9.2 for Windows). A restricted maximum likelihood method was applied to estimate the between-study variance and the combined pooled analysis (26, 28). In the overall model, heterogeneity was evaluated between randomized comparisons, applying the standard Q statistic followed by the calculation of the I2 value (29), an inconsistency index which can be interpreted as the percentage of total variation across several studies due to heterogeneity (30). Quantitatively, when various statistical models were evaluated and compared, the between-comparison variance in random-effects meta-analysis was estimated as T2 (29). For the stratified analysis, a fixed factor was added to the model indexing subgroups across randomized comparisons. The stratified analysis was accompanied by interaction tests based on the Q statistics.

By definition, there is no heterogeneity (i.e., variation over and above chance) between comparisons within a trial with multiple comparisons. This means that considering individual comparisons (from the same study) as if they were independent trials could potentially lead to an underestimation of the between-trial heterogeneity. Since between-drug variability is a key concern in this report, this was taken into account by applying a within-trial pooled approach (ignoring dose) for sensitivity to address this potential caveat.

The following stratified and meta-regression analyses were prespecified to explore possible clinical differences according to effect sizes: 1) NSAID subgroups, 2) duration of treatment (weeks), 3) disease duration (years). A post hoc stratified analysis was conducted to explore possible clinical differences among doses according to Thomson Reuters Micromedex 1.0 (31), categorized as below, within, or above the recommended dose for RA.

Prespecified sensitivity analyses explored differences in effect size and the robustness of conclusions, as follows: 1) effect of risk of bias in included trials, defined as adequate handling of allocation concealment, blinding, and adequacy of the ITT analysis; 2) effect of various statistical approximations categorized according to the degree of statistical judgments needed (none/minor/major). Furthermore, due to the large number of identified unreported trials, a post hoc sensitivity analysis was conducted to explore differences in effect sizes according to data availability, categorized according to the following: CRP data reported in peer-reviewed scientific journals (published), data identified only via other online sources (online), or data obtained only via data transfer agreements with pharmaceutical companies (company data). Finally, a post hoc meta-regression analysis (32) explored the relationship between the log 80% inhibition concentration (IC80) ratio for COX-2:COX-1 inhibition and the changes in CRP level per randomized comparison. Values for COX inhibition were obtained from Warner and Mitchell (33).

RESULTS

Searches for eligible studies.

The bibliographic search identified 4,536 references, and 3,014 titles and abstracts were screened after removal of duplicates (Figure 1). Eighteen trials with CRP as a prespecified outcome were found eligible for inclusion in the systematic review. Of these 18 trials, 5 were not included in the meta-analysis. Two of these 5 trials had only graphic presentation of the CRP data in the publication, but the quality of the graphs was poor and reliable data could not be extracted (34, 35), and 3 had no numerical CRP data reported in the results section and data could not be identified through other sources (36–38). Thirteen trials were included in the meta-analysis. Data from 9 trials came from published work (39–47). From 1 trial (48), more complete CRP data were provided by the manufacturer. From another trial (49), more complete CRP data were identified online from Drugs@FDA Search for FDA-approved drugs (50). In the last 2 trials, no numerical CRP data were reported in the published results section, but data were obtained via other sources, and we were therefore able to include these trials in the meta-analysis; in 1 trial (51), CRP data were obtained online from the UK Medicines and Healthcare Products Regulatory Agency (52), and in the other trial (53), CRP data were provided by the manufacturer.

Figure 1.

Trial flow diagram. The meta-analysis was performed according to the recommendations of the Cochrane Collaboration and reported according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) guidelines. Shown is the PRISMA flow chart of the identification of eligible randomized controlled trials included in the meta-analysis. CENTRAL = Cochrane Central Register of Controlled Trials; incl. crit. = inclusion criteria; CRP = C-reactive protein; pt. = patient.

Searches in other databases and relevant reviews identified the following extra studies. Three trials were identified but could not be included in the meta-analysis owing to the company's lack of ability to provide data; these included 1 trial of a new selective COX-2 inhibitor (GW406381) (54) and 2 placebo-controlled trials of lumiracoxib compared to naproxen (55) or diclofenac (56). Six additional trials were identified that could be included in the meta-analysis. These included 2 trials (57, 58) with CRP data provided by the manufacturer; 1 trial with CRP data obtained online from the manufacturer's clinical trial results database (59); 1 trial with CRP data obtained online from Drugs@FDA Search for FDA-approved drugs (60); 1 trial (61) initially excluded from the main search because CRP was not reported as an outcome, with CRP data identified online from Drugs@FDA Search for FDA-approved drugs (50); and 1 Japanese trial (62), which was identified via a review.

Ultimately, a total of 19 trials were included in the meta-analysis, covering 10 different orally administered NSAIDs. Most of the included trials used multiple trial arms, enabling us to include a total of 54 randomized comparisons, presenting CRP data from 13,689 participants. This averaged to 3,148 patient-years in the overall ITT population.

Study characteristics.

The study characteristics and the risk of bias assessment of the 19 included trials in the meta-analysis are presented in Table 1. Trials were reported between 1993 and 2007. Patients were ≥18 years of age, had RA according to the ACR criteria (22) or equivalent, and had a disease duration of >3 months. The mean trial duration was 10 weeks (range 2–26 weeks). The mean disease duration was 10 years (range 8–11 years). On average, 78% of the patients were female (range 69–91%). Except for 1 trial (46), all trials were sponsored by the pharmaceutical industry. According to the Thomson Reuters Micromedex 1.0 recommended dosage for RA, 10 of the 54 randomized comparisons used a dose above the recommended dose, 39 used the recommended dose, and 5 used a dose below the recommended dose.

Table 1. Study characteristics and risk of bias assessment of randomized, placebo-controlled trials included in the meta-analysis*
Author, year (ref.)Trial duration, weeksNo. of patients randomizedDisease duration, years% femaleRisk of biasNSAIDDaily dose, mgNo. of patients receiving interventionNo. of patients receiving placebo
  • *

    All trials included only patients ≥18 years of age with rheumatoid arthritis (RA) and with a disease duration of >3 months. All patients received at least 1 disease-modifying antirheumatic drug along with nonsteroidal antiinflammatory drug (NSAID). NA = not available.

  • Shown as randomization and concealment of allocation/blinding of participants and caregivers/incomplete outcome data. A = adequate (low risk of bias); I = inadequate (high risk of bias); U = unclear risk of bias.

  • ↔ = within; ↑ = above; ↓ = below recommended dose for RA.

  • §

    C-reactive protein (CRP) data were not available for this randomized comparison.

  • CRP data provided by the manufacturer.

  • #

    The year 2002 is the date of the clinical trial report.

Gibofsky et al, 2007 (45)125081179A/A/IValdecoxib10↔170171
     Naproxen1,000↔167 
Wong et al, 2007 (46)237969U/U/IRofecoxib25↔1212
     Indomethacin150↔13 
Abe et al, 2006 (62)43741185A/U/UCelecoxib50↓9895
      Celecoxib200↔89 
      Celecoxib400↔92 
Williams et al, 2006 (49)121,0931176U/A/UValdecoxib10↔226220
     Valdecoxib20↑219 
      Valdecoxib40↑209 
      Naproxen1,000↔219 
Geusens et al, 2004 (51)261,1241079A/A/ALumiracoxib200↔280284
     Lumiracoxib400↔281 
      Naproxen§1,000↔279 
Bensen et al, 2002 (61)121,0901077A/U/AValdecoxib10↔209222
     Valdecoxib20↑212 
      Valdecoxib40↑221 
      Naproxen1,000↔226 
Collantes et al, 2002 (43)12891882A/U/IEtoricoxib90↔353357
     Naproxen1,000↔181 
Furst et al, 2002 (42)128941076A/U/IMeloxicam7.5↔175177
      Meloxicam15↔184 
      Meloxicam22.5↑177 
      Diclofenac150↔181 
Geusens et al, 2002 (48)121,058983A/U/IRofecoxib25↔315299
     Rofecoxib50↑297 
      Naproxen1,000↔147 
Matsumoto et al, 2002 (44)12816977A/U/IEtoricoxib90↔323323
     Naproxen1,000↔170 
Novartis, 2002 (59)#131,239878A/A/ILumiracoxib200↔315309
      Lumiracoxib400↔313 
      Celecoxib400↔302 
Curtis et al, 2001 (58)8581NANAA/U/UEtoricoxib10↓78123
     Etoricoxib60↓126 
      Etoricoxib90↔134 
      Etoricoxib120↔120 
Truitt et al, 2001 (57)12909NANAA/U/URofecoxib12.5↓148301
     Rofecoxib25↔311 
      Naproxen1,000↔149 
Schnitzer et al, 1999 (53)86581077A/U/URofecoxib5↓158168
     Rofecoxib25↔171 
      Rofecoxib50↑161 
Simon et al, 1999 (41)121,1491173A/A/ACelecoxib200↔240231
     Celecoxib400↔235 
      Celecoxib800↑218 
      Naproxen1,000↔225 
GD Searle, 1998 (60)121,1031073A/A/ACelecoxib200↔228221
      Celecoxib400↔219 
      Celecoxib800↑217 
      Naproxen1,000↔218 
Lemmel et al, 1997 (40)346810NAA/A/AMeloxicam7.5↔159147
     Meloxicam15↔162 
Mizushima et al, 1997 (47)4220891U/U/ILornoxicam12↔7073
     Lornoxicam18↑77 
Loose et al, 1993 (39)252NANAU/U/ANaproxen1,000↔2923

Effect on CRP level.

The meta-analysis of the 54 randomized comparisons and the prespecified stratified analysis according to the 10 different NSAIDs are presented in Figure 2, Table 2, and Supplementary Table 2 (available on the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131). The present meta-analysis of the 54 randomized comparisons identified a nonsignificant effect size (SMD 0.01 [95% CI −0.03, 0.06], P = 0.62) with an I2 of 14% inconsistency. Despite an apparently low degree of statistical heterogeneity, it is clear from Figure 2 that data were not consistent across drugs and randomized comparisons. For each NSAID subgroup, there was no clear dose dependence.

Figure 2.

Forest plot of 54 randomized comparisons and the stratified analysis according to the 10 different nonsteroidal antiinflammatory drugs (NSAIDs), analyzed using a random-effects model. Data are presented in ranking order with the most cyclooxygenase 2–selective NSAIDs at the top and the most cyclooxygenase 1–selective NSAIDs at the bottom (33). Within the NSAID subgroups, the randomized comparisons are listed according to the daily dose applied. CRP = C-reactive protein; I2 = inconsistency index.

Table 2. Results of all the stratified meta-analyses*
 kSMD (95% CI)T2P for interaction
  • *

    Analyzed using a restricted maximum likelihood meta–regression model. k = number of randomized comparisons (data points); SMD = standardized mean difference; 95% CI = 95% confidence interval; T2 = estimated between-study variance; NSAID = nonsteroidal antiinflammatory drug.

  • P < 0.05.

All trials540.01 (−0.03, 0.06)0.00560.624
NSAID vs. placebo  0.00330.112
 Lumiracoxib40.13 (0.01, 0.25)  
 Rofecoxib80.08 (−0.03, 0.18)  
 Etoricoxib60.03 (−0.09, 0.15)  
 Valdecoxib7−0.05 (−0.17, 0.07)  
 Meloxicam50.07 (−0.07, 0.22)  
 Celecoxib100.01 (−0.10, 0.11)  
 Diclofenac10.14 (−0.21, 0.50)  
 Lornoxicam2−0.06 (−0.41, 0.30)  
 Naproxen10−0.11 (−0.20, −0.02)  
 Indomethacin1−0.30 (−1.30, 0.71)  
Randomization and concealment of allocation  0.00590.112
 Adequate450.01 (−0.03, 0.06)  
 Unclear9−0.02 (−0.17, 0.12)  
 Inadequate0  
Blinding  0.00570.711
 Adequate210.02 (−0.05, 0.09)  
 Unclear330.00 (−0.06, 0.06)  
 Inadequate0  
Intent-to-treat analysis  0.00570.112
 Adequate17−0.01 (−0.08, 0.07)  
 Unclear170.09 (0.00, 0.18)  
 Inadequate20−0.02 (−0.09, 0.04)  
Statistical approximations  0.00450.092
 None380.04 (−0.01, 0.10)  
 Minor4−0.10 (−0.22, 0.02)  
 Major12−0.01 (−0.10, 0.08)  
Data availability  0.00600.284
 Published31−0.01 (−0.07, 0.05)  
 Online130.00 (−0.08, 0.08)  
 Company data100.09 (−0.02, 0.20)  
Recommended daily dose  0.00590.554
 Below50.10 (−0.07, 0.27)  
 Within390.00 (−0.05, 0.05)  
 Above100.02 (−0.08, 0.13)  

Compared to the meta-analysis of the 54 randomized comparisons, the prespecified stratified analysis of the 10 different NSAIDs showed a decrease in the between-study variance (T2 changed from 0.0056 to 0.0033), demonstrating a better model fit per se; thus, a model investigating each NSAID separately is less heterogeneous. Lumiracoxib statistically and consistently (I2 = 0%) increased the CRP level (SMD 0.13 [95% CI 0.01, 0.25], P = 0.037), corresponding to an average increase in CRP level of 3.6 mg/liter (i.e., 16% increase from baseline), whereas naproxen statistically and consistently (I2 = 0%) decreased the CRP level (SMD −0.11 [95% CI −0.20, −0.02], P = 0.022), corresponding to a decrease in the CRP level of 3.1 mg/liter (i.e., 14% decrease from baseline). Effect sizes were not statistically significant for the other NSAIDs. However, etoricoxib, with an I2 of 73%, gave inconsistent results; of the 3 extracted studies (43, 44, 58), 2 of them (43, 58) demonstrated a trend toward an increase in the CRP level, and the third (44) demonstrated a clear decrease in the CRP level. The 2 studies showing seemingly opposite effects (43, 44) were replicate phase III trials, identically designed, using the same laboratory for measuring CRP level.

Sensitivity and post hoc analyses.

Table 2 presents some additional results from stratified analyses. Sensitivity analyses included examination of whether the individual study's risk of bias was judged as being either low or high. We also assessed the potential influence of different dosing regimens and data availability. None of these analyses showed statistically significant effects (i.e., the between-study variability [T2] was not reduced). However, different statistical assumptions reduced the between-study variance (T2 changed from 0.0056 to 0.0045). The prespecified meta–regression analysis regarding disease duration resulted in an increase in T2 (0.0061), whereas meta–regression analysis using trial duration as a covariate resulted in a minor decrease in T2 (0.0055). This indicates a slightly worse model in the former and a marginally improved statistical model in the latter. The meta–regression analysis exploring the relationship between the log IC80 ratio for COX-2:COX-1 inhibition and the changes in CRP level per all 54 randomized comparisons indicated that decreasing COX-2 selectivity reduced CRP level (slope β = −0.044 [95% CI −0.073, −0.014], P = 0.0030, T2 = 0.0032) (Figure 3).

Figure 3.

Results of meta–regression analysis exploring the relationship between the log 80% inhibition concentration (IC80) ratio for cyclooxygenase 2 (COX-2):COX-1 inhibition and the changes in C-reactive protein (CRP) level per randomized comparison. Shown are standardized mean differences (SMDs) plotted according to the anticipated COX-2:COX-1 selectivity ratio (log IC80 COX-2:COX-1) (33) for each of the 10 nonsteroidal antiinflammatory drugs. The surface area of the circles is proportional to the precision and thus the weight of the estimate used in the meta–regression analysis; the regression line illustrates the predicted association (slope β = −0.044, P = 0.0030).

Analyses using the within-trial pooled approach corroborated the above analysis, although results were less robust. For lumiracoxib, the 95% CI overlapped with the null hypothesis, although the point estimate still supported a potential increase in CRP level. Naproxen, however, was consistently associated with a decreased CRP level that was independent of the statistical model used (see Supplementary Figure 1, available on the Arthritis & Rheumatism web site at http://onlinelibrary.wiley.com/journal/10.1002/(ISSN)1529-0131).

DISCUSSION

The present meta-analysis, which was based on 54 randomized comparisons, addresses CRP level as a surrogate for the antiinflammatory properties of NSAIDs in RA patients. Overall, there was no change in CRP levels after NSAID treatment (P = 0.62). Some NSAIDs appeared to have a small effect on the CRP level, but only lumiracoxib and naproxen showed statistically significant results. The empirical evidence for the selective COX-2 inhibitor lumiracoxib showed an increased CRP level (16%), whereas the nonselective NSAID naproxen consistently decreased the CRP level (14%). This raises the question of whether this difference has any clinical meaning: is there a relationship between the divergent effects of the COX-2 inhibitor lumiracoxib and the nonselective COX inhibitor naproxen on the CRP level and their effects on the CV system? There is some indirect evidence that the answer may be yes.

First, there is a relationship between the CRP level and CV risk (11). This effect is thought to be mediated by an effect of CRP on disease pathogenesis through various mechanisms (14). COX-2 expression is induced in atherosclerotic tissue (63–65), and inhibition of COX-2 could decrease the inflammatory condition of atherosclerosis (66, 67). Also, a reduction in inflammation status is associated with a reduced risk of death (68). Second, COX-2 inhibitors such as lumiracoxib increase the incidence of CV complications, while naproxen (a nonselective NSAID) is associated with significantly fewer CV events than are COX-2 inhibitors (1, 4–7). In a meta-analysis of 18 RCTs comparing selective COX-2 inhibitors against naproxen, the overall pooled odds ratio of myocardial infarction was 1.93 (95% CI 1.22, 3.05) (P < 0.05) in favor of naproxen (69). In another meta-analysis of observational studies, the effect of naproxen compared to non-naproxen NSAIDs supported the claim that naproxen has a cardioprotective effect compared to non-naproxen NSAID treatment (7). The rate ratio of myocardial infarction was 0.86 (95% CI 0.75, 0.99) in favor of naproxen (7). Another recent network RCT meta-analysis (6) showed that naproxen did not increase or decrease CV complications. Finally, it has been proposed that the excessive CV complications seen with many NSAIDs may be related to an imbalance in the inhibition of the 2 COX enzymes, with increased CV complications in proportion to COX-2 selectivity (70, 71), although this has not been clearly proven in terms of specific clinical end points (6).

Of course, we cannot rule out the possibility that other mechanisms associated with NSAIDs could play a major role in relation to CV complications. As summarized by Trelle et al, these mechanisms could be differential effects on prostacyclin and thromboxane A2 synthesis, blood pressure, endothelial function and nitric oxide production, volume retention, and other renal effects (6). Taken together, it is not unreasonable to extrapolate from an effect on CRP level to an effect on CV outcomes, although the connections remain somewhat speculative.

Several limitations of the present review should be considered. First, the included patients were all treated with a broad range of comedications (e.g., glucocorticoids and disease-modifying antirheumatic drugs), which could influence CRP level (17, 72). The effect of NSAID treatment on CRP level may therefore be seen as an effect of synergy with comedication. Only 19 RCTs were eligible for inclusion in the meta-analysis. This is a consequence of including only placebo-controlled (add-on) trials that provided suitable and unbiased CRP data for meta-analysis. Two RCTs included in the systematic review gave incomplete CRP data and were not included in the meta-analysis. Three further RCTs were not included in the meta-analysis because companies were not able to provide data, and data could not be identified via other sources. Selective outcome reporting could therefore be a limitation in the present meta-analysis (73). Since no RCTs of ibuprofen fulfilled the inclusion criteria, the present meta-analysis is not representative of all NSAIDs. Another potential caveat regarding the interpretation of the present meta-analysis is that these results concern RA only. A study of osteoarthritis patients, which directly compared the effects of etoricoxib, celecoxib, and ibuprofen versus placebo, found comparable results for serum CRP level for these 3 NSAIDs (74). Although those results are not entirely inconsistent with the findings of the present meta-analysis, our results may not be generalizable beyond the disease that was investigated, RA.

A strength of this meta-analysis is that the CRP level is an objective outcome measure and is therefore not affected by allocation and concealment or by blinding of participants/caregivers in the trials (25). The stratified analyses confirmed this, and it seems unlikely that CRP changes were affected by different levels of risk of bias among the RCTs in the present meta-analysis. A further strength of our data is the comprehensive search for eligible trials. The search in other online sources available in the public domain (such as reports in FDA and company online records) gave data of acceptable quality that were not otherwise published in peer-reviewed journals. Missing data were also obtained via direct contact with involved pharmaceutical companies. These time-consuming processes added further data to our final data set. We are therefore convinced that we have used all available sources possible for our analyses.

We conclude that NSAIDs as a group have no effect on the CRP level, although they are analgesic and may be antiinflammatory in RA assessed by parameters other than the CRP level. On the other hand, it is important to consider GI and CV risk factors when choosing the optimal NSAID for a particular patient. When dealing with a patient who has both GI and CV risk factors, using nonselective NSAIDs in combination with a gastroprotective agent may be a rational choice (1). Given the proposed cardioprotective properties of naproxen, and based on the present analysis, we suggest that this nonselective NSAID could be a potential choice for patients with cardiac risk factors in rheumatology practice.

AUTHOR CONTRIBUTIONS

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

Study conception and design. Tarp, Bartels, Bliddal, Furst, Boers, Danneskiold-Samsøe, Rasmussen, Christensen.

Acquisition of data. Tarp, Bartels, Christensen.

Analysis and interpretation of data. Tarp, Bliddal, Furst, Boers, Christensen.

ROLE OF THE STUDY SPONSOR

Mundipharma International had no role in the study design, data collection, data analysis, data interpretation, writing of the manuscript, agreement to submit the manuscript for publication, or approval of the content of the submitted manuscript. Publication of this article was not contingent upon the approval of Mundipharma International.

ADDITIONAL DISCLOSURE

Merck & Company, Inc. provided us with unpublished data and reviewed the manuscript. However, comments provided by Merck & Company, Inc. were discretionary only. Merck & Company, Inc. had no influence on the data or on the conclusions drawn by the authors.

Ancillary