Adverse Effects of Cyclooxygenase 2 Inhibitors on Renal and Arrhythmia Events:  Meta-analysis of Randomized Trials FREE

It is estimated that more than 30 million people worldwide take nonsteroidal anti-inflammatory drugs (NSAIDs) daily for treatment of pain and inflammation.¹ Conventional NSAIDs block both cyclooxygenase (COX) enzyme isoforms, COX-1, and COX-2.^2- 4 However, COX-1 inhibition by conventional NSAIDs is known to cause toxic gastrointestinal (GI) effects,^5- 6 as well as exert adverse renal effects,⁷ such as decreased renal perfusion, decreased glomerular filtration rate (GFR), edema, increased blood pressure, and interstitial nephritis.^8- 10 Overall, an estimated 2.5 million individuals in the United States annually experience adverse renal effects caused by use of NSAIDs.¹¹

With decreased risk of adverse GI effects, a class of drugs that selectively inhibits COX-2 enzyme was introduced for analgesia and the treatment of arthritis. Rofecoxib and celecoxib were the first 2 members of selective COX-2 inhibitor class that were approved by the US Food and Drug Administration (FDA) in December 1998 and May 1999, respectively. However, rofecoxib was recently recognized to significantly increase risk of myocardial infarction.^12- 14 Accumulating evidence on the cardiovascular safety of many COX-2 inhibitors has been systematically studied and suggests a potential class effect of COX-2 inhibitors on cardiovascular risk.^14- 20

Although the nephrotoxic potential by COX inhibition via either nonselective NSAIDs or selective COX-2 inhibitors has also been recognized, the adverse renal effects of selective COX-2 inhibition are less clear.²¹ By October 2000, the FDA had received 233 reports of renal failure associated with the administration of rofecoxib and celecoxib.²² Some studies showed that selective COX-2 inhibitors had mild and transient effects on renal blood flow and sodium excretion in salt-depleted or elderly patients.^23- 25 Other trials showed COX-2 inhibitors may increase blood pressure.^26- 28 Due to inconsistent results from individual trials,¹¹ the renal safety profile of selective COX-2 inhibitors has yet to be clearly documented. Additionally, the evidence for potential effects of COX-2 inhibitors on risk of arrhythmia, a possible cardiorenal-related event, also has not been clearly documented.

To comprehensively quantify adverse risks and potential class effect for renal outcomes and arrhythmia, we therefore conducted a comprehensive class-wide meta-analysis of randomized double-blinded trials of the COX-2 inhibitors: rofecoxib, celecoxib, valdecoxib, parecoxib, etoricoxib, and lumiracoxib. We also conducted time-cumulative meta-analysis to assess the temporal consistency and robustness of the evidence for the adverse effects on renal and arrhythmia events.

We conducted a systematic search of EMBASE and MEDLINE (through June 2006) for double-blind randomized clinical trials of the COX-2 inhibitors: rofecoxib, celecoxib, valdecoxib, parecoxib, etoricoxib, and lumiracoxib. We expanded our search for trials in the Cochrane Controlled Trials Register, the Computer Retrieval of Information on Scientific Projects (CRISP) database of the National Institutes of Health, direct contact with relevant investigators, references of retrieved articles, relevant FDA reports, and the online clinical trial information centers, and trial result repositories (http://www.ClinicalTrialResults.org and http://www.ClinicalStudyResults.org.).

We focused on studies with renal end points of interest (peripheral edema, hypertension, renal dysfunction) and arrhythmia. We excluded pooling analyses of other adverse events and any trials with no control group, no relevant events in either drug or control group, not double-blinded, abnormal baseline renal function, and combined simultaneous intervention of more than 1 COX-2 inhibitor. For multiple reports of a same study, the duplicate trial report from FDA and pharmaceutical company sources was used to supplement published data when applicable, with FDA data given precedence. From the systematic search, 114 informative reports were included in this analysis. The study selection process is summarized in Figure 1.

Using a standardized data extraction form, 2 independent investigators (J.J.Z. and E.L.D.) extracted and tabulated all data. Discrepancies were resolved via referencing the original report(s) and group discussions. Information extracted include first author, publication year, mean age of participants, sex proportion, comorbidity status (osteoarthritis, rheumatoid arthritis, other pain, hypertension, coronary heart disease, cancer, neurological dysfunction, healthy status), trial duration, type of COX-2 inhibitor agent, type of control(s), drug dose (weighted average of doses if multiple arms of the same COX-2 inhibitor were included in the same trial), number of participants in drug and control groups, and number of events of interest (peripheral edema, hypertension, renal dysfunction, and arrhythmia) in the drug and control groups.

The end point of hypertensive events included onset of hypertension, clinically recognized increased blood pressure event, or clinically recognized aggravated hypertension event as defined using routine clinical criteria by each trial ascertained via double-blinded manner. The renal dysfunction end point included significant changes of serum urea or creatinine levels as defined by the trial, clinically diagnosed kidney disease, or renal failure. The arrhythmia end point included atrial fibrillation, ventricular fibrillation, tachycardia, cardiac arrest, sudden cardiac death, or unspecified arrhythmia. However, bradycardia, an arrhythmia event of low clinical significance, was not included as an arrhythmia end point.

Relative risks (risk ratios, RRs) by an intent-to-treat analytic approach were calculated as the measure of association. Natural log transformations were performed on the RR measures. Relative risks in each analysis were pooled via DerSimonian and Laird random-effects models.²⁹ All analyses were stratified by each COX-2 inhibitor. Because oral valdecoxib and intravenous parecoxib share the identical functional compound, analyses were conducted pooling these 2 agents. Because peripheral edema, hypertension, and renal dysfunction share common pathophysiologies and are already well recognized to be adverse renal effects of conventional NSAIDs, a composite renal end point was formed. Our primary analyses examined pooled RRs of each COX-2 inhibitor vs controls with risk of composite renal events and arrhythmia events. Secondary analyses examined renal event-specific RRs of COX-2 inhibitor vs controls.

Additionally, we conducted subgroup analyses for composite renal end point using meta-regressions to assess potential effect modification by type of control (placebo, NSAID, mixed/other; other included aspirin, acetaminophen, rizatriptan, doloteffin, morphine, or salicin), drug dose (dosage dichotomy for each drug; eg, rofecoxib >25 vs ≤ 25 mg/d; celecoxib >400 vs ≤400 mg/d), duration (≥6 vs <6 months), mean age (≥65 vs <65 years), sex, and comorbidity (osteoarthritis, rheumatoid arthritis, other morbidity). A sparse number of trials precluded further informative stratification for other comorbidites and for specific arrhythmia subtypes. In addition, a time-cumulative meta-analysis was conducted for any COX-2 inhibitor found with increased risk of renal events and arrhythmia. Analyzed by calendar year of trial report, sequential series of cumulative random-effects pooling were conducted to identify the earliest calendar year, at the end of which, the increased risk of the specified end point became significantly apparent.

Conventional random-effects weighting was used in all analyses. To avoid statistical duplication of data, multiple control groups in a trial were collapsed as 1 independent control group for comparison to a COX-2 inhibitor agent. For studies reporting 0 events in a trial group, yet an informative trial as a whole, the classic half-integer correction was applied to calculate the RRs and variances. Further sensitivity analyses of arrhythmia data were conducted using Mantel-Haenszel–weighted pooling of trials and using the Fisher exact test of overall number of events and participants,³⁰ both established methods for sparse-events data without need of continuity correction.

To assess heterogeneity, the I² statistic^31- 32 was used to describe the percentage of total variation across studies that is due to heterogeneity rather than chance; a value of 0% indicates no observed heterogeneity, while larger values between 0% and 100% show increasing heterogeneity. Although we included unpublished trials in our analysis, we used both the Egger test and Begg test for assessment of potential publication bias³³ and examined relative symmetry of individual study estimates around the overall estimate using Begg funnel plots in which log RRs were plotted against their corresponding standard errors, stratified by COX-2 inhibitor.

All primary analyses were conducted using STATA version 8.2 (STATACorp, College Station, Tex); Mantel-Haenszel pooled analyses were conducted using EPISHEET version 2003 (developed by Rothman³⁴). All tests were 2-sided; P≤.05 was considered statistically significant.

The analysis included 114 informative reports, consisting of 127 trial populations (40 rofecoxib, 37 celecoxib, 29 valdecoxib/parecoxib, 15 etoricoxib, and 6 lumiracoxib trials) and included 116  094 total participants (Table 1, Table 2, Table 3, and Table 4). There were a total of 6394 composite renal events (2670 peripheral edema, 3489 hypertension, 235 renal dysfunction) and 286 arrhythmia events. Descriptive summary statistics of participants, detailed number of renal events, age, sex, duration, dose, control type, and comorbidities for each COX-2 inhibitor are summarized in Table 5. The number of arrhythmia events is shown in Table 6.

Overall pooled analysis of composite renal events ignoring COX-2 inhibitor type indicated substantial heterogeneity (I² = 57, 95% CI, 47-64; P<.001). Different COX-2 inhibitor agent was an important explanatory factor for the observed heterogeneous effects (overall P for interaction by agent = .02), indicating no class effect. Based on stratification by type of COX-2 inhibitor, there was overall decreased and nonsignificant heterogeneity of renal effect estimates within drug categories, with exception of etoricoxib. Primary results are summarized in Table 7. Forest plots for COX-2 drug categorical component renal effects and for arrhythmia events are available at http://www.Cox2DrugReview.org, or by request.

Results indicated that those taking rofecoxib experienced significantly increased risks of composite renal events compared with those in the control groups (RR, 1.53; 95% CI, 1.33-1.76). For specific renal end points, rofecoxib robustly increased risk of all 3 sub–end points: peripheral edema (RR, 1.43; 95% CI, 1.23-1.66), hypertension (RR, 1.55; 95% CI, 1.29-1.85), and renal dysfunction (RR, 2.31; 95% CI, 1.05-5.07). Although there was suggestion that valdecoxib plus parecoxib was associated with composite renal events (RR, 1.24; 95% CI, 1.00-1.55; P = .054), other COX-2 inhibitors overall were not significantly associated with increased renal events. Furthermore, celecoxib appeared to be associated with lower risk of both renal dysfunction (RR, 0.61; 95% CI, 0.40-0.94) and hypertension (RR, 0.83; 95% CI, 0.71-0.97) compared with controls. The I² statistic did not show significant heterogeneity for any individual events.

In addition, the recent Pre SAP celecoxob trial for prevention of colorectal adenoma recurrence¹¹¹ (which was published after our study’s protocol review period) suggested renal risks but a sensitivity analysis further incorporating the pre SAP trial results still consistently indicated a lack of COX-2 inhibitor renal class effect (P=.02) and no overall risk of composite renal events across 38 celecoxib trials (RR, 1.06; 95% CI, 0.90-1.24).

For arrhythmia events, only those taking rofecoxib were at increased risk compared with controls (RR, 2.90; 95% CI, 1.07-7.88). Despite relatively few arrhythmia events in rofecoxib trials, results were robust, for sensitivity analysis via Mantel-Haenszel pooling and Fisher exact test, methods more exact and optimal for sparse-events data, both further affirmed the increased risk of arrhythmia with rofecoxib (Mantel-Haenszel RR, 6.52; 95% CI, 1.48-28.8; Fisher exact P value = .004). Celecoxib showed no effect on arrhythmia, whereas valdecoxib plus parecoxib were related to a marginally lower arrhythmia risk. Additionally, results for etoricoxib and lumiracoxib for risk of arrhythmia were inconclusive due to limited number of available trials.

Potential sources of between-trial heterogeneity were further explored via meta-regression stratified analyses; lumiracoxib was not further stratified due to limited studies. Stratified results are summarized in Table 8. Consistently, increased risk of renal events was evident for rofecoxib regardless of comparison to either placebo (RR, 1.70; 95% CI, 1.35-2.14), nonselective NSAIDs (RR, 1.32; 95% CI, 1.08-1.61), or mixed comparison of other controls (RR, 1.68; 95% CI, 1.31-2.15); rofecoxib effects vs placebo control had a nonsignificant trend toward being stronger than NSAID controls (P = .10 ). Moreover, the stratified results suggested that both higher dose (>25 mg/d) and longer trial duration (≥6 months) may potentially further increase risk of renal events (P = .05 and P = .006, respectively); results also suggested that adverse effects of rofecoxib may be stronger among rheumatoid arthritis patients than other patients (P<.001). Stratified results for celecoxib, valdecoxib plus parecoxib, and etoricoxib did not reveal significant effect modifications.

In time-cumulative analyses by calendar year of trial report to examine earliest year of apparent adverse effects, the trial evidence showed that it was evident by the end of the year 2000 that rofecoxib was associated with overall adverse renal events (P<.001; and for all subsequent years), as well as specific events of hypertension and peripheral edema (P<.01 for both; and for all subsequent years); the evidence for adverse effects on renal dysfunction became apparent by 2005 (Figure 2). Furthermore, rofecoxib's adverse effects on arrhythmia first became significant by the end of year 2004 (P = .05), and became further apparent by the end of 2005 (P = .04; Figure 2).

In this comprehensive meta-analysis of 114 randomized trials of COX-2 inhibitors comprised of 116 094 participants, rofecoxib uniquely increased risks of renal events (peripheral edema, renal dysfunction, hypertension) and arrhythmia events, with apparent adverse effects by the end of year 2000 and 2004, respectively. However, the results did not show adverse effects of other COX-2 inhibitors on renal events and arrhythmia, indicating no overall evidence for a COX-2 inhibitor class effect.

The underlying mechanisms for potential nephrotoxic effects of selective COX-2 inhibitors are not fully understood, although several explanations have been proposed. In the most basic framework, conventional NSAIDs block both COX-1 and COX-2 enzyme isoforms, whereas selective COX-2 inhibitors primarily antagonize COX-2. Prostaglandins (PGs), derived from arachidonic acid by various COX enzymes, can function as important mediators of inflammation and modulate a variety of physiological processes, including maintenance of gastric mucosal integrity, renal hemodynamics, renin synthesis and release, and tubular reabsorption of sodium and water.¹¹¹ Because PGs are involved in renal function, nonselective NSAIDs are already recognized to exhibit adverse renal effects, including renal failure due to hemodynamic changes, acute tubular necrosis, tubuolinterstitial nephritis, or papillary necrosis, as well as disturbances and disorders of electrolyte balance (hyponatremia, hyperkalemia) such as hypertension and edema.^8- 10

Although COX-2 was originally designated as an “inducible” enzyme because of its up-regulation by inflammatory and proliferating stimuli, COX-2 is constitutively expressed in adult mammalian kidney tissues, including the cortex, macula densa, thick ascending limb, interstitial cells in inner medulla, and papilla and podocyte.^113- 116 COX-2 expression has been found to be up-regulated in renal ischemia, salt depletion status, which suggests COX-2-derived prostanoids may play a role in maintaining renal medulla blood supply, renal salt excretion, and systemic blood pressure.^113- 116 Furthermore, prostacyclin synthesis and renin-release from the kidney have also been found to be COX-2 dependent.¹¹⁷ Some evidence indicate that specific COX-2 inhibition may induce renal ischemia, electrolyte imbalance, and abnormal blood pressure,⁸ ultimately leading to fluid and sodium retention as well as decreased GFR.^{23,25,118- 119}

However, a class effect for all COX-2 inhibitors has previously remained unclear.¹²⁰ Although all study agents are considered selective COX-2 inhibitors, the various agents are known to somewhat differ in terms of potency and specificity of COX-2 inhibition as well as their respective pharmacokinetics and metabolism in the human body.¹²¹ These characteristics may contribute to observed differences of the effect across various COX-2 inhibitors. Consistent with evidence that rofecoxib has stronger COX-2 selectivity than other COX-2 inhibitors,¹²⁰ our results indicate substantially different effects between rofecoxib and other agents in its class. Notably, we found a higher risk of renal dysfunction, hypertension, and peripheral edema with rofecoxib, whereas in contrast, we found lower risk of hypertension and renal dysfunction with celecoxib and no effect with other agents. Although this finding may be due to chance, our results are supported by multiple animal studies that showed rofecoxib and celecoxib to have different or opposite effects on blood pressure, glomerular injury, inflammatory reaction, and endothelial dysfunction—all of which rofecoxib was generally more adverse compared with celecoxib.^122- 124 Furthermore, clinical trials have also shown that rofecoxib elicits significantly greater blood pressure increases compared with celecoxib,²⁶ and that celecoxib may protect against the adverse renal effects of aspirin.¹²⁵ The degree and relative inhibition of different COX isoforms may be important, with modest differences between agents in the relative potency and selectivity of COX-2 vs COX-1 inhibition potentially mediating differential effects. Our results do not support a class effect of COX-2 inhibitors on renal events.

Our study is the first to our knowledge to highlight that rofecoxib may increase risk of arrhythmia and appears to differ in arrhythmia effects than other COX-2 inhibitors, although the exact mechanisms underlying such effects remain uncertain. Although myocardial infarction is a frequent cause of arrhythmia, the recognized cardiovascular class effect of rofecoxib, celecoxib, and valdecoxib^{12,14- 17,19- 20} suggests that cardiac events might not be the underlying explanation for the differential arrhythmia effects of rofecoxib from other COX-2 inhibitors.

On the other hand, another recognized and established cause of arrhythmia is disturbance of electrolyte balance, especially hyperkalemia. Consistent with the divergent adverse renal effects of rofecoxib compared with other COX-2 inhibitors, these renal mechanisms may, at least in part, explain the parallel increased risk of arrhythmia uniquely associated with rofecoxib but not associated with other agents. Although there were suggestions of renal risk but lower risk of arrhythmia in valdecoxib and parecoxib trials, results were not statistically significant and these trials included mostly atrial arrhythmias, which are generally less fatal.

In contrast, the arrhythmia results for rofecoxib appear more clinically relevant and significant, as the vast majority of rofecoxib arrhythmia events were ventricular fibrillation, cardiac arrest, and sudden cardiac death (Table 6). However, due to limited number of trials and cases, insufficient power was available to further evaluate specific types of arrhythmia. Further mechanistic investigations are warranted to study the divergent effects of COX-2 inhibitors on renal events and subtypes of arrhythmia.

Our study has several limitations that merit consideration. First, although tests for publication bias did not reveal such biases for any drug, we cannot exclude the potential for bias in reporting drug-related adverse events, as recently highlighted in a controversy over reporting of rofecoxib cardiovascular events.^13,126- 128 However, we believe such reporting bias is minimal because our meta-analysis had supplemented publication data with FDA reports and unpublished pharmaceutical company trial safety reports whenever available. Moreover, because renal effects of conventional NSAIDs are already well recognized, renal events are unlikely to have influenced decisions for publication.

Second, although diagnostic criteria for hypertensive events and renal dysfunction sometimes varied by trial-specific criteria or were not clearly defined because these adverse events were not commonly prespecified, it is difficult to conceive that trials did not use conventional definitions for hypertension and renal dysfunction consistent with routine clinical criteria. In addition, besides the inherent advantage of random-effects models' in conservatively incorporating any potential heterogeneity, the informative power gained from such trials substantially outweighs the alternative of wastefully excluding the majority of these highly relevant trials. Furthermore, this study's definition of hypertensive events is also consistent with composite hypertension end points frequently reported in clinical trials. Most important, because all included trials were optimally double-blinded and evaluated with intent-to-treat analysis in our study, there would not be inherent biases in the ascertainment of any end point or bias of any trial-specific RR.

Third, due to less power in stratified analyses, we were limited in our ability to fully and clearly assess heterogeneity by variety of characteristics, such as underlying comorbidity, and evaluate age and sex differences beyond crude trial means and trial proportions. Also, our time-cumulative results are conservative, as pharmaceutical companies would have had data and results of all industry-sponsored trials available much earlier, likely months and years prior to publication, submission to FDA, or uploading on an online pharmaceutical industry database. Additionally, due to the limited number of trials of the newer drugs etoricoxib and lumiracoxib, results for these agents still remain tentative and inconclusive; additional safety monitoring investigations are needed.

Nevertheless, this systematic review and meta-analysis, overall, has the advantage of having a greater sample size to detect adverse effects,¹²⁹ which previous trials of COX-2 inhibitors were often underpowered to assess.¹³⁰ Furthermore, this comprehensive class-wide evidence-based assessment has the unique advantage of assessing class effects, which may help guide policy decisions regarding the overall safety of the drug class and the development of future drugs in this class.

Notably for policy and clinical decision making, our results also suggest that a time-cumulative meta-analytic approach for examining available trial safety data would have helped clarify apparent adverse effects several years earlier than the current report. The knowledge of all potential adverse effects is important and indeed time-sensitive, for physicians and patients both need complete information about risks and benefits to properly use COX-2 inhibitors and other clinical treatments.¹³⁰ However, the current system of postmarketing surveillance has been recognized to possess a variety of shortcomings, including an overall lack of vigilance.^131- 132 Therefore, as alluded to by other reviews of clinical trial safety^{130,132- 133} and supported by other teams of systematic review experts,^14,134 future drug safety monitoring of emerging clinical treatments may benefit from continuous, cumulative meta-analytic aggregation of safety data for all drug-approval applications and experimental agents. Of further benefit, the establishment of an independent postmarketing surveillance system based on such an active and continuous data aggregation structure would have many advantages over the current passive reporting system.^131,135

In conclusion, our analysis of 114 randomized trials involving 116 094 participants indicates that rofecoxib increased risk of renal events and arrhythmia events. Overall, a class effect was not evident for renal events and arrhythmia events across all COX-2 inhibitors, although further safety monitoring current and emerging treatments are warranted and may benefit from a cumulative and active surveillance system.