Efficacy of Rofecoxib, Celecoxib, and Acetaminophen in Osteoarthritis of the Knee:  A Randomized Trial FREE

Osteoarthritis (OA) is the most common joint disorder, accounting for significant disability and large health care expenditures.^1- 3 Although nonsteroidal anti-inflammatory drugs (NSAIDs) have long been used to treat the pain and stiffness associated with OA, the American College of Rheumatology guidelines published in 1995,⁴ and updated in 2000,⁵ recommended acetaminophen as first-line therapy for the systemic treatment of symptomatic OA. This decision was partly due to concerns about gastrointestinal tract and other adverse effects associated with NSAIDs and also due to lack of data confirming their superior efficacy over simple analgesics. However, the severity of pain often prompts treatment with NSAIDs, which remain commonly used and preferred medicines by many patients with symptomatic OA.^1,6- 7

The mechanism of action of NSAIDs involves inhibition of prostaglandin synthesis. In humans, prostaglandin synthesis is catalyzed by at least 2 forms of cyclooxygenase, cyclooxygenase 1 (COX-1) and cyclooxygenase 2 (COX-2). COX-1, constitutively expressed in a variety of tissues, is responsible for the production of prostanoids that regulate physiological functions such as platelet aggregation and gastric mucosal protection. Although it is constitutively expressed in the brain, COX-2 is typically only induced in most other tissues by cytokines and other soluble mediators.^8- 9 COX-2 has been detected in leukocytes and human rheumatoid synoviocytes, and it mediates synthesis of prostanoids generated in inflammation and pain.^10- 13

Most NSAIDs are dual inhibitors of COX-1 and COX-2 and therefore can be associated with gastrointestinal tract toxicity (such as perforation, ulcer formation, and gastrointestinal tract bleeding) due to COX-1–mediated reduction in protective prostanoids.¹⁴ Selective COX-2 inhibitors may suppress pathological responses mediated by prostanoids (eg, pain and inflammation) without inducing toxicity associated with the inhibition of COX-1. Rofecoxib and celecoxib selectively inhibit COX-2. Administration of these agents has been shown to provide relief of symptoms of OA^15- 18 with a reduced risk of gastrointestinal toxicity relative to dual COX inhibitors.^19- 21

The objective of this study was to estimate the efficacy of rofecoxib, celecoxib,¹⁸ and acetaminophen^4- 5 in adult patients with OA of the knee.

The study was conducted at 29 clinical centers in the United States from June 1999 to February 2000. Men and nonpregnant women with symptomatic OA of the knee for at least 6 months were eligible for study participation if they were at least 40 years old; fulfilled American College of Rheumatology clinical criteria for OA of the knee²²; and had an American College of Rheumatology functional class rating of I, II, or III.²³ The knee designated as the "study joint" was the primary source of pain or disability in the lower extremity.

Consistent with patient enrollment in previous studies,^15- 16 patients who entered the study were previous users of either a single, prescription-strength NSAID or high doses of acetaminophen for control of OA symptoms for at least 30 days prior to entry. Entry criteria for NSAID users included demonstrating, while taking the NSAID, a screening Western Ontario and McMaster Universities Osteoarthritis Index (WOMAC²⁴) visual analog scale (VAS) score of less than 80 mm for the assessment of walking pain at visit 1 (screening); a minimum VAS score of 40 mm for the assessment of walking pain after discontinuing the NSAID; an increase (ie, worsening) from screening in the walking pain VAS score of at least 15 mm, and a worsening from screening in the Investigator Global Assessment of Disease Status (IGADS) (performed by the physician investigator) of at least 1 point on a 5-point Likert scale (range, 0 [very well] to 4 [very poor]) at visit 2.

During the prespecified washout period, patients who had discontinued NSAIDs could take acetaminophen for intolerable pain, but acetaminophen was discontinued at least 24 hours before efficacy assessments. Since previous acetaminophen users were not taking NSAIDs, they were required to have a history of therapeutic benefit with regular doses of acetaminophen (1200-4000 mg/d) as exclusive therapy of OA, a score of 2 (fair), 3 (poor), or 4 (very poor) on IGADS, and the same minimum VAS score of 40 mm required of the NSAID previous users after discontinuing acetaminophen therapy for the assessment of walking pain at visits 1 and 2.

Patients were excluded from the study if they had a concurrent medical or arthritic disease or abnormal laboratory results (values outside normal reference range or determined by the investigator to be of clinical significance) that had potential to confound or interfere with the efficacy evaluation or pose an additional risk to the patient; history of allergy to study drugs, hypersensitivity to aspirin, ibuprofen, or any other NSAID, or sulfonamide-containing compounds; or received an investigational drug within 30 days of screening.

Consistent with the typical duration of other OA studies^15- 16 and previously published recommendations,²⁵ we conducted a double-blind, randomized, active comparator–controlled trial of 6 weeks' treatment comparing COX-2 inhibitors at indicated once-daily doses for treatment of knee OA: rofecoxib (12.5 mg/d and 25 mg/d), celecoxib, 200 mg/d, and the highest recommended dose of acetaminophen—4000 mg (1000 mg 4 times daily^4- 5). Patients who satisfied entry criteria discontinued their prior NSAID therapy according to a prespecified schedule (>5 plasma half-lives of prior NSAID). Patients were instructed to return to the site for visit 2 on significant worsening of knee pain or related symptoms, or at the end of their allowed washout period, whichever came first. For acetaminophen users, visit 2 was scheduled within 3 to 7 days of screening.

NSAID users were allowed to take 325-mg acetaminophen tablets during the washout phase as rescue therapy for OA pain at a daily dose of acetaminophen restricted to 2600 mg (8 tablets). No other rescue medication was allowed during the study. All patients discontinued use of acetaminophen at least 12 hours before visit 1 (screening) and visit 2 (day 1).

At visit 2, enrolled patients were randomly assigned (via computer-generated assignment) to treatment with either: rofecoxib 12.5 mg/d, rofecoxib 25 mg/d, celecoxib 200 mg/d, or acetaminophen 1000 mg 4 times per day for 6 weeks. Patients and investigators were blinded to treatment; exact-matching placebo tablets were used to blind the study. Patients took rofecoxib 12.5 or 25 mg or celecoxib 200 mg or 1000 mg of acetaminophen or matching placebo each morning from between 7 and 10 AM, and subsequently took matching placebo or acetaminophen 1000 mg 3 additional times to complete 4-per-day dosing.

Early efficacy data were collected on days 1 through 6 using patient take-home diaries. Preplanned analysis included determination of efficacy using the entire WOMAC Osteoarthritis Index Version VA 3.0 (a VAS ranging from 0 [best] to 100 mm [worst]; assessed by composite subscales and specified questions outlined) and patient global assessment of response to therapy (PGART) on a 5-point categorical scale (range, 0 [none, no response] to 4 [excellent response]). It was prespecified that responses to WOMAC pain walking on a flat surface, night pain, pain at rest and morning stiffness, and PGART would be recorded on take-home forms for days 1 through 6 and analyzed to determine efficacy over the first 6 days. Patients completed questions about night pain and morning stiffness prior to their first dose on days 2 through 6; questions about walking pain and pain at rest were completed at bedtime on days 1 through 6. Clinical efficacy (WOMAC and PGART) and safety measures (physical examination, and patient interview) were administered at scheduled office visits at weeks 2, 4, and 6. Patients completed the entire 24-item WOMAC and PGART at each study visit. Safety evaluation was based on physical examination, laboratory testing, and reporting of adverse events. Cardiovascular and adverse gastrointestinal tract events were adjudicated by a previously reported mechanism involving an external committee blinded to treatment.²⁰

The trial was designed to enroll approximately 200 patients who had used NSAIDs and a target of 100 to 200 patients who had used acetaminophen to ensure that each treatment group would have a minimum of 50 patients who had used NSAIDs and a target of 25 who had used acetaminophen. The analysis plan was first to evaluate the treatments within each subgroup to determine whether the treatment effects were consistent for each prior-use subgroup. Then, if consistent, the plan was to combine the prior-use subgroups to obtain overall estimates of treatment effect. As stated in the protocol, a between-group difference of 10 mm on the WOMAC VAS scale was the anticipated effect size between rofecoxib and acetaminophen. With 50 evaluable patients per treatment and subgroup, the half-width of a 95% confidence interval (CI) for a treatment mean would be 8.3 mm, assuming a within-group SD of 30 mm. With 75 (100) patients per group, the power to detect a treatment difference of 10 mm on the WOMAC scale was 52% (65%).

All analyses in this efficacy trial were conducted using a modified intent-to-treat approach, whereby all patients who took at least 1 dose of study medication were included for analysis. For the analysis of WOMAC data, a patient had to have a baseline value, which was the value obtained after washout from prior therapy and at least 1 on-treatment value to be included in the analysis. For PGART, only 1 on-treatment value was required. Missing values were imputed by carrying forward the most recent, previous, nonmissing value. Missing WOMAC data were not imputed nor carried forward. Only 1 patient, in the celecoxib group, did not receive treatment and could not provide on-treatment data.

For the efficacy analyses, all tests were 2-sided and P values ≤.050 were considered statistically significant. For each efficacy parameter, secondary analyses were conducted by previous user subgroup (NSAIDs or acetaminophen). Similar trends were observed in each user subgroup and therefore all results were summarized for the total patient population.

For the 4 WOMAC question scores and the 3 WOMAC composite subscales (ie, pain, stiffness, and physical function), analysis of covariance was used to assess statistical significance of treatment differences in 6-day and 6-week mean changes from baseline to determine 95% CIs and corresponding P values. The model included terms for baseline covariate and treatment group. Changes from baseline in the individual WOMAC question scores were analyzed for early efficacy assessments on days 1 through 6 (early efficacy) as well as for the entire 6-week treatment period. Six-week data were analyzed for changes from baseline in the 3 WOMAC subscales. A 6-day and 6-week average was calculated for each patient. The 6-week average for a patient was the mean of the change from baseline to weeks 2, 4, and 6. The 6-day average was calculated for each patient in the same way.

The percentage of patients with a good or excellent response to therapy (PGART) was calculated for each treatment group at weeks 2, 4, and 6 with week 6 prespecified to be primary. A logistic regression model was used for analysis of the PGART data, with estimates of odds ratios (ORs), corresponding P values, and 95% CIs. The model included terms for the IGADS at baseline and treatment group. A time-to-event analysis was conducted for the first report of a good or excellent response for first 6-day diary data; the cumulative incidence was calculated by the Kaplan-Meier estimate and compared by Wilcoxon test for ranked survival data.

Numbers and rates of adverse events, gastrointestinal tract symptoms (ie, acid reflux, dyspepsia, epigastric discomfort, heartburn, nausea, and vomiting), and events of special interest (ie, edema, hypertension and gastrointestinal tract bleeding, perforation, and ulceration) were tabulated for each treatment group.

Each patient provided written informed consent prior to enrollment. Institutional review board approval of the study protocol was obtained for each investigational site. Statistical analyses were conducted with SAS version 6.12 (SAS Inc, Cary, NC).

A total of 515 patients were screened; 133 did not meet entry criteria and 382 patients were enrolled (Figure 1). Treatment groups were comparable in age, race, sex, prior medication use, OA duration, and baseline IGADS and WOMAC scores (Table 1). Overall, 68.3% of patients were women, and 85.3% were white. Mean age was 62.6 years (range, 39-91 years). Seventy-seven percent of the patients used NSAIDs prior to study entry; 88 patients (23%) were previous acetaminophen users.

Overall, 79% of patients completed the study. Thirty-one percent in the acetaminophen group withdrew compared with 18% to 19% taking COX-2 inhibitors. Overall, lack of efficacy was the most commonly cited reason for withdrawal. For this reason, 17% of patients were withdrawn from the acetaminophen group vs 8% taking rofecoxib 25 mg and 12.5 mg and 9% taking celecoxib 200 mg in the other groups. A total of 23 patients (6%) discontinued due to a clinically adverse event (4%-7% per group).

At baseline, mean scores for all efficacy end points were similar among treatment groups. Over 6 days, mean decreases from baseline in the score for walking pain were greatest for rofecoxib 25 mg/d (32.2 mm) and smallest for acetaminophen (20.6 mm) (Table 2). Compared with acetaminophen, improvement in this score was statistically significantly greater in patients treated with rofecoxib, 25 mg/d (P<.001); rofecoxib, 12.5/d mg (P = .004); or celecoxib (P = .04). The difference between the rofecoxib 25 mg/d and celecoxib groups was statistically significant (P = .05). Over 6 weeks, the mean decreases from baseline for walking pain were 30.3 mm for the acetaminophen group and 42.0 mm for the rofecoxib 25 mg/d group (P = .001). There were no other statistically significant differences among groups.

Over 6 days, mean decreases from baseline in the average night pain score ranged from 25.2 mm for rofecoxib 25 mg/d to 18.7 mm for celecoxib (Table 2). Improvement in this score was statistically significantly greater in patients treated with rofecoxib 25 mg/d vs those treated with celecoxib (P = .04) or acetaminophen (P = .04).

Over 6 weeks, the mean decreases in the average night pain score ranged from 22.6 mm in the celecoxib group to 32.7 mm in the rofecoxib 25 mg/d group. The decrease in pain scores for those taking 25 mg/d of rofecoxib was statistically significantly greater than for those taking acetaminophen (P = .006), as well as for those taking rofecoxib 12.5 mg/d (P<.02) and celecoxib (P = .002).

Over 6 days, mean decreases from baseline in rest pain score ranged from 21.8 mm for rofecoxib 25 mg/d to 12.5 mm for acetaminophen (Table 2). Relief of rest pain was statistically significantly greater in patients treated with rofecoxib 25 mg/d vs those taking celecoxib (P = .02) or acetaminophen (P<.001); the decrease in rest pain score for those taking rofecoxib 12.5 mg/d was greater than for those taking acetaminophen (P = .02).

Over 6 weeks, mean decreases from baseline in rest pain score ranged from 21.7 mm for those taking acetaminophen to 31.1 mm for those taking rofecoxib 25 mg/d. The decrease in rest pain score for those taking rofecoxib 25 mg/d was statistically significantly greater than for those taking celecoxib (P = .02) and for those taking acetaminophen (P = .005).

Over 6 days, mean decreases from baseline morning stiffness scores ranged from 30.4 mm in the rofecoxib 25 mg/d group to 20.9 mm in the acetaminophen group. Mean reductions in this score were statistically significantly greater in patients treated with rofecoxib 25 mg/d (P = .003) and 12.5 mg/d (P = .02) compared with acetaminophen but there were no significant differences in scores between the celecoxib and acetaminophen groups or among the celecoxib and rofecoxib groups.

Over 6 weeks, mean changes from baseline in morning stiffness score ranged from 22.3 mm in the acetaminophen group to 36.2 mm in the rofecoxib 25-mg/d group (P<.001). The difference between effects provided by rofecoxib 12.5 mg/d and rofecoxib 25 mg/d was also statistically significant (P = .05). Differences between celecoxib and acetaminophen as well as differences among celecoxib and rofecoxib were not significant.

Improvements in the WOMAC pain, stiffness, and functional disability subscales in each treatment group during the 6-week treatment period are described in Table 2.

Across treatments, mean decreases from baseline in the 6-week domain score for pain ranged from 24.9 mm in the acetaminophen group to 35.4 mm in the 25-mg/d rofecoxib group. Improvement in this score was statistically significantly greater in patients treated with rofecoxib 25 mg/d vs those taking rofecoxib 12.5 mg/d (P = .02), celecoxib (P = .03), or acetaminophen (P = .001).

Mean decreases from baseline in the 6-week domain score for stiffness ranged from 21.6 mm in the acetaminophen group to 35.0 mm in the rofecoxib 25-mg/d group. Improvement in this score was statistically significantly greater in patients treated with rofecoxib 25 mg/d vs celecoxib (P = .04) or acetaminophen (P<.001); the difference between rofecoxib groups was not statistically significant (P = .051).

Improvement in the average 6-week functional disability score was statistically significantly greater in the rofecoxib 25-mg/d group (29.7 mm) than in the acetaminophen group (19.5 mm; P = .001); there were no other statistically significant differences among treatment groups.

By day 6, cumulative incidences of first report of a good or excellent response on PGART were 75% in the 25-mg/d and 71% in the 12.5-mg/d rofecoxib groups, 68% for celecoxib, and 54% for acetaminophen groups (rofecoxib 25 mg/d vs acetaminophen, P = .05; differences for all other comparisons are not statistically significant). Median time to first report of good or excellent PGART was day 3 for rofecoxib 25 mg/d, day 3 for celecoxib, day 4 for rofecoxib 12.5 mg/d, and day 6 for acetaminophen 4000 mg/d (rofecoxib 25 mg/d vs acetaminophen, P = .05; differences for all other comparisons are not statistically significant).

At week 6, the percentages of patients who had good or excellent response on PGART were 60% in the 25-mg/d and 56% in the 12.5-mg/d rofecoxib groups; 46%, celecoxib; and 39%, acetaminophen groups (Figure 2). A positive response was statistically significantly more likely in patients treated with rofecoxib 25 mg/d vs celecoxib (P = .03) or acetaminophen (P = .003). A good or excellent response rate in the rofecoxib 12.5-mg group was also statistically significantly higher than in the acetaminophen group (P = .02) but not significantly higher than in the celecoxib group. Compared with acetaminophen, both rofecoxib groups were more likely to experience a good or excellent response to therapy: rofecoxib 12.5 mg/d (OR, 2.05; 95% CI, 1.13-3.71; P = .02) and 25 mg/d (OR, 2.47; 95% CI,1.36-4.48; P = .003). The OR of celecoxib vs acetaminophen was 1.30 (95% CI, 0.72-2.36; P = .38). The OR of rofecoxib 25 mg/d vs celecoxib was 1.89 (95% CI, 1.05-3.40; P = .03) and for rofecoxib 12.5 mg vs celecoxib 200 mg was 1.57 (95% CI, 0.86-2.82; P = .10).

Table 3 and Figure 1 describe the clinical adverse event profile. Incidences of specific gastrointestinal events were generally low and comparable among treatment groups. No patient experienced gastrointestinal tract bleeding, perforation, or ulceration during the trial. A total of 7 patients (2%) had hypertension during the study (1-3 patients per group); none of these resulted in study withdrawal. Lower extremity, pedal, and ankle edema were the most common types of edema reported (0%-3.2% per group). Two patients (1, rofecoxib 25-mg/d; 1, acetaminophen) were withdrawn from the study due to edema (pedal or lower extremity). Two patients (1, rofecoxib 25 mg/d; 1, celecoxib) also experienced fluid retention; none of the patients had congestive heart failure. There were no myocardial infarctions in any group during the trial. One previous celecoxib user had received rofecoxib 25 mg/d in the study and had a stroke diagnosed 1 week after completing the trial and cessation of study therapy.

In this study, rofecoxib 25 mg/d provided greater therapeutic benefits than maximal daily doses of 4000 mg/d of acetaminophen in treating patients with OA of the knee for all prespecified end points and benefit over rofecoxib 12.5 mg/d and celecoxib 200 mg/d. Previous studies have failed to demonstrate convincingly therapeutic benefit of full anti-inflammatory doses of dual COX-1 and COX-2 inhibiting NSAIDs compared with acetaminophen in the treatment of OA. A 4-week study by Bradley et al,²⁶ involving 184 patients randomized to 3 treatment arms, concluded that the efficacy of acetaminophen in OA of the knee was similar to that of 2 different dosages of ibuprofen. Ibuprofen given at analgesic (1200 mg/d) and anti-inflammatory (2400 mg/d) dose levels provided improvements in the Health Assessment Questionnaire pain and disability scores that were indistinguishable from those obtained from patients treated with acetaminophen at 4000 mg/d (P>.90). Differences in improvement from baseline in pain walking did not reach statistical significance. However, the mean improvement in rest pain provided by both ibuprofen dose levels was superior to that achieved with acetaminophen.

A second study,²⁷ involving 178 patients assigned to receive either 750 mg/d of the dual COX-1 and COX-2 inhibitor naproxen, or 2600 mg/d of acetaminophen, demonstrated greater efficacy of the NSAID for only 1 end point: rest pain. No statistically significant difference was noted in other end points, including pain on motion, 50-foot walk time, and physician assessment of disease activity. Failure to demonstrate superiority of NSAIDs over acetaminophen in these previous trials, in part, may have been because earlier studies examined a more limited profile of less responsive end points, and in part because of inadequate power to detect differences in these studies.

Recently, Pincus et al²⁸ reported the results of a 6-week OA trial, similar in size to our study, comparing acetaminophen with diclofenac plus misoprostol that are consistent with the results we report herein. For WOMAC end points, diclofenac provided statistically significantly greater responses compared with acetaminophen. Previous studies comparing rofecoxib and celecoxib to dual-inhibiting NSAIDs have shown comparable efficacy of the COX-2-specific agents compared with either high-dose ibuprofen¹⁶ or naproxen.^17- 18

Ours is the first study, to our knowledge, that assesses the effects of COX-2 inhibitors and maximal daily dosages of acetaminophen in the treatment of symptomatic OA across this broad range of patient-reported end points. Across multiple clinical end points, rofecoxib 25 mg/d provided greater benefit in adult patients with OA of the knee than did maximal doses of acetaminophen. Differences in response to these 2 treatments were observed within 6 days of initiating therapy and were maintained over the 6-week treatment period. Compared with acetaminophen, 12.5 mg/d of rofecoxib also resulted in significantly greater proportion of positive PGART responses over 6 weeks, and both rofecoxib 12.5 mg/d and celecoxib 200 mg/d provided more rapid relief of pain walking in the first 6 days of therapy than did acetaminophen. By the end of 6 weeks of therapy, approximately 39% of patients treated with acetaminophen assessed their response to be good or excellent, and, therefore, had also responded to therapy.

Since OA is a symptomatic disease, with poor correlation of objective measures of disease with symptoms,²⁵ most clinical studies focus on patient-reported symptomatic outcome measures (ie, WOMAC and PGART), which are generally assessed over several weeks. Such a study design was previously recommended by an expert panel²⁵ and is similar to the one we used. Since there is no current consensus on the magnitude of effects that are clinically important, this study focused on statistically significant differences in treatment effects.

In this study, the renal and vascular safety profile of rofecoxib and celecoxib was similar to that of acetaminophen. Incidences of edema-related events were low in each treatment group, with lower extremity, ankle, and pedal edema being the most common types reported. Hypertension and cardiovascular events rarely occurred in this trial. In a recent report based on selected gastrointestinal tract safety and efficacy trials on COX-2 inhibitors, the annualized rates of myocardial infarction for rofecoxib were 0.74%; for celecoxib, 0.80%; and for placebo, 0.52%.²⁹ The placebo estimate was obtained from a meta-analysis of 4 randomized trials of aspirin for cardioprotection.²⁹ A more recent meta-analysis that included placebo-controlled trials using the cardiovascular event end point, defined by the Anti-Platelet Trialists' Collaboration,³⁰ showed no significant difference in rates of these events between rofecoxib and placebo. In our study, no significant differences were found comparing COX-2 inhibitors with acetaminophen although the trial was of relatively short duration.

Our study had several potential limitations. We prespecified 5 key efficacy end points instead of a single primary end point, raising potential concerns for interpretation of statistical differences due to multiplicity of analyses. However, the findings were consistent across all end points, which in aggregate indicates a rank order of efficacy as follows: rofecoxib 25 mg/d > rofecoxib 12.5 mg/d = celecoxib 200 mg/d > acetaminophen. The consistency of the data across all 5 key end points provides support for selective COX-2 inhibitors, expecially rofecoxib 25 mg/d, having greater efficacy than acetaminophen 4000 mg/d in the treatment of symptomatic OA of the knee. It is possible that a higher dose of celecoxib would have resulted in greater efficacy although previous studies have demonstrated that 200 mg/d is the dose level that provides maximal efficacy in OA,^17- 18 perhaps because of lack of proportional increase in plasma levels beyond this dose.²⁹ Further statistical power also may have been provided by larger group sizes. Additional studies to evaluate the relative efficacy of these agents in OA may be helpful.

The efficacy and safety profiles of selective COX-2 inhibitors and acetaminophen need to be considered especially when treating patients with OA who tend to be older, regularly require analgesics, and are especially at risk for NSAID-related adverse events.²⁸ Selective COX-2 inhibitors are more expensive than acetaminophen and some NSAIDs; thus, individualized decisions need to be made regarding optimal medical management and modification of treatment based on patient response, published efficacy and safety data, and overall cost considerations.^{5- 6,14,17,28}