We included 36 trials that provided data for 17,206 adults: 9402 participants received celecoxib 200 mg/day, and 7804 were assigned to receive either tNSAIDs (N = 1869) or placebo (N = 5935). Celecoxib was compared with placebo (32 trials), naproxen (6 trials) and diclofenac (3 trials). Studies were published between 1999 and 2014. Studies included participants with knee, hip or both knee and hip OA; mean OA duration was 7.9 years. Most studies included predominantly white participants whose mean age was 62 (± 10) years; most participants were women. There were no concerns about risk of bias for performance and detection bias, but selection bias was poorly reported in most trials. Most trials had high attrition bias, and there was evidence of selective reporting in a third of the studies.
Celecoxib versus placebo
Compared with placebo celecoxib slightly reduced pain on a 500-point Western Ontario and McMaster Universities Arthritis Index (WOMAC) pain scale, accounting for 3% absolute improvement (95% CI 2% to 5% improvement) or 12% relative improvement (95% CI 7% to 18% improvement) (4 studies, 1622 participants). This improvement may not be clinically significant (high quality evidence).
Compared with placebo celecoxib slightly improved physical function on a 1700-point WOMAC scale, accounting for 4% absolute improvement (95% CI 2% to 6% improvement), 12% relative improvement (95% CI 5% to 19% improvement) (4 studies, 1622 participants). This improvement may not be clinically significant (high quality evidence).
There was no evidence of an important difference for withdrawals due to adverse events (Peto OR 0.99, 95% CI 0.85 to 1.15) (moderate quality evidence due to study limitations).
Results were inconclusive for numbers of participants experiencing any serious AEs (SAEs) (Peto OR 0.95, 95% CI 0.66 to 1.36), gastro-intestinal events (Peto OR 1.91, 95% CI 0.24 to 14.90) and cardiovascular events (Peto OR 3.40, 95% CI 0.73 to 15.88) (very low quality evidence due to serious imprecision and study limitations). However, regulatory agencies have warned of increased cardiovascular events for celecoxib.
Celecoxib versus tNSAIDs
There were inconclusive results regarding the effect on pain between celecoxib and tNSAIDs on a 100-point visual analogue scale (VAS), showing 5% absolute improvement (95% CI 11% improvement to 2% worse), 11% relative improvement (95% CI 26% improvement to 4% worse) (2 studies, 1180 participants, moderate quality evidence due to publication bias).
Compared to a tNSAID celecoxib slightly improved physical function on a 100-point WOMAC scale, showing 6% absolute improvement (95% CI 6% to 11% improvement) and 16% relative improvement (95% CI 2% to 30% improvement). This improvement may not be clinically significant (low quality evidence due to missing data and few participants) (1 study, 264 participants).
Based on low or very low quality evidence (downgraded due to missing data, high risk of bias, few events and wide confidence intervals) results were inconclusive for withdrawals due to AEs (Peto OR 0.97, 95% CI 0.74 to 1.27), number of participants experiencing SAEs (Peto OR 0.92, 95% CI 0.66 to 1.28), gastro-intestinal events (Peto OR 0.61, 0.15 to 2.43) and cardiovascular events (Peto OR 0.47, 95% CI 0.17 to 1.25).
In comparisons of celecoxib and placebo there were no differences in pooled analyses between our main analysis with low risk of bias and all eligible studies. In comparisons of celecoxib and tNSAIDs, only one outcome showed a difference between studies at low risk of bias and all eligible studies: physical function (6% absolute improvement in low risk of bias, no difference in all eligible studies).
No studies included in the main comparisons measured quality of life. Of 36 studies, 34 reported funding by drug manufacturers and in 34 studies one or more study authors were employees of the sponsor.