Types of studies

We included studies if they were full publications of double‐blind trials of single‐dose oral celecoxib against placebo for the treatment of moderate to severe postoperative pain in adults, with at least 10 participants randomly allocated to each treatment group. We included multiple‐dose studies if appropriate data from the first dose were available, and included cross‐over studies provided that data from the first arm were presented separately.

We excluded studies if they were:

• posters or abstracts not followed up by full publication;

• reports of trials concerned with pain other than postoperative pain (including experimental pain);

• trials using healthy volunteers;

• trials where pain relief was assessed by clinicians, nurses or carers (i.e. not patient‐reported);

• trials of less than four hours' duration or which failed to present data over four to six hours postdose.

Types of participants

We included studies of adult participants (15 years old or above) with established moderate to severe postoperative pain. For studies using a visual analogue scale (VAS), pain of at least moderate intensity was assumed when the VAS score was greater than 30 mm (Collins 1997). We included trials of patients with postpartum pain provided the pain investigated resulted from episiotomy or Caesarean section (with or without uterine cramp). We excluded trials investigating pain due to uterine cramps alone.

Types of interventions

Orally administered celecoxib or matched placebo for relief of postoperative pain.

Types of outcome measures

Data collected included the following if available:

• patient characteristics;

• pain model;

• patient‐reported pain at baseline (physician, nurse, or carer reported pain would not be included in the analysis);

• patient‐reported pain relief expressed hourly over four to six hours using validated scales, or reported total pain relief (TOTPAR) at four to six hours;

• patient‐reported pain intensity expressed hourly over four to six hours using validated pain scales, or reported summed pain intensity difference (SPID) at four to six hours;

• patient‐reported global evaluation of treatment using a validated scale;

• number of participants using rescue medication, and the time of assessment;

• time to use of rescue medication;

• withdrawals ‐ all‐cause, adverse event;

• adverse events ‐ participants experiencing one or more adverse event and any serious adverse event, and the time of assessment.

Electronic searches

We searched the following electronic databases:

• the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2013, Issue 5 of 12);

• MEDLINE via Ovid (1966 to 31 May 2013);

• EMBASE via Ovid (1980 to 31 May 2013);

• Oxford Pain Database (Jadad 1996a);

• ClinicalTrials.gov (on 31 May 2013) for update only.

See Appendix 1 for the MEDLINE search strategy, Appendix 2 for the EMBASE search strategy, and Appendix 3 for the CENTRAL search strategy.

Searches for the original review were up to July 2008, for the first update to January 2012, and the second update to May 2013.

Searching other resources

We manually searched reference lists of retrieved studies. We did not search abstracts, conference proceedings, and other grey literature. We did not contact manufacturers. For this update we searched ClinicalTrials.gov for any unpublished and ongoing studies, and attempted to contact the study sponsors for further information.

Selection of studies

Two review authors independently assessed and agreed the search results for studies that might be included in the review. We resolved disagreements by consensus or referral to a third review author.

Data extraction and management

Two review authors independently extracted data and recorded it on a standard data extraction form. One author entered data suitable for pooling into RevMan 5.2 (RevMan 2012).

Assessment of risk of bias in included studies

Two review authors independently assessed each study using a three‐item, five‐point scale (Jadad 1996b) and agreed a consensus score.

We also completed a Risk of bias in included studies table, using methods adapted from those described by the Cochrane Pregnancy and Childbirth Group. Two authors independently assessed risk of bias for each study using the criteria outlined in the Cochrane Handbook for Systematic Reviews of Interventions (Higgins 2011), with any disagreements resolved by discussion.

The following were assessed for each study.

• Random sequence generation (checking for possible selection bias). The method used to generate the allocation sequence was assessed as: low risk of bias (any truly random process, e.g. random number table; computer random number generator);  high risk of bias (any non‐random process, e.g. odd or even date of birth; hospital or clinic record number) ‐ these studies would be excluded; unclear risk of bias.

• Allocation concealment (checking for possible selection bias). The method used to conceal allocation to interventions prior to assignment assessed whether intervention allocation could have been foreseen in advance of or during recruitment, or changed after assignment. The methods were assessed as: low risk of bias (e.g. telephone or central randomisation; consecutively numbered sealed opaque envelopes); high risk of bias (open random allocation; unsealed or non‐opaque envelopes, alternation; date of birth) ‐ these studies would be excluded; unclear risk of bias.

• Blinding of outcome assessment (checking for possible detection bias). The methods used to blind study participants and outcome assessors from knowledge of which intervention a participant received were assessed. Studies were considered to be at low risk of bias if they stated that they were blinded and described the method used to achieve blinding (e.g. identical tablets; matched in appearance and smell), or at unknown risk if they stated that they were blinded but did not provide an adequate description of how it was achieved. Single‐blind and open studies would be excluded.

• Size (checking for possible biases confounded by small size). Small studies have been shown to overestimate treatment effects, probably due to methodological weaknesses (Nuesch 2010). Studies were considered to be at low risk of bias if they had ≥ 200 participants, at unknown risk of they had 50 to 200 participants, and at high risk if they had < 50 participants.

Measures of treatment effect

We used relative risk (or 'risk ratio', RR) to establish statistical difference. We used numbers needed to treat (NNT) and pooled percentages as absolute measures of benefit or harm.

We use the following terms to describe adverse outcomes in terms of harm or prevention of harm.

• When significantly fewer adverse outcomes occur with celecoxib than with control (placebo or active) we use the term the number needed to treat to prevent one event (NNTp).

• When significantly more adverse outcomes occur with celecoxib compared with control (placebo or active) we use the term the number needed to harm or cause one event (NNH).

Unit of analysis issues

We accepted only randomisation to the individual patient.

Dealing with missing data

The only likely issue with missing data in these studies was from imputation using last observation carried forward when a patient requests rescue medication. We have previously shown that this does not affect results for up to six hours after taking study medication (Barden 2004).

Assessment of heterogeneity

We examined heterogeneity visually using L'Abbé plots (L'Abbé 1987).

Data synthesis

We followed the QUOROM guidelines (Moher 1999). For efficacy analyses we used the number of participants in each treatment group who were randomised, received medication, and provided at least one post‐baseline assessment. For safety analyses we used the number of participants randomised to each treatment group who took the study medication. We planned analyses for different doses.

For each study we converted the mean TOTPAR, SPID, VAS TOTPAR, or VAS SPID (Appendix 4) values for active treatment and placebo to %maxTOTPAR or %maxSPID by division into the calculated maximum value (Cooper 1991), and calculated the proportion of participants in each treatment group who achieved at least 50%maxTOTPAR using verified equations (Moore 1996; Moore 1997a; Moore 1997b). We then converted these proportions into the number of participants achieving at least 50%maxTOTPAR by multiplying by the total number of participants in the treatment group. We used this information on the number of participants with at least 50%maxTOTPAR for active treatment and placebo to calculate relative benefit or relative risk and number needed to treat to benefit (NNT).

We accepted the following pain measures for the calculation of TOTPAR or SPID:

• five‐point categorical pain relief (PR) scales with comparable wording to 'none, slight, moderate, good or complete';

• four‐point categorical pain intensity (PI) scales with comparable wording to 'none, mild, moderate, severe';

• VAS for pain relief;

• VAS for pain intensity.

If none of these measures were available, we used the number of participants reporting 'very good or excellent' on a five‐point categorical global scale with the wording 'poor, fair, good, very good, excellent' for the number of participants achieving at least 50% pain relief (Collins 2001).

For each treatment group we extracted the number of participants reporting treatment‐emergent adverse effects, and calculated relative benefit and risk estimates with 95% confidence intervals (CI) using a fixed‐effect model (Morris 1995). We calculated NNT and number needed to treat to harm (NNH) with 95% CIs using the pooled number of events and the method devised by Cook and Sackett (Cook 1995). We assumed a statistically significant difference from control when the 95% CI of the relative risk or relative benefit did not include one.

Subgroup analysis and investigation of heterogeneity

We planned subgroup analyses to determine the effect of dose. A minimum of two trials and 200 participants had to be available in any subgroup or sensitivity analysis (Moore 1998), which was restricted to the primary outcome (50% pain relief over four to six hours). We determined significant differences between NNT, NNTp, or NNH for different groups in subgroup and sensitivity analyses using the z test (Tramèr 1997).

Sensitivity analysis

We planned sensitivity analyses for pain model (dental versus other postoperative pain), trial size (39 or fewer versus 40 or more per treatment arm), and quality score (two versus three or more).