Types of studies

We included randomised controlled trials (RCTs), irrespective of blinding, language, publication status, date of publication, study setting, sample size, or whether the incidence of MRSA infection was the primary outcome of the trial. We planned to include cluster randomised clinical trials if the effect estimate was available after adjusting for clustering effect. No other study designs (i.e. quasi‐randomised studies and non‐randomised studies) were included.

Types of participants

People undergoing surgery, irrespective of age, type of surgery, whether surgery was elective or emergency, and whether MRSA colonisation was identified by routine screening (in general, we expected that if MRSA colonisation was identified, it would have been eradicated before surgery). We excluded studies recruiting people with established MRSA SSIs, as these are covered in another review (Gurusamy).

Types of interventions

1. Comparison of antibiotic prophylaxis (irrespective of the antibiotic) compared with placebo (or no treatment).

2. Comparison of different antibiotic prophylaxis (and regimens). This includes different doses, routes, or timings of administration.

We included studies evaluating a combination of antibiotics in terms of the combined regimen rather than as single antibiotics.

Types of outcome measures

Electronic searches

In March 2013 we searched the following electronic databases to identify reports of relevant RCTs: 

• The Cochrane Wounds Group Specialised Register (searched 28 February 2013);

• The Cochrane Central Register of Controlled Trials (The Cochrane Library CENTRAL) (2013, Issue 1);

• Database of Abstracts of Reviews of Effects (DARE) ) (The Cochrane Library 2013, Issue 1);

• NHS Economic Evaluation Database ) (The Cochrane Library 2013, Issue 1);

• Health Technology Assessment (HTA) Database ) (The Cochrane Library 2013, Issue 1);

• Ovid MEDLINE (1948 to February Week 3 2013);

• Ovid MEDLINE (In‐Process & Other Non‐Indexed Citations, 28 February 2013);

• Ovid EMBASE (1974 to 2013 Week 08);

• EBSCO CINAHL (1982 to 21 February 2013).

We used the following search strategy in The Cochrane Central Register of Controlled Trials (CENTRAL):

#1 MeSH descriptor Methicillin Resistance explode all trees
#2 MeSH descriptor Staphylococcal Infections explode all trees
#3 MeSH descriptor Staphylococcus aureus explode all trees
#4 (#2 OR #3)
#5 (#1 AND #4)
#6 MeSH descriptor Methicillin‐Resistant Staphylococcus aureus explode all trees
#7 (methicillin NEXT resistan*) or (meticillin NEXT resistan*) or MRSA
#8 (#5 OR #6 OR #7)
#9 MeSH descriptor Wound Infection explode all trees
#10 MeSH descriptor Sepsis explode all trees
#11 MeSH descriptor Soft Tissue Infections explode all trees
#12 MeSH descriptor Surgical Wound Dehiscence explode all trees
#13 surg* NEAR/5 infect*
#14 surg* NEAR/5 wound*
#15 surg* NEAR/5 site*
#16 surg* NEAR/5 incision*
#17 surg* NEAR/5 dehisc*
#18 wound* NEAR/5 dehisc*
#19 (deep NEXT infection*) or "deep sepsis" or (infected NEXT
collection*)
#20 (#9 OR #10 OR #11 OR #12 OR #13 OR #14 OR #15 OR #16 OR #17 OR
#18 OR #19)
#21 MeSH descriptor Anti‐Bacterial Agents explode all trees
#22 MeSH descriptor Cephalosporins explode all trees
#23 MeSH descriptor Tetracycline explode all trees
#24 MeSH descriptor Penicillins explode all trees
#25 antibiotic* or penicillin* or beta‐lactam* or cephalosporin* or clindamycin or trimethoprim* or tetracycline* or doxycycline or minocycline or linezolid or vancomycin or daptomycin or telavancin or rifampicin or gentamycin or gentamicin or fluoroquinolone
#26 (#21 OR #22 OR #23 OR #24 OR #25)
#27 (#8 AND #20 AND #26)

The search strategies for Ovid MEDLINE, Ovid EMBASE and EBSCO CINAHL can be found in Appendix 1. We combined the Ovid MEDLINE search with the Cochrane Highly Sensitive Search Strategy for identifying randomised trials in MEDLINE: sensitivity‐ and precision‐maximising version (2008 revision) (Lefebvre 2011). We combined the EMBASE search with the Ovid EMBASE filter developed by the UK Cochrane Centre (Lefebvre 2011). We combined the CINAHL searches with the trial filters developed by the Scottish Intercollegiate Guidelines Network (SIGN 2013). We did not restrict studies with respect to language, date of publication or study setting.

We searched the metaRegister of Controlled Trials (mRCT) (http://www.controlled‐trials.com/mrct/), which includes the ISRCTN Register and the NIH ClinicalTrials.gov Register among others. We also searched the World Health Organization's International Clinical Trials Registry Platform (ICTRP) (http://apps.who.int/trialsearch/). The ICRTP portal includes national trial registry databases from a number of countries (last searched 11 December 2012).

Searching other resources

We searched the references of the included trials to identify further relevant trials. We also contacted experts in MRSA infection to identify further trials.

Selection of studies

Two review authors (KG and RK) identified the trials for inclusion independently. We have listed the excluded studies with reasons for their exclusion. Any differences were resolved through discussion.

Data extraction and management

Both review authors extracted the following data independently:

1. Year and language of publication of trial report.

2. Country.

3. Year in which trial was conducted.

4. Inclusion and exclusion criteria.

5. Sample size.

6. Type of surgery.

7. Details of antibiotic treatment including dose, route, frequency, and duration.

8. Outcomes (described above).

9. Risk of bias (described below).

10. Source of funding.

When multiple reports had been published for a trial, we obtained information from all the reports. We contacted the authors of the individual trials to seek out any unclear or missing information. If there was any doubt about whether the trials shared the same participants, completely or partially (by identifying common authors and centres), we contacted the study authors to clarify whether the trial report had been duplicated. We resolved any differences in opinion through discussion.

Assessment of risk of bias in included studies

We followed the instructions in the Cochrane Handbook for Systematic Reviews of Intervention to assess risk of bias (Higgins 2011b). According to empirical evidence (Kjaergard 2001; Moher 1998; Schulz 1995; Wood 2008), the risk of bias of the trials was assessed according to the following bias risk domains:

Measures of treatment effect

For dichotomous variables, we calculated the risk ratio (RR) with 95% confidence interval (CI). We used Peto's odds ratios (Peto OR) for outcomes with a proportion of less than 1% events. For continuous variables, we planned to calculate the mean difference (MD) with 95% CI for outcomes such as hospital stay and standardised mean difference (SMD) with 95% CI for quality of life (where different scales might be used). For time‐to‐event outcomes such as survival at maximal follow‐up, we planned to calculate the hazard ratio (HR) with 95% CI.

Unit of analysis issues

The unit of analysis was individual people undergoing surgical procedures. We did anticipate the need for people to have a second operation during the same admission within the trials. However, we decided that if we found any person having a second operation, provided that s/he did not have MRSA infection before the second surgery, we would consider each surgery to be a separate unit of analysis.

Dealing with missing data

We performed an intention‐to‐treat analysis whenever possible (Newell 1992). We imputed missing data for binary outcomes using various scenarios such as best‐best scenario, worst‐worst scenario, best‐worst scenario, and worst‐best scenario (Gurusamy 2009). In the best‐best scenario, the people with the missing outcomes were considered not to have developed a complication. In the worst‐worst scenario, the people with the missing outcomes were considered to have developed a complication. In the best‐worst scenario, the people with the missing outcomes were considered not to have developed a complication in the intervention group and to have developed a complication in the control group. In the worst‐best scenario, the people with the missing outcomes were considered to have developed a complication in the intervention group and not to have developed a complication in the control group.

For continuous outcomes, we planned to use available‐case analysis. We planned to impute the standard deviation from P values according to the instructions in the Cochrane Handbook for Systematic Reviews of Intervention (Higgins 2011c), and use the median for the meta‐analysis when the mean was not available. If it was not possible to calculate the standard deviation from the P value or the CIs, we planned to impute the standard deviation as the highest standard deviation in the other trials included under that outcome, fully recognising that this form of imputation would decrease the weight of the study for calculation of MD and bias the effect estimate to no effect in case of SMD (Higgins 2011c).

For time‐to‐event outcomes, where HR and 95% CIs were not reported, we planned to obtain the logarithm of hazard ratios (ln(HR)) and the standard error (SE) of ln(HR) according to the methods described by Parmar 1998.

Assessment of heterogeneity

We explored heterogeneity by means of the Chi² test with significance set at P value 0.10, and measured the degree of heterogeneity by means of the I² test (Higgins 2002). We also used overlapping of CIs on the forest plot to determine heterogeneity.

Assessment of reporting biases

We planned to use visual asymmetry on a funnel plot to explore reporting bias in the presence of at least 10 trials (Egger 1997; Macaskill 2001). We planned to perform the linear regression approach described by Egger 1997 to determine the funnel plot asymmetry.

Data synthesis

For the comparison of antibiotic versus placebo (or no intervention), we planned to perform the meta‐analysis only if there was sufficient clinical homogeneity in terms of people included in the trials. This was based on our clinical judgement. For the comparison of different antibiotics, we planned to perform the meta‐analysis only if there was sufficient clinical homogeneity in terms of people included in the trials and in terms of antibiotics used (i.e. similar class of antibiotics). We performed the meta‐analyses using the software package RevMan 5 (RevMan 2011), following the recommendations of The Cochrane Collaboration (Higgins 2011a). We used both a random‐effects model (DerSimonian 1986), and a fixed‐effect model (DeMets 1987), for the meta‐analyses. In case of discrepancy between the two models identified from the pooled estimates and their CIs, we have reported both results; otherwise we have reported the results of the fixed‐effect model. Calculations for dichotomous outcomes to produce risk ratio (RR) or Peto OR do not include trials in which no events occurred in either group in the meta‐analysis, whereas risk difference (RD) calculations do. We planned to report the RD when the results using this association measure were different from RR or Peto OR. However, RR or Peto OR were the measures that we used to arrive at conclusions, since they perform better than RD when there are differences in the control event rate (proportion of people who develop the event in the control). We planned to use the generic inverse variance method to combine the HRs for time‐to‐event outcomes.

Subgroup analysis and investigation of heterogeneity

We planned to perform the following subgroup analyses.

1. Different antibiotics (or class of antibiotics).

2. Different doses and durations of antibiotics.

3. Different types of surgery including different types of wounds (clean, clean‐contaminated, contaminated, and dirty or infected) (the risk of SSI varies according to wound types) (Garner 1986).

4. Different patient characteristics (presence of systemic illness such as diabetes or other immunocompromised individuals).

5. People routinely screened and treated versus those who were not routinely screened.

We planned to use a P value of less than 0.05 for the Chi² test to identify the differences between subgroups and to investigate whether heterogeneity in effect estimates were attributable to differences in the above characteristics.

Sensitivity analysis

We performed a sensitivity analysis by imputing data for binary outcomes using various scenarios such as best‐best scenario, worst‐worst scenario, best‐worst scenario, and worst‐best scenario (Gurusamy 2009). We also planned to perform a sensitivity analysis by testing the effect of removing trials at unclear or high risk of bias and excluding the trials in which the mean and the standard deviation were imputed.