Description of studies

From the initial search strategy we identified 39 potentially eligible studies, and after further analyses 7 studies met the inclusion criteria of the review. After updating the search strategy, one additional study was identified, so eight studies met the above criteria and were included in the current systematic review. The remaining studies were excluded.

Studies were described according to the antibiotic used, timing and route of administration, number of doses given, technique of hernia repair, criteria for the diagnosis of infection and follow‐up time.

In one study on the use of different surgical techniques including hernia repair, Evans 1973 used intra‐muscular (IM) cephaloridine 1 g in the anaesthesia induction and two further intra‐muscular doses in the postoperative period. Definition of infection was: purulent exudates at the wound site or need for pus drainage from the wound. Follow‐up was 4 weeks.

Andersen 1980 used a single dose of 1 g topical subfascial ampicillin before closure of non‐mesh hernia repair and colecystectomy. The criteria for wound infection considered in this study was the collection of pus in the wound site requiring revision. Patients in the study were followed for one year.

In the study on antimicrobial prophylaxis for mastectomy and non‐mesh hernia repair, Platt 1990 used a single dose of 1g cefonicid, administered intra‐venouosly 90 minutes before surgery. Wound infection was defined as the presence of erythema and drainage, purulent exudates, or non‐closing open wound. Probable infection was considered when erythema extending at least 2 cm to any direction was present, or when wound infection was diagnosed even if established infection criteria were not met.

Lazorthes 1992 combined cefamandole 750 mg to local anaesthesia for patients undergoing non‐mesh hernia repair. Wound abscess was defined as all wounds with discharge in which pathogens grew regardless of whether or not the discharge was purulent or serohematic. Discharge of pus, even when germs were not found, was also considered a wound abscess. Patients were assessed one month after the intervention.

Taylor 1997 administered amoxicillin‐clavulanic acid 1.2 g intra‐venouosly before the incision in patients undergoing non‐mesh hernia repair. Infection was defined as purulent wound discharge or spreading erythema indicative of cellulites; wound breakdown; or dehiscence with clinical evidence of infection. Patients were assessed 4 and 6 weeks after surgery.

Morales 2000 is a multicenter study that assessed the efficacy of a single dose of cefazolin 2 g administered intra‐venouosly during anaesthetic induction for patients undergoing mesh hernia repair. The authors defined the following criteria for wound infection: a) cutaneous erythema greater than 2 cm on both sides of the incision, b) purulent exudates through the wound, c) organism isolated from culture of non‐ purulent exudates, and d) an open wound that was not closed afterwards. The wound was assessed 30 days after the surgery.

Yerdel 2001 used ampicillin‐sulbactam 1.5 g IV before the incision in patients undergoing mesh hernia repair. The criteria of the Center for Disease Control (1992) for wound infection were applied. One year follow‐up period was considered.

Oteiza 2004 used amoxycilin‐clavulanic acid 2 g IV 15‐30 minutes before the incision in patients undergoing elective mesh hernia repair, Lichtenstein or plug‐mesh technique. Wound infection was defined as the purulent exudate or non purulent exudate with positive culture or the surgeon declares that incisional infection is present. A month follow‐up period was considered.

The remaining studies were excluded for several reasons:

Studies focused on several clean surgical techniques, with no consideration of hernia pathology (Bozzetti 1975; Gaskill 1989; Lord 1977; Lord 1983; Luke 1991; Menzies 1989; Wersink 1991);
Studies focused on clean surgical techniques including hernia pathology, but data for this subgroup of patients could not be collected (Houck 1989; Lewis 1995; Nundy 1983);
Final results are not yet available for one study (Karran 1992);
Studies reported results of patients series having no antibiotic prophylaxis (Hedawoo 1995; Wantz 1996), otherwise, they were non‐controlled studies of antibiotic prophylaxis (Angio 2001; Dazzi 1994; Gervino 2000; Massaioli 1995; Spallitta 1999; Van‐Damme 1981; Sultan 1989);
Both study arms received antibiotic prophylaxis (Musella 2001; Shwed 1991; Reggiori 1996);
Studies used historical controls (Abo‐Rahmy 1998), they were comparative retrospective or non‐randomized studies, or misbalances in patients and techniques used in both treatment groups were present (Barreca 2000; Escartín 1999; Gilbert 1993; Hair 2000; Platt 1992; Ryan 1967; Vara 1993);
Studies did not use antibiotic but local antiseptic treatment (Gilmore 1977).

Risk of bias in included studies

A data form including ten methodological parameters for the initial classification of the studies was designed (Table 1).

Two studies did not provide information on the random allocation concealment procedures (Andersen 1980; Lazorthes 1992). Consequently, they are considered as "unclear". The concealment procedure of Evans 1973 (coin tossing) was inappropriate. The remaining studies used appropriate concealment methods (sealed envelopes: Morales 2000, computer programs: Platt 1990; Taylor 1997; Yerdel 2001, Oteiza 2004).

Morales 2000; Platt 1990; Taylor 1997; Yerdel 2001 were blinded studies; Andersen 1980 was described by the author as a triple‐blinded trial despite the fact that the control group received no intervention, and Evans 1973,Lazorthes 1992 and Oteiza 2004 were not blinded studies.

The patient comparison groups were homogeneous in the majority of included studies, with respect to epidemiological characteristics, techniques used in the
hernia repair, and associated comorbilidities. However, the study Andersen 1980 differed with regard to one factor, and the study Evans 1973 differed by more than one factor.
Apart from Evans 1973 and Lazorthes 1992, the statistical methods were clearly reported in the studies.

All studies except Lazorthes 1992, clearly described a set of wound infection criteria.

All studies used a penicillin derivative antibiotic. The route of administration was intravenous in five studies, subcutaneous/subfascial in two studies (Andersen 1980; Lazorthes 1992) and intramuscularly in one study (Evans 1973). Lazorthes 1992; Morales 2000; Platt 1990; Taylor 1997; Yerdel 2001 and Oteiza 2004 used a single preoperative dose. Andersen 1980 administered subfascial antibiotic before closing the aponeurosis. Evans 1973 used one preoperative and two postoperative intramuscularly doses.

Andersen 1980; Evans 1973; Lazorthes 1992 and Oteiza 2004 included patients with no drug intervention as the control group. The remaining studies used placebo.

Effects of interventions

The total number of patients included in the meta‐analysis was 2907 (prophylaxis group: 1421, control group: 1486). There was no obvious evidence of significant heterogeneity when all the eight studies were pooled (p=0.18). The overall infection rates were 2.88% in the prophylaxis group, and 4.30% in the control group (OR 0.65, 95% CI 0.35 ‐ 1.21). Despite the odds ratio not being significant, the risk differences for this outcome were in fact of borderline significance. This result should be considered cautiously, as a sign of what the expected benefit could be, in case of reaching significance.

Analysis of the group of patients with herniorrhaphy showed no evidence of heterogeneity (p=0.42). The number of patients treated with prophylaxis was 924 and the infection rate for this group was 3.78%. The number of patients in the control group was 943 and the infection rate was 4.87% (OR 0.84, 95%CI 0.53 ‐ 1.34). As in the previous paragraph, the risk differences showed a borderline significative effect. This result should be considered cautiously, as a sign of what the expected benefit could be, in case of reaching significance.

Analysis of the group of patients with hernioplasty showed significant heterogeneity (p=0.07). The number of patients treated with prophylaxis was 497 and the infection rate for this group was 1.2%. The number of patients in the control group was 543 and the infection rate was 3.3% (OR 0.48; CI 95%: 0.07 ‐ 3.37). The risk differences were not significant for this outcome, and so the intervention studied should be viewed as able to produce a benefit or a harm; the confidence interval reflects this dual character by computing to which extent the intervention could be beneficial or harmful (Altman 1998).

Factors like number of antibiotics administered and methodological quality are not related to differences in the treatment effect between the included studies, as shown by the non significant results of the metaregression model (details can be sought from the authors).

Sensitivity analyses
Analysis of the Peto odds ratio under the fixed effect model provided statistically significant estimates for the global effect size (Peto OR 0.65, 95% CI 0.43‐0.96) and for the hernioplasty subgroup (Peto OR 0.37, 95% CI 0.15‐0.91). On the other hand, in the herniorrhaphy subgroup no significant results were obtained (Peto OR 0.76, 95%CI 0.48‐1.19). An explanation for these results, in contrast with the main analysis, could be the low rates of infection observed in all the studies but two (Taylor and Yerdel show an incidence around 9% while the other trials show incidences below 5%).

The subgroup analysis performed with the highest quality trials (Platt 1990; Taylor 1997; Yerdel 2001) showed no qualitative differences with respect to the main analyses.

None of the included studies is highly influential on the analysis, since the exclusion of any of them did not change the global qualitative conclusion. Global heterogeneity appeared when any of the following studies was excluded: Andersen 1980; Evans 1973; Platt 1990; Morales 2000.