Wound Edge Protectors in Open Abdominal Surgery to Reduce Surgical Site Infections: A Systematic Review and Meta-Analysis

Importance Surgical site infections remain one of the most frequent complications following abdominal surgery and cause substantial costs, morbidity and mortality. Objective To assess the effectiveness of wound edge protectors in open abdominal surgery in reducing surgical site infections. Evidence Review A systematic literature search was conducted according to a prespecified review protocol in a variety of data-bases combined with hand-searches for randomized controlled trials on wound edge protectors in patients undergoing laparotomy. A qualitative and quantitative analysis of included trials was conducted. Findings We identified 16 randomized controlled trials including 3695 patients investigating wound edge protectors published between 1972 and 2014. Critical appraisal uncovered a number of methodological flaws, predominantly in the older trials. Wound edge protectors significantly reduced the rate of surgical site infections (risk ratio 0.65; 95%CI, 0.51–0.83; p = 0.0007; I2 = 52%). The results were robust in a number of sensitivity analyses. A similar effect size was found in the subgroup of patients undergoing colorectal surgery (risk ratio 0.65; 95%CI, 0.44–0.97; p = 0.04; I2 = 56%). Of the two common types of wound protectors double ring devices were found to exhibit a greater protective effect (risk ratio 0.29; 95%CI, 0.15–0.55) than single-ring devices (risk ratio 0.71; 95%CI, 0.54–0.92), but this might largely be due to the lower quality of available data for double-ring devices. Exploratory subgroup analyses for the degree of contamination showed a larger protective effect in contaminated cases (0.44; 95%CI, 0.28–0.67; p = 0.0002, I2 = 23%) than in clean-contaminated surgeries (0.72, 95%CI, 0.57–0.91; p = 0.005; I2 = 46%) and a strong effect on the reduction of superficial surgical site infections (risk ratio 0.45; 95%CI, 0.24–0.82; p = 0.001; I2 = 72%). Conclusions and Relevance Wound edge protectors significantly reduce the rate of surgical site infections in open abdominal surgery. Further trials are needed to explore their effectiveness in different risk constellations.


Objective
To assess the effectiveness of wound edge protectors in open abdominal surgery in reducing surgical site infections.

Evidence Review
A systematic literature search was conducted according to a prespecified review protocol in a variety of data-bases combined with hand-searches for randomized controlled trials on wound edge protectors in patients undergoing laparotomy. A qualitative and quantitative analysis of included trials was conducted.

Findings
We identified 16 randomized controlled trials including 3695 patients investigating wound edge protectors published between 1972 and 2014. Critical appraisal uncovered a number of methodological flaws, predominantly in the older trials. Wound edge protectors significantly reduced the rate of surgical site infections (risk ratio 0.65; 95%CI, 0.51-0.83; p = 0.0007; I 2 = 52%). The results were robust in a number of sensitivity analyses. A similar effect size was found in the subgroup of patients undergoing colorectal surgery (risk ratio 0.65; 95%CI, 0.44-0.97; p = 0.04; I 2 = 56%). Of the two common types of wound protectors double ring devices were found to exhibit a greater protective effect (risk ratio 0.29; 95%CI, 0.15-0.55) than single-ring devices (risk ratio 0.71; 95%CI, 0.54-0.92), but this might largely be due to the lower quality of available data for double-ring devices. Exploratory subgroup analyses for the degree of contamination showed a larger protective effect in contaminated cases (0.44; 95%CI, 0.28-0.67; p = 0.0002, I 2 = 23%) than in clean-contaminated surgeries Introduction An estimated 234,2 million surgeries are performed annually resulting in more than 7 million complications worldwide [1]. Of these surgical site infections (SSI) are one of the most frequent complications following abdominal surgery and cause considerable morbidity, mortality and health-care costs. Despite the implementation of preventive measures SSI rates in prospective trials with adequate follow-up and standardized SSI definition in abdominal surgical patients remain high and vary from 15% to 30% [2][3][4][5]. In the U.S. an estimated 300,000 to 500,000 SSIs occur annually [6][7][8]. SSIs are associated with a twofold increased relative risk of in-hospital mortality [8] and over one third of postoperative deaths in patients with SSIs are attributable to the infection [9]. Furthermore, several studies have shown an increase in the length of hospital stay between 6-24 days [8,[10][11][12][13]. The resulting direct costs have to be added to the indirect costs resulting in substantial expenses to the health care system and the society. [14][15][16].
The incidence of postoperative SSIs varies substantially depending on the degree of intraoperative contamination (clean, clean-contaminated, contaminated or dirty operations as defined by the Centers for Disease Control and Prevention, CDC [17]) and the site of operation [18][19][20]. Colorectal surgery has been associated with high rates of SSIs in studies with adequate design and follow-up [5,21,22]. While in the past definitions of SSIs varied, over the last decade the SSI definition of the CDC has gained universal acceptance [17].
As the most frequent pathogens causing postoperative SSIs following abdominal surgery are endogenous pathogens from the skin or gastrointestinal tract [17,23] it has been proposed that protecting the wound edges from bacterial invasion during surgery might reduce SSIs [24]. To this end impervious circular wound edge protectors (CWEP) have been proposed as in abdominal surgery as a simple, easy-to use and cost-effective intervention to reduce SSI. There are different devices available, but they all fall into two main categories: a.) devices with a single semirigid plastic ring placed into the abdominal cavity after laparotomy with an impervious drape attached, which comes out of the abdomen to protect the incisional edges (single-ring device) (S1 Fig); b.) devices with two semi-rigid plastic rings connected by an impervious drape. By placing one of the rings inside the abdomen after laparotomy and one externally, the drape protects the wound edges during surgery (double-ring device) (S1 Fig). While CWEPs effectively protect wound edges from bacterial invasion [25] their clinical effectiveness has been disputed and their widespread use has been hampered by conflicting trial results with some authors reporting beneficial effects [26][27][28] while others found no benefit [29][30][31]. Two previous smaller reviews based on limited patient numbers and low-quality evidence reported a beneficial effect of CWEPs, but called for adequately powered, high-quality trials [32,33] which have been conducted since then. The objective of our meta-analysis, therefore, is to evaluate the effectiveness of CWEPs in reducing SSIs compared to non-CWEP usage in patients undergoing open abdominal surgery based on all available RCTs and in a number of risk constellations.

Methods
The review is reported in line with current PRISMA guidelines [34] (S1 PRISMA Checklist). The systematic review and meta-analysis was conducted according to a pre-specified protocol, which is available upon request.

Eligibility Criteria
We included all RCTs in human subjects of any age comparing the use of an intraoperative CWEP with standard care in open abdominal surgery (elective and emergent). Trials were included if CWEPs were the only difference between trial arms. The CWEP intervention was defined as any intervention that covered the wound edges (i.e. skin, subcutaneous tissue, muscle and fascia) during surgery with an impervious plastic drape. We included trials using singlering as well as double-ring devices (S1 Fig). We excluded trials evaluating adhesive incision drapes that are placed on the skin and cut during laparotomy as they do not cover the incisional edges during surgery. We included any trial that reported SSI based on clinical signs, but excluded trials that used purely microbiological outcome definitions as purely bacteriological outcome definitions inadequately differentiate between bacteriological contamination and true SSI. Furthermore, we considered SSIs at the laparotomy site only, but not at alternative sites (e.g. drainage sites which were not covered by the CWEP). No restrictions were defined for year of publication, language or length of follow-up. Moreover, the references of the included articles were hand-searched to identify additional relevant studies. We used the patient-intervention-comparison-outcome (PICO) scheme to build our search strategy using search terms describing the intervention (CWEP) and the outcome (SSI). Sensitive search strategies were developed for all databases using wildcards and adjacency terms were appropriate. The search strategy for the MEDLINE search via OvidSP is shown in Fig 1. The search strategies for the other databases were adapted to the specific vocabulary of each database. We combined our search of the MEDLINE database via Ovid with the Cochrane Highly Sensitive Search Strategy for identifying RCTs in MEDLINE, sensitivity maximizing version, Ovid format (2008 revision) [35].

Trial Selection and data extraction
Titles and abstracts of studies were screened by two reviewers (A.L.M., T.C.M.) independently. In case of disagreement, a consensual decision between the two reviewers under involvement of a third independent reviewer (J.K.) was reached. Data was entered in RevMan 5 software 5.2. [36]. The following data items were extracted: year of publication, country of origin, study design, number of participants (total and for each group), type of surgery performed, risk factors for SSIs, grade of surgical contamination, inclusion and exclusion criteria, retractor type (single-ring vs. double-ring), description of the intervention, description of the control intervention, described and reported outcomes, SSI definition used, duration of follow-up, outcome data and effect sizes, number of patients lost to follow-up/withdrawals, funding and competing interests.

Assessment of risk of bias in individual studies
Bias was judged using The Cochrane Collaboration tool of for assessing quality and risk of bias [37]. Trials were defined as having an overall high risk of bias, if they were assessed high risk in any of the following domains: random sequence generation, allocation concealment, blinding of outcome assessors or if the duration of follow-up was inadequate (less than 30 days). As blinding of surgeons performing the CWEP intervention is impossible, double-blinding is defined as blinding of patients and SSI outcome-assessors.

Statistical methods
Risk ratio (RR) was used as summary measure for the dichotomous outcome: presence or absence of SSI. Results are presented with 95% confidence interval (CI). RevMan 5 software (version 5.2.) was used for data analysis [36]. Clinical heterogeneity was assessed. Data analysis was pre-specified using a random effects model given the variation in type of surgery and SSI definition (clinical heterogeneity). However, since random effects models do not give reliable estimates when few studies are included [38] and give more weight to small trials at the expense of large trials [39], analyses were also performed using a fixed-effects model. Statistical heterogeneity was assessed by using I 2 statistics and results of over 60% were considered as substantial heterogeneity. Missing data (e.g. death, re-laparotomy, lost to follow-up) were treated in the following prespecified way: when trials specified missing data for each treatment group, deaths and re-laparotomies were counted as SSI in both the intervention and control group, because a.) SSIs are a frequent cause of re-laparotomy and/or death, b.) death and re-laparotomy are a clinical 'worst case' outcome; therefore, not considering death and re-laparotomies would underestimate the risk for patients. Missing data for other reasons (e.g. lost to follow-up) were considered non-SSI cases in both groups given the overall small number of SSIs. When trials reported missing data, but did not specify the trial arm, only complete case data was used for analysis. Similarly, for trials that did not specify group numbers post-randomization and before drop-out, complete case data was used for analysis. For the pre-specified sensitivity analyses other imputation methods were applied: a.) all patients excluded post-randomization/with missing data were assumed to have developed SSIs in both groups; b.) patients excluded post-randomization were assumed to have developed SSI in the CWEP group, but not have developed SSI in the control group. Furthermore, a pre-specified sensitivity analysis after removing trials at high overall risk of bias was conducted.
The risk of publication bias was assessed by means of a funnel plot for the endpoint SSI.

Additional analyses
In addition to the sensitivity analyses described above, the following pre-specified subgroup analyses (effectiveness of CWEP vs. non-CWEP on SSI rate) were performed: a.) in colorectal vs. non-colorectal surgery; b.) for different degrees of intraoperative contamination (clean, clean-contaminated, contaminated or dirty operations as defined by the CDC [17]); c.) for single-ring CWEPs; d.) for double-ring CWEPs. Furthermore, two post-hoc sensitivity analyses were performed: a.) including only multicenter trials; b.) excluding underpowered trials with less than 40 patients per group.

Trial characteristics
We identified 990 studies, 987 studies by database searching and 3 studies by hand searching. Of these 392 duplicates were removed (Fig 2). Of the remaining 598 studies titles and abstracts were assessed for eligibility and 573 studies were consequently removed according to pre-specified criteria. Of the remaining 25 trials full-text articles were retrieved, of which 9 had to be excluded on final analysis as they investigated the wrong outcome (bacteriological outcome), used a CWEP soaked in povidone-iodine (unclear if observed effect is due to CWEP, or disinfectant) or studied a bundle of interventions (S1 Table). Sixteen RCTs were included in the final meta-analysis [40][41][42][43]27,[44][45][46][47]31,48,49,29,50,26,[51][52][53]. The characteristics of the 16 included RCTs are listed in Table 1. The trials reported SSI outcomes in a total of 3695 patients, of which 1846 received CWEPs as investigational intervention. One trial investigated three patient groups [29], all other trials investigated two parallel study groups. All trials were single-center studies with the exception of the trials by Pinkney et al.  [31]. In the reports by Sookhai [51] and Psaila [29] it remains unclear if trials were conducted at a single or two trial centers. Most trials were conducted in patients undergoing colorectal surgery [26,31,[40][41][42][43] or general abdominal surgery [27][28][29]45,49,53]. Two trials focused specifically on patients undergoing open appendectomy [44,48] and one on women undergoing cesarean section [52]. All studies prespecified SSI as an outcome, but there was considerable variation in the definitions applied (Table 2). Six trials referred to the SSI definition of the CDC [40,42,27,45,49,26]. In the remaining studies the authors used own definitions of SSI. Furthermore, length of follow-up varied widely. Five trials did not specify the duration of follow-up [27,29,31,41,52] and 9 trials had a length of follow-up of 30 days or more as required by current guidelines [26,28,31,40,42,43,46,49,50]. Baier et al. merely conducted a telephone interview on postoperative day 30 [40].

Risk of bias
Risk of bias is summarized in Fig 3 according to the Cochrane Collaboration proposal [37]. A risk of bias graph and a detailed assessment of bias for each individual study are provided in S2 Fig and S2 Table. Frequently, there was no report whether patients were excluded post randomization [27][28][29]31,43,50,52] or missing data of patients excluded post-randomization were not specified (attrition bias; Table 2) [41,42,44,48,53]. Many trials did not report if outcome assessors were blinded [29,31,41,43,48,52] or used an unblinded outcome assessment [40]. Only 4 trials described an adequate process of random sequence generation [26,44,46,49]. Similarly, few trials reported allocation concealment [26,42,46,48,49]. In the trial by Alexander-Williams, study personnel apparently had access to the patient allocation scheme [53]. Other forms of bias concerned the SSI definition employed and the length of follow-up or the failure to report conflicts of interest or funding sources [27][28][29]31,40,41,43,48,50,52,54]. In the trial by Alexander-Williams an industrial sponsor assisted with the design of the trial and preparation of the  give informed consent were excluded from the study.
• Patients undergoing laparoscopic colorectal resection were also excluded because of the concern of extraction site metastases in the abscence of wound protection.  randomization schedule [53]. Furthermore, a number of trials failed to report comparability of study arms with respect to SSI risk factors (see Table 1). Bätz et al failed to clearly define SSI in their trial [41]. Lee et al. conducted a post-hoc interim analysis, which lead to discontinuation of the trial [44]. The risk of publication bias was examined using a funnel plot (Fig 4). The asymmetry of the plot is due to 4 studies by Bätz et al. [41], Cheng et al. [42], Lee et al. [44], and Theodoridis et al. [52], that favor CWEPs but have a higher standard error of the effect estimate.

Results of individual studies and synthesis of results
The individual risk ratios and 95% confidence intervals (95%CI) for the random effects model meta-analysis of the 16 included RCTs are shown in Fig 5. The 6 trials published before 1990 all failed to show a statistically significant effect [29,31,41,43,48,53]. Of the 10 trials published after 1990, seven trials reported a significant reduction in SSIs in the CWEP group in their original report [26][27][28]42,44,46,50], a result which was not reproducible in our random effects model for the trials by Cheng et al. [42] and Horiuchi et al. [27] although both trials clearly favored CWEPs. Two trials after 1990 failed to show a significant reduction in SSI [40,49] and one trial did not elaborate its statistical analysis [52]. Cheng et al. reported the most favorable result for CWEPs (risk ratio, 0.07; 95%CI, 0.00-1.16) [42].

Sensitivity analyses of pooled data
In order to test the robustness of our results, we performed three pre-specified sensitivity analyses. First, we counted all patients excluded post-randomization (i.e. with missing data) as SSI in both groups. In the resulting random-effects model meta-analyses the pooled risk ratio was 0.70 (95%CI, 0.54-0.89; p = 0.004; I 2 = 55%) and 0.73 (95%CI, 0.64-0.83;p<0.0001;I 2 = 55%) in a fixed-effects model in 3695 patients. These results indicate that in the unlikely event that all missing patients would have developed a SSI, CWEPs still conferred a significant benefit. Secondly, we analyzed a potential 'worst case' scenario for CWEPs by imputing all missing data as SSI in the CWEP group and as non-SSI cases in the control group in the same 3695 patients. In this highly unrealistic event the pooled risk ratio for SSI was 0.76 (95%CI, 0.58-1.02; p = 0.06; I 2 = 63%) in a random effects model and 0.83 (95%CI, 0.72-0.95; p = 0.009; I 2 = 63%) in a fixed-effects model. Finally, we conducted a meta-analysis excluding all trials that were at high overall risk of bias [40,44,52,53] (see methods section for definition and Fig 3). In this case the pooled risk ratio was 0.61 (95%CI, 0.45-0.83; p = 0.002; I 2 = 59%) in a random-effects model and 0.67 (95%CI, 0.57-0.78; p<0.0001; I 2 = 59%) in a fixed effects-model in 3065 patients.

Summary of findings
SSIs remain one of the most frequent complications in open abdominal surgery. CWEPs are a simple and fast intervention that hold promise to reduce SSIs, but reports on its effectiveness have yielded conflicting results requiring a systematic review of available data. This was confirmed by our literature search, in which we identified 9 RCTs that reported positive results for CWEPS [26][27][28]41,42,44,46,50,52] while 7 RCTs failed to show a benefit [29,31,40,43,48,49,53]. Furthermore, two previous systematic reviews based on limited patient numbers have uncovered the methodological flaws in many previous studies investigating CWEPS [32,33] and have called for high-quality multicenter RCTs to be performed. Several high-quality trials including more than 1800 additional patients have been published since, necessitating our current reevaluation and meta-analysis, which elucidates the clinical effectiveness of CWEPs based on sound evidence.
We calculated a pooled risk ratio for the 3695 included patients of 0.65 (95%CI, 0.51-0.83) in a random-effects model. When a fixed-effect model was used the pooled risk ratio did not differ greatly (risk ratio 0.68; 95%CI, 0.60-0.79) indicating moderate inter-study heterogeneity. This was confirmed by the I 2 statistics (52%). Results indicate that CWEPs significantly reduce SSI in open abdominal surgery by 32-35% compared to controls. This would translate in a number-needed-to-treat of 15 (95%CI 10.9-23.3) i.e. 15 patients wound need to be treated with a CWEP to prevent one SSI. Given the large number of included RCTs and patients, these findings seem reliable. Furthermore, several of the included trials were conducted with high methodological standards and exhibited low-risk of bias [26,46,49] (Fig 3). In addition, we have included all randomized patients in our analyses as far as data were available and not just complete case data to prevent attrition bias as the proportion of missing outcome data in some trials compared with the observed event risk may well have induced clinically relevant bias.
Choosing a conservative way of imputation (deaths and relaparotomies were counted as SSIs) further supports our findings. Finally we conducted a number of sensitivity analysis to test the robustness of our results. Neither counting all patients excluded post-randomization as SSIs nor by excluding trials that were deemed at overall high risk of bias or underpowered trials changed the significance of our findings. By constructing a 'worst case' scenario (counting all missing data as SSI in the CWEP and as non-SSIs in the control group) however, showed a statistically significant result only in the fixed-effect model (RR 0.83; 95%CI, 0.72-0.95) but not for the random-effect model (RR 0.76; 95%CI, 0.58-1.02).
As CEWPs have been reported to be more effective in contaminated and colorectal operations, we have conducted several subgroup analyses. Pooling data from 11 RCTs and more than 1500 patients confirmed an overall significant effect of CWEPs in colorectal surgery, but the effect size (risk ratio 0.65; 95%CI, 0.44-0.97) was comparable to the one seen for all open abdominal patients. This is not surprising, if one assumes that most colorectal surgeries were clean-contaminated cases which were by far the most frequent type of surgeries conducted in the included RCTs (1456 cases out of 2228 operations that specified the degree of contamination). This might also explain that a similar risk ratio was found in the subgroup of clean-contaminated surgeries (risk ratio 0.63; 95%CI, 0.40-0.99). The subgroup analysis for clean, contaminated and dirty-infected surgeries have to be treated with caution and are exploratory at best, because of the limited number of patients, high CIs and high risk of bias of some of the included trials. However, our meta-analysis currently does not support the use of CWEPs in dirty surgeries. One might hypothesize that in dirty operations a contamination of the abdominal wall and subcutaneous tissue might occur upon laparotomy by spillage, i.e. before the CWEP can be put into place.
It was proposed that double-ring devices might be more effective than single-ring devices in protecting against SSIs [33]. Although we found the same phenomenon in our subgroup analysis (Fig 6B and 6C), as the pooled risk ratio for single ring devices was 0.71 (95%CI, 0.54-0.92), but 0.29 (95%CI, 0.15-0.55) for double-ring devices, we believe that this data does not sufficiently support the hypothesis. For single ring devices 11 RCTs with more than 2800 patients were available including two large multicenter trials with high methodological quality [46,49], while only 5 RCTs including 755 patients studied double-ring devices. With the exception of the study by Reid et al. [26] all of the double-ring studies exhibited considerable risk of bias [27,42,44,52]. Furthermore, three of these trials [42,44,52] were identified in the funnel plot analysis to exhibit considerable risk of publication bias (Fig 4).

Limitations
There are several limitations of our analysis. First, although our analysis was confined to studies with clinical definitions of SSI, numerous different definitions of SSI were used in the included RCTs (Table 2). This is in line with a report by Bruce et al. who reported more than 40 different SSI definitions in 80 publications in a 5-year period [55]. Most importantly, a number of authors did either not specify the length of follow-up [27,29,41] or used inadequate followup durations [44,52,53]. We have tried to eliminate the bias of these studies by excluding these trials in our sensitivity analysis. Only 6 trials used the internationally accepted SSI definition of the CDC [26,27,40,42,46,49].
Furthermore, a number of trials failed to report clear inclusion and exclusion criteria (Table 1). Therefore, the risk of SSIs cannot be accurately assessed in these patient populations. Although a number of risk factors for SSIs have been well documented in the literature, including age, obesity, smoking, diabetes mellitus, comorbidities, preoperative antibiotic prophylaxis and the duration of the surgical procedure [17], many of the RCTs that we have included in our meta-analysis neither reported these risk factors sufficiently nor adjusted for them in their analysis (Table 1). Not controlling for SSI risk factors appropriately makes it difficult to assess the real effect of CWEPs.
Finally, the methodological quality of some of the included trials was poor. A frequent problem was the inadequate reporting of randomization procedures, allocation concealment, control interventions, outcome definitions or the length of follow-up. Frequently, trials failed to report on patients excluded post-randomization or did not specify the group allocation of excluded patients. Moreover, some trials exhibited flaws that could have influenced the outcome, like the unblended wound assessment in the trial by Baier et al. [40] or the lack of allocation concealment in the trial by Alexander-Williams et al. [53].
It is noteworthy, though, that a number of trials including a recent methodologically well conducted multicenter trial from the UK [49] failed to show a benefit for CWEPs. It remains unclear why CWEPs in open abdominal surgery failed to show a benefit in these trials, but it might well be, that aspects not accounted for in the study conduct influenced the efficacy of CWEPs. Of the seven studies that reported negative results in their original publications, 4 were conducted in the UK [29,43,49,53] and two by the same team of authors in Sweden [31,48]. The remaining one was conducted in Germany and although the authors failed to show a significant benefit of CWEPs, results clearly favored the intervention group (20 SSIs in 98 cases vs. 30 SSIs in 101 cases) [40]. Positive results, on the other hand, have been reported from Australia [26], Germany [41,46], Ireland [28,50], Japan [27], Malaysia [42] and the USA [44]. Pinkney and colleagues [49] have recently pointed out that changing of gloves and instruments after removal of the CWEP and before closure of the fascia and skin would not be standard practice in the participating UK trial sites, while it is in other countries and hospitals. These aspects warrant further evaluation in future studies.

Conclusion
Our meta-analysis conducted of 16 RCTs and over 3600 patients showed that CWEPs are effective in reducing SSIs in open abdominal surgery compared to standard care. Our results are based on solid evidence from multiple trials, some with high methodological quality, and were consistent across multiple sensitivity analyses.