Revisiting the effectiveness of interventions to decrease surgical site infections in colorectal surgery: A Bayesian perspective

Abstract

To evaluate the evidence for interventions to decrease surgical site infections (SSIs) in colorectal operations using Bayesian meta-analysis. Interventions other than appropriate administration of prophylactic antibiotics to prevent SSIs have not been adopted widely, in part because of lack of recommendations for these interventions based on traditional meta-analyses. Bayesian methods can provide probabilities of specific thresholds of benefit, which may be more useful in guiding clinical decision making. We hypothesized that Bayesian meta-analytic methods would complement the interpretation of traditional analyses regarding the effectiveness of interventions to decrease SSIs. We conducted a systematic search of the Cochrane database for reviews of interventions to decrease SSIs after colorectal surgery other than prophylactic antibiotics. Traditional and Bayesian meta-analyses were performed using RevMan (Nordic Cochrane Center, Copenhagen, Denmark) and WinBUGS (MRC Biostatistics Unit, Cambridge, UK). Bayesian posterior probabilities of any benefit, defined as a relative risk of <1, were calculated using skeptical, neutral, and enthusiastic prior probabilities. Probabilities were also calculated that interventions decreased SSIs by ≥10%, and ≥20% using neutral prior probability distributions. A total of 9 Cochrane reviews met the search criteria. Using traditional meta-analysis methods, only laparoscopic colorectal surgery resulted in a significant reduction in SSIs and a recommendation for use of the intervention. Using Bayesian analysis, several interventions that did not result in "significant" decreases in SSIs using traditional analytic methods had a >85% probability of benefit. Also, nonuse of 2 interventions (mechanical bowel preparation and adhesive drapes) had a high probability of decreasing SSIs compared with their use. Bayesian probabilities and traditional point estimates of treatment effect yield similar information in terms of potential effectiveness. Bayesian meta-analysis, however, provides complementary information on the probability of a large magnitude of effect. The clinical impact of using Bayesian methods to inform decisions about which interventions to institute first or which interventions to combine requires further study.