Application of Bayesian Methods to Help Interpret the VANCO Trial Results.

Abstract

To determine whether a Bayesian analysis changes the results of the VANCO trial. A secondary analysis of a randomized clinical trial using Bayesian methods. Thirty-six US trauma centers. Patients ages 18-80 years with a tibial plateau or pilon fracture deemed high risk of infection and definitively treated with plate and screw fixation. Patients were randomly allocated to receive 1000 mg of intrawound vancomycin powder at their definitive fixation or to a control group that received no topical antibiotics. A deep surgical site infection requiring operative treatment within 6 months of definitive fixation. Secondary outcomes included gram-positive and gram-negative-only deep surgical site infections. Of the 980 patients randomized, 874 (89%) had at least 140 days of follow-up and were included in this Bayesian analysis. The estimated probability that intrawound vancomycin powder reduces the risk of a deep surgical site infection is >98% [relative risk (RR), 0.66; 95% credible interval (CrI), 0.46-0.98]. There is a >99% chance intrawound vancomycin powder reduces gram-positive infections and an 80% chance the magnitude of this risk reduction exceeds 35% (RR, 0.52; 95% CrI, 0.33-0.84) exists. It is unlikely (44%) that intrawound vancomycin powder prevents gram-negative surgical site infections (RR, 1.06; 95% CrI, 0.48-2.45). There is a high probability (>98%) that intrawound vancomycin powder reduces deep surgical site infections in patients with tibial plateau or pilon fractures at high risk of infection and even more likely it reduces deep infections with gram-positive pathogens (>99%). Therapeutic Level I. See Instructions for Authors for a complete description of levels of evidence.