High-dose dual antibiotic cement for hip hemiarthroplasty : a post-hoc Bayesian analysis of a randomized controlled trial.

Abstract

In this study we aimed to evaluate how a post-hoc Bayesian analysis of results from the large multicentre World Hip Trauma Evaluation (WHiTE) 8 (High-dose dual-antibiotic-loaded cement for hip hemiarthroplasty) trial might aid clinical interpretation. Using anonymized data, a post-hoc Bayesian Poisson regression adjusting for age, sex, and centre (random effects), with 200,000 iterations, was undertaken to compare high-dose dual-antibiotic cement versus single-antibiotic cement for the primary outcome reduction of deep surgical site infection (dSSI) at 90 days. Two set of priors were used: reference priors and a data-derived prior. We assessed the posterior probabilities (PP) of relative risk (RR) (for any benefit (RR > 1), modest benefit (> 1.25), and large benefit (> 1/5)) and risk difference (RD). There were 38/2,187 (1.7%) and 27/2,219 (1.2%) dSSIs identified at 90 days in the single-antibiotic and high-dose dual-antibiotic cement groups, respectively (original analysis (OR) 1.44 (95% CI 0.88 to 2.37)). With a minimally informative reference prior there was a posterior median (PM) RR for dSSI of 1.42 (95% credible interval (CrI) 0.87 to 2.36), with a 92% PP of a RR > 1 and an 81% PP of a RD > 0.25 in favour of high-dose dual-antibiotic cement. When utilizing the data-driven prior without downweighting, there was a PM RR for dSSI of 1.62 (95% CrI 1.03 to 2.59), with a 98% PP of a RR > 1 and a 93% PP of a RD > 0.25. This Bayesian re-analysis demonstrates that it is highly probable (> 90%) that there is at least some reduction of dSSI related to high-dose dual-antibiotic cement use when performing hemiarthroplasty; adding useful clinical nuance to the original trial findings in context of the noteworthy detrimental impact of dSSI in the study population. This study also highlights the potential benefits of a Bayesian approach to analysis due to improved interpretability of trial outcomes.