Bayesian reanalysis of postoperative delirium prevention trials from a published systematic review: Posterior probability estimates for clinical decision-making

Background and Aims: Postoperative delirium (POD) is a common complication associated with morbidity, mortality, and healthcare costs. Traditional frequentist meta-analyses dichotomize conclusions based on P value thresholds. We aimed to complement existing evidence by reanalyzing published POD prevention trial data using Bayesian methods to calculate posterior probabilities of intervention effectiveness. Material and Methods: We conducted a Bayesian hierarchical meta-analysis using randomized controlled trial data from a published systematic review (Meng et al. , British Journal of Anaesthesia 2024). Of 201 trials, 118 comparing eight intervention categories against usual care were analyzed. We employed weakly informative priors and calculated posterior probabilities that the true odds ratio (OR) lay below defined thresholds: P (OR <1), P (OR <0.9), P (OR <0.8), and P (OR <0.7). Sensitivity analyses examined risk-of-bias filtering, alternative priors, and zero-event handling, including the Sweeting reciprocal-of-other-arm correction. Results: All eight interventions demonstrated high posterior probability of POD reduction. Dexmedetomidine (OR 0.47; 95% credible interval [CrI], 0.37–0.58) and hemodynamic management (OR 0.45; 95% CrI, 0.31–0.64) showed the most favorable posterior profiles, with P (OR <0.8) of >99.9% and 99.9%, respectively. Pairwise comparisons suggested that hemodynamic management and dexmedetomidine were among the more effective interventions, although these comparisons were indirect and exploratory. Conclusions: Dexmedetomidine and hemodynamic management showed favorable Bayesian posterior estimates. This reanalysis is not intended to replace the source meta-analysis but to improve clinical interpretability through probability-based estimates; absolute benefits remain conditional on baseline risk.