Contextualizing the use of corticosteroids in severe Pneumocystis jirovecii pneumonia through a Bayesian lens

<h2>Abstract</h2><h3>Background</h3> A recent multicenter randomized clinical trial (RCT) of steroids for severe <i>Pneumocystis jirovecii</i> pneumonia (PCP) in patients without HIV failed to demonstrate a significant benefit in its primary outcome of mortality at 28 days. We sought to contextualize the findings leveraging the data from HIV trials through a Bayesian meta-analysis. <h3>Methods</h3> We assumed a moderate prior probability that steroids would reduce mortality in PCP outside of HIV. Using data from the HIV RCTs (489 patients, 6 trials), we contextualized a recent RCT in patients without HIV using Bayesian meta-analysis on the log odds ratio (OR) scale. We chose a vague prior for the efficacy parameter (μ∼N[0,0.2.7²]) and an evidence-based informative prior (log-normal [−1.975,0.67]) for heterogeneity (τ). <h3>Results</h3> The meta-analytic probability of benefit at 28 days was >99.9% with an odds ratio of 0.51 (95% credible interval 0.32–0.77) and an estimated adjusted risk difference of −11.4%. <h3>Conclusions</h3> Our results suggest that patients without strong contraindications who have severe PCP should be considered for corticosteroid treatment regardless of HIV status.