Advantages of a Two-Stage Randomized Trial Design to Evaluate Antimicrobial Treatment Strategies: a Simulation Study.

Abstract

Antimicrobial prescribing policies affect treated patients and, through altered transmission, their contacts. Two-stage randomized (2SR) designs can estimate these spillover effects but have not been applied to antimicrobial strategies. We built a stochastic agent-based model of a hospital ward with two competing strains (susceptible and resistant) to emulate a 2SR trial. Drug A covers only the susceptible strain, while Drug B covers both. Six clusters were randomized to a 90/10 or 50/50 Drug A/Drug B allocation, with individuals randomized to treatments within clusters. We estimated direct, indirect, total, and overall effects on mortality. Drug A recipients had higher mortality than Drug B recipients in every cluster, because Drug A does not cover the resistant strain (direct effect +4.75 to +7.08 percentage points [pp]). Among Drug A recipients, mortality was higher when the ward followed 90/10 than 50/50 (indirect effect +2.10 pp), driven by higher resistant-strain prevalence; the effect on Drug B recipients was minimal. The 50/50 strategy reduced total mortality (overall effect -4.00 pp), but this concealed a redistribution: resistant-strain deaths fell while susceptible-strain deaths rose. Findings were robust across sensitivity scenarios. The 2SR design captures spillover effects invisible to individually randomized trials.