Analysis of the impact of naturally-acquired immunity on Shigella vaccine performance to inform clinical trials: A simulation-based modeling study

Background Effective vaccines against Shigella , the leading bacterial cause of diarrheal disease globally, could prevent morbidity and mortality from shigellosis as well as limit antibiotic resistance. Several Shigella vaccine candidates are in late stages of development, and the design of large Phase 3 trials in target populations is underway. In high incidence settings, children are expected to rapidly accrue natural immunity such that after repeated pathogen exposure, unvaccinated infants may have similar levels of immunity as vaccinated infants. For this reason, in these settings estimates of vaccine efficacy (VE) may underestimate expected efficacy in a fully naïve population. To set expectations and support optimal planning of future Shigella vaccine trials, we aimed to quantify the potential impact of this phenomenon on VE estimates under different surveillance strategies and assumptions about Shigella immunity. Methods and findings We simulated Shigella vaccine trials in low-, medium-, and high-burden settings under different assumptions about vaccine- and infection-conferred immunity. We estimated VE against severe, Shigella -attributable disease and any Shigella -attributable diarrhea under six trial designs: active surveillance for Shigella infection or symptom-based reporting paired with single-outcome, stratified recurrent outcome, or crude recurrent-outcome Cox proportional hazards models. In each scenario, we evaluated bias, 95% confidence interval (CI) coverage, and the false negative rate. Only the trial design using active surveillance for infection and single outcome models was consistently unbiased across all examined scenarios. The other five designs underestimated VE by up to an absolute 20%, with a single outcome model paired with symptom-based reporting performing least poorly. Bias generally was worse at higher-burden sites. For realistically sized trials, the 95% CI coverage and false negative rate were also worse at higher-burden sites. The results were sensitive to input parameters which depended on assumptions about Shigella natural history based on limited published data. Conclusions The observed VE in Shigella vaccine trials is likely to underestimate true VE, though the magnitude of underestimation described here is conditional on the simulation’s inputs and assumptions. Careful trial design is needed to minimize this underestimation.