Causal Inference
Estimating the impact of Shigella vaccines on growth outcomes and implications for clinical trial design Allison Codi* Allison Codi Elizabeth Rogawski-McQuade Razieh Nabi David Benkeser
Shigella vaccines could prevent growth faltering in children in low-resource settings, prompting interest in trials to demonstrate a vaccine effect (VE) on linear growth. However, the population-level VE on height-for-age z-score (HAZ) is expected to be small (<0.01) because most participants won’t experience shigellosis, such that the effect is diluted by children who were not at risk for Shigella-attributable growth faltering. Thus, standard VE estimation methods are likely to be underpowered at realistic trial sizes and at risk of producing null or inverse results.
We use causal inference to develop an alternative estimand that quantifies VE for HAZ in the subset of children with Shigella. We adjust for confounders of infection and growth and include children for whom the vaccine prevented infection. These children would be expected to experience the largest VE but would be excluded in a naïve comparison of children observed to have shigellosis in the trial (i.e., a comparison of growth among unvaccinated cases and vaccinated breakthrough cases). We compare power of our estimand to the population-level measure through trial simulations using data from recent observational studies of shigellosis for parameterization.
Simple population-level comparisons of HAZ by vaccine arm showed extremely limited power (<5%) and a negative point estimate for VE in more than 40% of simulated trials of realistic size (n=10,000-20,000). Analysis of VE for HAZ in the subset of children with vaccine-preventable Shigella has improved power (~20-30%) and decreased likelihood of demonstrating negative VE (<10% of simulated trials). While our method offers dramatic improvement over the population-level comparison, realistically sized trials will still likely be underpowered. Nevertheless, these novel methods may be relevant beyond the specific application to Shigella vaccines for studying other post-infection endpoints in a broad range of vaccine trials.