TY - JOUR T1 - Using simulation to aid trial design: ring-vaccination trials JF - bioRxiv DO - 10.1101/071498 SP - 071498 AU - Matt D.T. Hitchings AU - Rebecca F. Grais AU - Marc Lipsitch Y1 - 2016/01/01 UR - http://biorxiv.org/content/early/2016/08/25/071498.abstract N2 - Background The 2014-5 West African Ebola epidemic highlights the need for rigorous, rapid clinical trials under difficult circumstances. Challenges include temporally and spatially patchy transmission, and the responsibility to deliver public health interventions during a randomized trial. An innovative design such as ring vaccination with an immediate arm and a delayed arm can address these issues, but complex trials raise complex analysis issues.Methods and Findings We present a stochastic, compartmental model for a ring vaccination trial of a vaccine for an Ebola-like disease. After identification of an index case, a ring of primary contacts is recruited and either vaccinated immediately or after a delay of 21 days. The primary outcome of the trial is effectiveness calculated from cumulative incidence in the two arms, counting cases only from a pre-specified window in which the immediate arm is assumed to be fully protected and the delayed arm is not protected. The results of simulating the trial are used to calculate the sample size necessary for 80% power and the estimates of effectiveness are reported under a variety of assumptions regarding the trial design and implementation.The three key components of sample size calculations – attack rate in controls, estimate of incidence difference between the arms, and intracluster correlation coefficient – are dependent on trial design and implementation in a way that can be quantitatively predicted by the model. Under baseline parameter assumptions, we found that a total of 8,900 study participants were needed to achieve 80% power to detect a difference in attack rate between the two arms, whereas a standard approach with the same parameters returns a necessary sample size of 7,100 individuals. Such a study would on average return a vaccine effectiveness estimate of 69.81%, with average 95% confidence interval (41.2%, 84.2%).We found that for this design the necessary sample size and estimated effectiveness are sensitive to properties of the vaccine – in particular, pre-exposure and post-exposure efficacy; to two setting-specific parameters over which investigators have little control – rate of infections from outside the ring and overall attack rate in the controls; and to three parameters that are determined by the study design – the time window in which cases are counted, intensity of case-detection and administrative delay in vaccinating individuals.This approach replaces assumptions about parameters in the trial with assumptions about disease dynamics and vaccine characteristics at the individual level.Conclusions Incorporating simulation into the trial design process can improve robustness of sample size calculations. Simulation can identify optimal values for study design parameters that can be controlled. For this specific trial design, vaccine effectiveness depends on properties of the ring vaccination design and on the measurement window, as well as the epidemiologic setting. Rejecting the null likely indicates one or more types of vaccine efficacy at the individual level, but the magnitude of the effect will vary across settings. ER -