RCT vs cRCT study designs
Sixth annual Clinic on Meaningful Modeling of Epidemiological Data
June 1-12, 2015, African Institute for Mathematical Sciences, Muizenberg, Cape Town, RSA
Return to the Main Page.
Return to the Schedule.
Return to the list of potential projects.
This group will compare randomized controlled trial (RCT) and community-randomized controlled trial designs (CRCT) in the context of growing and declining epidemics with large amounts of geographic heterogeneity. This project will build off of the following paper, and the associated GitHub code repository and supplementary appendix:
Things to consider
- This group is recommended for:
- Participants interested in epidemiological study design.
- Participants interested in simulating studies.
- Participants interested in classical epidemiological analyses.
- Participants interested in studying interventions during acute emerging epidemics.
- Participants with some background in coding (to allow engagement with an already large code repository).
- This group will have the opportunity to engage in any of the following:
- Simulate infection in populations of individuals in geographically distinct clusters under various study designs & for varying intervention efficacies.
- Generate data from these simulated trial populations that is comparable with real data collected from a trial.
- Analyze these data with generalized linear mixed models (GLMMs).
- Sessions that are particularly important for group members to attend or review before Week 2:
- Thursday June 4 - Lecture and Computer Session: Study Design and Analysis in Epidemiology: Where does modeling fit? (Bellan) and Lab 3: Study Design in Epidemiology
- Saturday June 6 - Lecture: Participatory coding for Variability, Sampling Distributions, and Simulation Lecture (Bellan, with Pulliam)
- Monday June 8 - Real-world example (optional lecture): Faculty Research Lecture #3 (Bellan)
- Thursday June 11 - Lecture: Modeling for policy (Williams)
There are well-known tradeoffs between RCT and CRCT designs. RCT designs have greater statistical power (probability of detecting that an efficacious intervention is efficacious) for a given sample size because randomization of interventions at the individual level is better than randomization at the cluster level, since individuals within clusters have similar risk levels. However, RCTs cannot be used to evaluate interventions that can only be applied at the cluster level (e.g. the construction of a new hospital, or new district-level policies) or assess the effect of indirect benefits of an intervention (e.g. vaccine-derived herd immunity, which protects unvaccinated individuals who are near vaccinated individuals). The difference in statistical power between RCT and CRCT designs is known to depend on the amount of variation between-clusters relative to the amount of individual variation. However, less is known about the effect of spatiotemporal variation on their relative statistical power. This project seeks to characterize this effect.
- Subnational Ebola epidemic data from West Africa or simulated data.
- Bellan, SE, JRC Pulliam, CAB Pearson, D Champredon, SJ Fox, L Skrip, AP Galvani, M Gambhir, BA Lopman, TC Porco, LA Meyers, J Dushoff. (2015) The statistical power and validity of Ebola vaccine trials in Sierra Leone: A simulation study of trial design and analysis. Lancet Infectious Diseases.
- Randomized controlled trial lab
- Participatory coding