Causal inference has progressed dramatically in the past few decades, but the assumption of no interference between units remains difficult to relax. Interference occurs when the treatment of one unit can affect the outcome of another, making it difficult to justify the assumption for outcomes that depend on social interactions, such as infectious disease.
In cluster-randomized trials, the assumption of no interference is often justified by the distance between units, but this condition may not be feasible or even sufficient in some settings. In trials of infectious disease prevention, the estimate of the treatment effect will be attenuated if a fraction of the exposures in the treatment clusters come from outside these clusters.
This source of interference, the fraction of contacts sufficient for transmission that are within-cluster, is potentially measurable. We make use of epidemic models to infer the influence of interference on the force of infection upon members of a cluster. This allows for the development of a weighting of the treatment indicator that allows estimation of the treatment effect in the absence of interference in a proportional hazards or frailty model.
|
ASA Meetings Department
732 North Washington Street, Alexandria, VA 22314
(703) 684-1221 • meetings@amstat.org
Copyright © American Statistical Association.