[ABE-L] Seminário do PIPGEs UFSCar/USP

[Ricardo Sandes Ehlers]

Caros redistas,

O PIPGEs UFSCar/USP convida para uma palestra no dia 12/4/2018 às 10h no
auditório Luiz Antonio Fávaro, ICMC/USP.

Estimation of intervention and spatial spillover effects in cluster
randomised trials

Karin Anaya-Izquierdo, University of Bath, UK

We describe a spatial regression methodology for the analysis of cluster
randomised trials (CRTs) with count outcomes, taking indirect effects into
account.
The assumption  between-cluster independence in CRT's is potentially
violated in trials against infectious diseases, whose clusters are often
defined geographically, potentially inducing spatial correlation and
indirect effects.  We use spatial regression models with Gaussian random
effects, where the individual outcomes have marginal distributions
overdispersed with respect to the Poisson and the corresponding
intervention effects have a marginal interpretation. Two types of effect
are distinguished and estimated: spillover dependence, which is
cross-cluster correlation between individual outcomes; and spillover
indirect effect, which is change in the intervention effect depending on
the proximity of individuals to those in the intervention arm. Orthogonal
regression is used to avoid bias arising from collinearity, a phenomenon
which has become known as spatial confounding. We also show that
coefficients from spatial models with a certain form of homoscedasticity
can be interpreted simply as intervention effects. The standard intrinsic
conditional autoregression (ICAR) model does not have this property, but we
use a normalized version which does. In order to quantify the proximity of
individuals in the intervention arm we use Tukey’s half space depth. We fit
the models in a Bayesian framework using integrated nested Laplace
approximations (INLA), and illustrate the methodology using data from a
pair-matched CRT done in Venezuela against the mosquito Aedes aegypti,
which is a vector of dengue and Zika.

Saudações,
Ricardo
-------------- Próxima Parte ----------
Um anexo em HTML foi limpo...
URL: <https://lists.ime.usp.br/archives/abe/attachments/20180409/a3d32867/attachment.html>