Abstract:
|
Investigations of infectious diseases in cohort studies of clusters - households, villages, or small groups - often report risk ratios as a summary of the relationship between a binary covariate and outcome. Variants of Poisson regression are used for estimation of risk ratios when outcomes are binary. When the marginal outcome probability is correctly specified, modified Poisson regression can deliver consistent estimates of the risk ratio and robust standard errors, even when the correlation structure within clusters is unknown. Epidemiologists have warned that risk ratios may be biased when outcomes are contagious, but the nature and severity of this bias is not well understood. In this study, we assess the epidemiologic meaning of the risk ratio when outcomes are contagious by formulating a generic characterization of infectious disease transmission that includes possible transmission from an exogenous source. We exhibit analytically and by simulation the circumstances under which the estimated risk ratio can be seriously misleading. We explain these findings in the epidemiologic language of confounding and relate the directional bias to Simpson's paradox.
|