Poisson regression with adjustment for contamination and non-compliance in cohort studies conducted to estimate intervention effectiveness

BackgroundThe effectiveness of an intervention such as cancer screening can be estimated by conducting a cohort study estimating the rate ratio between an exposed group invited to screening and a control group not invited to screening. A common issue is non-compliance, where not all individuals in the study group are exposed. Ignoring non-compliance can result in biased estimates of the exposure effect, but excluding non-exposed individuals may also be problematic as they may differ in risk profile from those who were exposed. A similar problem arises when members of the control group are inadvertently exposed (contamination). Observational studies face additional challenges due to confounding.ObjectiveTo report the development of a method to adjust rate ratio estimates in cohort studies for contamination and non-compliance, that also adjusts for confounding.MethodDerivation of the new method is outlined.ResultsThe method is illustrated in two examples.ConclusionThe results are comparable with a stratified estimate, but through the use of a Poisson regression model the range of possible analyses is extended, for example to tests of confounding factors and interaction.