chapter  23
48 Pages

Estimation of the causal effects of time-varying exposures

In this chapter we describe methods for the estimation of the causal effect of a timevarying exposure on an outcome of interest from longitudinal data collected in an observational study. The terms “exposure” and “treatment” will be used synonymously and

interchangeably. We assume a fixed study population, that is, a closed cohort with a welldefined, known start of follow-up date for each subject. Time will refer to time since start of follow-up, which we also refer to as time since baseline. We only consider estimation of the effect of exposures occurring at or after the start of follow-up because the estimation of the effects of pre-baseline exposures is not possible without making strong, untestable assumptions. We refer to the exposure received at start of follow-up as the baseline exposure. Baseline covariates refer to covariates, including pre-baseline exposure, that occur prior to the baseline exposure. We classify exposures as either fixed or time-varying.