NorPen webinar och årliga möte

*Presentation var en introduktion till G methods och ett smakprov på den längre NorPEN2021 kursen om G-methods i relation till läkemedelsepidemiologi. Presentationen kommer vara på engelska.

G-methods is a class of methods for estimating the causal effects of time-varying treatment strategies in longitudinal studies where time-varying confounders may be affected by past treatment. G-methods specifically aim to estimate Robins’s g-formula, a function of only measured study variables.  Under assumptions that include no unmeasured confounding, the g-formula indexed by a particular time-varying treatment strategy equals the (counterfactual) outcome mean in the study population had all individuals adhered to that strategy. The g-formula is usually a high-dimensional function when the dimension of measured confounders is high and/or there are many follow-up times. Different g-methods (e.g. inverse probability weighting, parametric g-computation, targeted maximum likelihood estimation) constitute different estimation methods for this function of the longitudinal data.  In this presentation, we will introduce counterfactual causal reasoning that motivates the g-formula as a target of statistical analysis and give a high-level overview of some different estimation methods.