Causal Effects of Time‐Varying Treatments: The G‐Formula

ABSTRACT

The g formula is a cornerstone method for estimating causal effects of time‐varying treatments using longitudinal observational data in the presence of time‐varying confounders that are affected by prior treatment. Standard regression techniques often fail in this setting because adjustment for such covariates can distort the very effects under study. The g‐formula addresses this problem by expressing the mean potential outcome under specified static or dynamic treatment regimes as a function of the joint distribution of covariates, treatments and outcomes, which can be approximated via parametric or semi‐parametric models and simulation. This review presents the g‐formula, emphasizing intuitive explanations. After outlining the causal framework and the core identification assumptions—consistency, sequential exchangeability and positivity—the article describes practical parametric g‐formula: model specification for covariate and outcome processes, implementation via forward simulation, and the interpretation of marginal causal contrasts between clinically relevant regimes. The g‐formula is then situated within the family of g‐methods alongside inverse probability weighting and targeted maximum likelihood estimation, highlighting complementary strengths and limitations. A dedicated section discusses concrete applications, including analyses of highly active antiretroviral therapy and AIDS or death, dynamic ‘when to start’ antiretroviral strategies in HIV and electronic health record‐based evaluations of blood pressure treatment targets. Practical guidance on modelling choices, diagnostics and transparent reporting is provided to support applied researchers considering g‐formula in clinical and epidemiological investigations.