Marginal versus conditional causal effects
AbstractConditional methods of adjustment are often used to quantify the effect of the exposure on the outcome. As a result, the stratums-specific risk ratio estimates are reported in the presence of interaction between exposure and confounder(s) in the literature, even if the target of the intervention on the exposure is the total population and the interaction itself is not of interest. The reason is that researchers and practitioners are less familiar with marginal methods of adjustment such as inverse-probability-weighting (IPW) and standardization and marginal causal effects which have causal interpretations for the total population even in the presence of interaction. We illustrate the relation between marginal causal effects estimated by IPW and standardization methods and conditional causal effects estimated by traditional methods in four simple scenarios based on the presence of confounding and/or effect modification. The data analysts should consider the intervention level of the exposure for causal effect estimation, especially in the presence of variables which are both confounders and effect modifiers.
Greenland S, Rothman KJ . Introduction to stratified analysis. In: Rothman KJ, Greenland S, Lash TL, Editors. Modern epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008. p. 258-82.
Greenland S, Lash TL, Rothman KJ.Concepts of interaction. In: Rothman KJ,Greenland S, Lash TL, Editors. Modern epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008. p. 71-83.
Hernan MA, Robins JM. Causal inference. London, UK: Chapman and Hall; 2015.
Greenland S, Mansournia MA. Limitations of individual causal models, causal graphs,and ignorability assumptions, as illustrated by random confounding and design unfaithfulness. Eur J Epidemiol 2015.
Jewell NP. Statistics for epidemiology.London, UK: Taylor & Franci; 2003.
Greenland S, Pearl J, Robins JM.Confounding and collapsibility in causal inference. Statist Sci 1999; 14(1): 29-46.
Greenland S. Absence of confounding does not correspond to collapsibility of the rate ratio or rate difference. Epidemiology 1996;7(5): 498-501.
Glymour MM, Greenland S. Causal diagrams. In: Rothman KJ, Greenland S, Lash TL, Editors. Modern epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008. p. 183-209.
Mansournia MA, Hernan MA, Greenland S.Matched designs and causal diagrams. Int J Epidemiol 2013; 42(3): 860-9.
Mansournia MA, Greenland S. The relation of collapsibility and confounding to faithfulness and stability. Epidemiology 2015; 26(4): 466-72.
Robins JM, Hernan MA, Brumback B.Marginal structural models and causal inference in epidemiology. Epidemiology 2000; 11(5): 550-60.
Hernan MA, Brumback B, Robins JM.Marginal structural models to estimate the causal effect of zidovudine on the survival of HIV-positive men. Epidemiology 2000;11(5): 561-70.
Mansournia MA, Danaei G, Forouzanfar MH, Mahmoodi M, Jamali M, Mansournia N, et al. Effect of physical activity on functional performance and knee pain in patients with osteoarthritis : analysis with marginal structural models. Epidemiology 2012; 23(4): 631-40.
Hubbard AE, Laan MJ. Population intervention models in causal inference. Biometrika 2008; 95(1): 35-47.
Vansteelandt S, Keiding N.Invited commentary:G-Computation–lost in translation? Am J Epidemiol 2011; 173(7):739-42.
Greenland S, Maldonado G. The interpretation of multiplicative-model parameters as standardized parameters. Stat Med 1994; 13(10): 989-99.
Greenland S. Introduction to regression modeling. In: Rothman KJ, Greenland S, Lash TL, Editors. Modern epidemiology. Philadelphia, PA: Lippincott Williams & Wilkins; 2008. p. 418-55.