Bias Formulas for Sensitivity Analysis for Direct and Indirect Effects

A key question in many studies is how to divide the total effect of an exposure into a component that acts directly on the outcome and a component that acts indirectly, ie, through some intermediate. For example, one might be interested in the extent to which the effect of diet on blood pressure is mediated through sodium intake and the extent to which it operates through other pathways. In the context of such mediation analysis, even if the effect of the exposure on the outcome is unconfounded, estimates of direct and indirect effects will be biased if control is not made for confounders of the mediator-outcome relationship. Often data are not collected on such mediator-outcome confounding variables; the results in this paper allow researchers to assess the sensitivity of their estimates of direct and indirect effects to the biases from such confounding. Specifically, the paper provides formulas for the bias in estimates of direct and indirect effects due to confounding of the exposure-mediator relationship and of the mediator-outcome relationship. Under some simplifying assumptions, the formulas are particularly easy to use in sensitivity analysis. The bias formulas are illustrated by examples in the literature concerning direct and indirect effects in which mediator-outcome confounding may be present.

[1]  Tyler J VanderWeele,et al.  Alternative Assumptions for the Identification of Direct and Indirect Effects , 2011, Epidemiology.

[2]  L. Keele,et al.  A General Approach to Causal Mediation Analysis , 2010, Psychological methods.

[3]  T J VanderWeele,et al.  Direct and Indirect Effects for Neighborhood-Based Clustered and Longitudinal Data , 2010, Sociological methods & research.

[4]  L. Keele,et al.  Identification, Inference and Sensitivity Analysis for Causal Mediation Effects , 2010, 1011.1079.

[5]  Stijn Vansteelandt,et al.  Estimating Direct Effects in Cohort and Case–Control Studies , 2009, Epidemiology.

[6]  Tyler J. VanderWeele,et al.  Concerning the consistency assumption in causal inference. , 2009, Epidemiology.

[7]  N. Perkins,et al.  Quantification of collider-stratification bias and the birthweight paradox. , 2009, Paediatric and perinatal epidemiology.

[8]  Sharon Schwartz,et al.  Opening the Black Box: a motivation for the assessment of mediation. , 2009, International journal of epidemiology.

[9]  Juni Palmgren,et al.  Sensitivity Analysis for Principal Stratum Direct Effects, with an Application to a Study of Physical Activity and Coronary Heart Disease , 2009, Biometrics.

[10]  J. Robins,et al.  Minimal sufficient causation and directed acyclic graphs , 2009, 0906.1720.

[11]  R. Gallop,et al.  Mediation analysis with principal stratification , 2009, Statistics in medicine.

[12]  T. VanderWeele.  Mediation and mechanism , 2009, European Journal of Epidemiology.

[13]  Tyler J. VanderWeele,et al.  Marginal Structural Models for the Estimation of Direct and Indirect Effects , 2009, Epidemiology.

[14]  T. VanderWeele.  Simple relations between principal stratification and direct and indirect effects , 2008 .

[15]  Els Goetghebeur,et al.  Estimation of controlled direct effects , 2008 .

[16]  Mark J van der Laan,et al.  The International Journal of Biostatistics Direct Effect Models , 2011 .

[17]  T. VanderWeele.  Sensitivity Analysis: Distributional Assumptions and Confounding Assumptions , 2008, Biometrics.

[18]  Christos Davatzikos,et al.  Are brain volumes based on magnetic resonance imaging mediators of the associations of cumulative lead dose with cognitive function? , 2008, American journal of epidemiology.

[19]  D. Mackinnon.  Introduction to Statistical Mediation Analysis , 2008 .

[20]  James M Robins,et al.  Directed acyclic graphs, sufficient causes, and the properties of conditioning on a common effect. , 2007, American journal of epidemiology.

[21]  Kevin G Lynch,et al.  Causal Mediation Analyses with Rank Preserving Models , 2007, Biometrics.

[22]  S. Geneletti.  Identifying direct and indirect effects in a non‐counterfactual framework , 2007 .

[23]  M. Hernán,et al.  The birth weight "paradox" uncovered? , 2006, American journal of epidemiology.

[24]  Dylan S. Small,et al.  Defining and Estimating Intervention Effects for Groups that will Develop an Auxiliary Outcome , 2006, math/0609457.

[25]  D. Shen,et al.  Past adult lead exposure is linked to neurodegeneration measured by brain MRI , 2006, Neurology.

[26]  Mark J van der Laan,et al.  Estimation of Direct Causal Effects , 2006, Epidemiology.

[27]  Chen Avin,et al.  Identifiability of Path-Specific Effects , 2005, IJCAI.

[28]  Richard F MacLehose,et al.  A further critique of the analytic strategy of adjusting for covariates to identify biologic mediation. , 2004, Epidemiologic perspectives & innovations : EP+I.

[29]  J. Robins,et al.  A Structural Approach to Selection Bias , 2004, Epidemiology.

[30]  D. Rubin.  Direct and Indirect Causal Effects via Potential Outcomes * , 2004 .

[31]  T. Blakely.  Commentary: estimating direct and indirect effects-fallible in theory, but in the real world? , 2002, International journal of epidemiology.

[32]  A. Wilcox,et al.  On the importance--and the unimportance--of birthweight. , 2001, International journal of epidemiology.

[33]  Judea Pearl,et al.  Direct and Indirect Effects , 2001, UAI.

[34]  B. Gordon,et al.  Past adult lead exposure is associated with longitudinal decline in cognitive function , 2000, Neurology.

[35]  James M. Robins,et al.  Association, Causation, And Marginal Structural Models , 1999, Synthese.

[36]  J. Pearl.  Causal diagrams for empirical research , 1995 .

[37]  J. Robins,et al.  Identifiability and Exchangeability for Direct and Indirect Effects , 1992, Epidemiology.

[38]  D. A. Kenny,et al.  Process Analysis , 1981, Plastics Process Analysis, Instrumentation, and Control.

[39]  Tyler J. VanderWeele,et al.  Conceptual issues concerning mediation, interventions and composition , 2009 .

[40]  J. Robins.  Addendum to “a new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect” , 1987 .

[41]  J. Robins.  A new approach to causal inference in mortality studies with a sustained exposure period—application to control of the healthy worker survivor effect , 1986 .

[42]  D. A. Kenny,et al.  The moderator-mediator variable distinction in social psychological research: conceptual, strategic, and statistical considerations. , 1986, Journal of personality and social psychology.