## Abstract

In the context of causal mediation analysis, prevailing notions of direct and indirect effects are based on nested counterfactuals. These can be problematic regarding interpretation and identifiability especially when the mediator is a time-dependent process and the outcome is survival or, more generally, a time-to-event outcome. We propose and discuss an alternative definition of mediated effects that does not suffer from these problems, and is more transparent than the current alternatives. Our proposal is based on the extended graphical approach of Robins and Richardson (in: Shrout (ed) Causality and psychopathology: finding the determinants of disorders and their cures, Oxford University Press, Oxford, 2011), where treatment is decomposed into different components, or aspects, along different causal paths corresponding to real world mechanisms. This is an interesting alternative motivation for *any* causal mediation setting, but especially for survival outcomes. We give assumptions allowing identifiability of such alternative mediated effects leading to the familiar mediation g-formula (Robins in Math Model 7:1393, 1986); this implies that a number of available methods of estimation can be applied.

This is a preview of subscription content, access via your institution.

## References

Aalen O, Borgan O, Gjessing H (2008) Survival and event history analysis: a process point of view. Springer, New York

Aalen O, Cook R, Røysland K (2015) Does Cox analysis of a randomized survival study yield a causal treatment effect? Lifetime Data Anal 21(4):579–593

Aalen O, Stensrud M, Didelez V, Daniel R, Røysland K, Strohmaier S (2018) Time-dependent mediators in survival analysis: modelling direct and indirect effects with the additive hazards mode. Biometrical J (submitted)

Avin C, Shpitser I, Pearl J (2005) Proceedings of the nineteenth international joint conference on artificial intelligence (IJCAI-05), Edinburgh, Scotland, UK, pp 357–363

Constantinou P, Dawid AP (2017) Extended conditional independence and applications in causal inference. Ann Stat 45(6):2618–2653

Daniel R, De Stavola BL, Cousens SN, Vansteelandt S (2015) Causal mediation analysis with multiple mediators. Biometrics 71(1):1–14

Dawid AP (2000) Causal inference without counterfactuals (with discussion). J Am Stat Assoc 95:407–448

Dawid AP (2002) Influence diagrams for causal modelling and inference. Int Stat Rev 70:161–189

Dawid AP (2015) Statistical causality from a decision-theoretic perspective. Ann Rev Stat Appl 2(1):273–303

Dawid AP, Didelez V (2010) Identifying the consequences of dynamic treatment strategies: a decision-theoretic overview. Stat Surv 4:184–231

Didelez V (2015) Proceedings of the 31st annual conference on uncertainty in artifical intelligence — causality workshop, pp 40–45

Didelez V (2012) Discussion of ‘Experimental designs for identifying causal mechanisms’ by Imai, Tingley, Yamamoto. J R Stat Soc Ser A 176:39

Didelez V (2016) Commentary: Should the analysis of observational data always be preceded by specifying a target experimental trial? Int J Epidemiol 45(6):2049–2051

Didelez V, Dawid A, Geneletti S (2006) Proceedings of the twenty-second annual conference on uncertainty in artificial intelligence. AUAI Press, Arlington, pp 138–146

Fosen J, Ferkingstad E, Borgan O, Aalen O (2006a) Dynamic path analysis—a new approach to analyzing time-dependent covariates. Lifetime Data Anal 12:143–167

Fosen J, Borgan O, Weedon-Fekjaer H, Aalen O (2006b) Dynamic analysis of recurrent event data using the additive hazard model. Biometrical J 48:381–398

Fulcher I, Tchetgen Tchetgen E, Williams P (2017) Mediation analysis for censored survival data under an accelerated failure time model. Epidemiology 28(5):660–666

Geneletti S (2007) Identifying direct and indirect effects in a noncounterfactual framework. J R Stat Soc Ser B 69(2):199–215

Hernán M (2010) The hazards of hazard ratios. Epidemiology 21(1):13–15

Hernán MA, Alonso A, Logan R, Grodstein F, Michels KB, Stampfer MJ, Robins JM (2008) Observational studies analyzed like randomized experiments: an application to postmenopausal hormone therapy and coronary heart disease. Epidemiology 19(6):766–779

Huang YT, Yang HI (2017) Causal mediation analysis of survival outcome with multiple mediators. Epidemiology 28(3):370–378

Lange T, Hansen J (2011) Direct and indirect effects in a survival context. Epidemiology 22(4):575–581

Lin SH, Young J, Logan R, VanderWeele T (2017) Mediation analysis for a survival outcome with time-varying exposures, mediators, and confounders. Stat Med 36(26):4153–4166

Lok J (2016) Defining and estimating causal direct and indirect effects when setting the mediator to specific values is not feasible. Stat Med 35(22):4008–4020

Pearl J (2001) Proceedings of the 7th conference on uncertainty in artificial intelligence (UAI-01). Morgan Kaufmann, Burlington, pp 411–420

Pearl J (2009) Causality, 2nd edn. Cambridge University Press, Cambridge

Pearl J, Robins J (1995) Proceedings of the eleventh annual conference on uncertainty in artificial intelligence (UAI-95). Morgan Kaufmann, San Francisco, pp 444–453

Robins JM, Richardson TS (2011) Alternative graphical causal models and the identification of direct effects. In: Shrout P (ed) Causality and psychopathology: finding the determinants of disorders and their cures. Oxford University Press

Robins JM (1986) A new approach to causal inference in mortality studies with sustained exposure periods—application to control for the healthy worker survivor effect. Math Model 7:1393–1512

Robins JM, Greenland S (1992) Identifiability and exchangeability of direct and indirect effects. Epidemiology 3:143–155

Robins J, Wasserman L (1997) Proceedings of the thirteenth conference annual conference on uncertainty in artificial intelligence (UAI-97). Morgan Kaufmann, San Francisco, pp 409–420

Røysland K, Gran J, Ledergerber B, von Wyl V, Young J, Aalen O (2011) Analyzing direct and indirect effects of treatment using dynamic path analysis applied to data from the swiss hiv cohort study. Stat Med 30:2947–2958

Røysland K, Ryalen P, Nygard M, Lange T, Aalen O, Didelez V (2018) Causal reasoning in survival analysis: re-weighting and local independence graphs (in preparation)

Shpitser I (2018) Identification in graphical causal models. In: Drton M, Lauritzen S, Maathuis M, Wainwright M (eds) Handbook of graphical models (to appear)

Shpitser I, Tchetgen E Tchetgen (2016) Causal inference with a graphical hierarchy of interventions. Ann. Stat. 44(6):2433–2466

Strohmaier S, Røysland K, Hoff R, Borgan Ø, Pedersen T, Aalen O (2015) Dynamic path analysis—a useful tool to investigate mediation processes in clinical survival trials. Stat Med 34(29):3866–3887

Strohmaier S, Haase N, Wetterslev J, Lange T (2016) A simple to implement algorithm for natural direct and indirect effects in survival studies with a repeatedly measured mediator. Tech. Rep. 2, University of Copenhagen, Department of Biostatistics

VanderWeele T (2011) Causal mediation analysis with survival data. Epidemiology 22(4):582–585

VanderWeele T (2015) Explanation in causal inference: methods for mediation and interaction. Oxford University Press, Oxford

VanderWeele TJ, Vansteelandt S, Robins JM (2014) Effect decomposition in the presence of an exposure-induced mediator-outcome confounder. Epidemiology 25(2):300–306

Vansteelandt S, Daniel R (2017) Interventional effects for mediation analysis with multiple mediators. Epidemiology 28(3):258–265

Vansteelandt S, Vandenberghe S, Steen J (2017) Mediation analysis of time-to-event endpoints accounting for repeatedly measured mediators subject to time-varying confounding, Ghent University Technical Report

Zheng W, van der Laan M (2017) Longitudinal mediation analysis with time-varying mediators and exposures, with application to survival outcomes. J Causal Inference 5(2):20160006

## Acknowledgements

I would like to thank Odd Aalen, Rhian Daniel, Ilya Sphitser, Mats Stensrud, Susanne Strohmaier, and Stijn Vansteelandt for helpful discussions.

## Author information

### Affiliations

### Corresponding author

## Additional information

### Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

## Appendix

### Appendix

The following extension results in a greater range of situations where identifiability can be obtained by including possibly time-varying covariates.

Let the data consist of a time-ordered sequence of measurements *A*, *C*(1), *M*(1), *N*(1), *C*(2), *M*(2), *N*(2), \(\ldots \), \(C(\tau ), M(\tau ), N(\tau )\), where *C*(*t*) are sets of covariates. We assume that none of the elements in any *C*(*t*) are graphical children of *A*, or in other words, *A* is assumed to have no direct effect on *C*(*t*). Let further \(C^M(t), C^T(t)\subset C(t)\) such that under \(\text{ do }(A^T, A^M)\)

With analogous reasoning as above, identifiability is then obtained as follows:

The assumptions are illustrated in the graph of Fig. 7. Note that, in fact, \(C^T(t)\) could be in time after *M*(*t*), if it is possible to have measurements between *M*(*t*) and *N*(*t*), similarly to *C* in Fig. 5 (left).

Our assumptions and results are closely related to those of Shpitser (2018). He derives path-specific effects in graphs, first assuming no unobserved variables. This requires the assumption of ‘no recanting witness’ which, in our setting, would be a variable directly affected by both \(A^T\) and \(A^M\)—this would also violate our Assumptions A1.C and A2.C as the conditional independence can then not hold. For graphs with unobserved variables (semi-Markovian graphs), the corresponding criterion is that of ‘no recanting district’, where a district is a sequence of nodes affected by unobserved variables. For instance, in Fig. 5, if *C* was unobserved, then (*N*(1), *M*(2)) would be a district; in fact it would be a ‘recanting district’ because there are directed edges from \(A^T\) as well as from \(A^M\) into nodes of the district. Hence, assumption A1.C and A2.C can be seen as characterising covariates that allow us to avoid recanting districts. In contrast, in Fig. 6, (*N*(1), *N*(2)) is a district as frailty is a latent variable; but it is not recanting because there is no edge from \(A^M\) pointing at any nodes in the district. Similar connections to recanting districts are also noted by Vansteelandt et al. (2017). As Shpitser (2018) covers other forms of identifiability, not only based on the g-formula, his results are of greater generality than ours but not immediately applicable to the survival case as they still rely on nested counterfactuals.

## Rights and permissions

## About this article

### Cite this article

Didelez, V. Defining causal mediation with a longitudinal mediator and a survival outcome.
*Lifetime Data Anal* **25, **593–610 (2019). https://doi.org/10.1007/s10985-018-9449-0

Received:

Accepted:

Published:

Issue Date:

### Keywords

- Causal inference
- Mediation analysis
- Graphical models
- Causal graphs
- Path specific effects