1: \begin{abstract}

2: Interventional effects have been proposed as a solution to the

3: unidentifiability of natural (in)direct effects under

4: mediator-outcome confounders affected by the exposure. Such

5: confounders are an intrinsic characteristic of studies with

6: time-varying exposures and mediators, yet the generalization of the

7: interventional effect framework to the time-varying case has

8: received little attention in the literature. We present an

9: identification result for interventional effects in a general

10: longitudinal data structure that allows flexibility in the

11: specification of treatment-outcome, treatment-mediator, and

12: mediator-outcome relationships. Identification is achieved under the

13: standard no-unmeasured-confounders and positivity assumptions. We

14: also present a theoretical and computational study of the properties

15: of the identifying functional based on the efficient influence

16: function (EIF). We use the EIF to propose a sequential regression

17: estimation algorithm that yields doubly robust,

18: $\sqrt{n}$-consistent, asymptotically Gaussian, and efficient

19: estimators under slow convergence rates for the regression

20: algorithms used. This allows the use of flexible machine learning

21: for regression while permitting uncertainty quantification through

22: confidence intervals and p-values. A free and open source \texttt{R}

23: package implementing our proposed estimators is made available on

24: GitHub. We apply the proposed estimator to an application from a

25: comparative effectiveness trial of two medications for opioid use

26: disorder. In the application, we estimate the extent to which

27: differences between the two treatments' on subsequent risk of opioid

28: use is mediated by craving symptoms.

29: \end{abstract}

30: