Mediation analysis in longitudinal intervention studies with an ordinal treatment-dependent confounder
Pith reviewed 2026-05-23 06:20 UTC · model grok-4.3
The pith
Under monotonicity, mediational effects with an ordinal post-treatment confounder are identified up to a stratum-specific sensitivity parameter.
A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.
Core claim
If the intervention always affects the treatment-dependent confounder only in one direction (monotonicity), the mediational effects are identified up to a stratum-specific sensitivity parameter and their empirical non-parametric expressions are derived. The feasibility of the monotonicity assumption can be assessed using empirical data, based on restrictions on the marginal distributions of counterfactuals of the treatment-dependent confounder.
What carries the argument
The monotonicity assumption that the intervention affects the ordinal treatment-dependent confounder in only one direction, which permits identification of direct and indirect effects up to a sensitivity parameter.
If this is right
- Natural direct and indirect effects admit non-parametric expressions that can be estimated from observed data once a value for the sensitivity parameter is supplied.
- The monotonicity assumption can be tested or falsified by checking whether observed data satisfy the implied bounds on counterfactual confounder distributions.
- The method permits quantification of the proportion of an intervention effect on a time-to-event outcome that is mediated through a longitudinal variable such as weight change.
- Defining the outcome as restricted disease-free time and the mediator functionally avoids bias from conditioning on post-treatment variables.
Where Pith is reading between the lines
- The identification strategy may extend to settings with more than three levels of the ordinal confounder or to other types of time-to-event data.
- In practice, the sensitivity parameter could be varied over a range informed by subject-matter knowledge to produce bounds on the mediational effects.
- The functional treatment of the mediator suggests the framework could accommodate mediators observed at irregular time points.
Load-bearing premise
The intervention affects the treatment-dependent confounder only in one direction.
What would settle it
Empirical data in which the intervention increases the ordinal confounder for some units and decreases it for others, or marginal distributions of counterfactual confounder values that violate the restrictions implied by monotonicity.
read the original abstract
In interventional health studies, causal mediation analysis can be employed to investigate mechanisms through which the intervention affects the targeted health outcome. Identifying direct and indirect (i.e. mediated) effects from empirical data become complicated, however, when the mediator-outcome association is confounded by a variable itself affected by the treatment. Here, we investigate identification of mediational effects under such post-treatment confounding in a setting with a longitudinal mediator, time-to-event outcome and a trichotomous ordinal treatment-dependent confounder. If the intervention always affects the treatment-dependent confounder only in one direction (monotonicity), we show that the mediational effects are identified up to a stratum-specific sensitivity parameter and derive their empirical non-parametric expressions. The feasibility of the monotonicity assumption can be assessed using empirical data, based on restrictions on the marginal distributions of counterfactuals of the treatment-dependent confounder. We avoid pitfalls related to post-treatment conditioning by treating the mediator as a functional entity and defining the time-to-event outcome as a restricted disease-free time. In an empirical analysis, we use data from the Finnish Diabetes Prevention Study to assess the extent to which the effect of a lifestyle intervention on avoiding type 2 diabetes is mediated through weight reduction in a high-risk population, with other health-related changes acting as treatment-dependent confounders.
Editorial analysis
A structured set of objections, weighed in public.
Referee Report
Summary. The manuscript develops identification results for natural direct and indirect effects in a longitudinal mediation setting with a time-to-event outcome and a trichotomous ordinal treatment-dependent confounder. Under a monotonicity assumption (the intervention affects the confounder in only one direction), the effects are identified up to a stratum-specific sensitivity parameter; the authors derive the corresponding non-parametric expressions, show how to assess monotonicity feasibility via restrictions on counterfactual marginal distributions, and avoid post-treatment conditioning by defining the mediator as a functional entity and the outcome as restricted disease-free time. The approach is illustrated with data from the Finnish Diabetes Prevention Study examining mediation of a lifestyle intervention on type 2 diabetes incidence through weight reduction.
Significance. If the derivations hold, the work supplies a practical, assumption-conditional identification strategy for a common but difficult configuration in health-intervention studies: longitudinal mediators, survival outcomes, and ordinal post-treatment confounding. Explicitly conditioning on the sensitivity parameter and providing an empirical check for monotonicity are strengths that support sensitivity analyses rather than point identification. The Finnish Diabetes Prevention Study application demonstrates relevance to a high-risk population and shows how the functional-mediator and restricted-time definitions sidestep standard pitfalls.
minor comments (2)
- [Abstract] Abstract: the phrase 'stratum-specific sensitivity parameter' is introduced without indicating its range or how it enters the identifying expressions; adding one sentence on its interpretation would improve accessibility.
- The manuscript would benefit from an explicit statement, early in the methods, of how the functional definition of the mediator (as opposed to a standard time-varying covariate) alters the g-computation or weighting formulas relative to existing longitudinal mediation literature.
Simulated Author's Rebuttal
We thank the referee for the positive evaluation of our manuscript and the recommendation for minor revision. No specific major comments were provided in the report, so we have no points requiring point-by-point rebuttal or revision at this stage. We are prepared to address any minor comments that may arise in a subsequent round.
Circularity Check
No significant circularity
full rationale
The paper derives non-parametric expressions for mediational effects conditional on the monotonicity assumption for the ordinal confounder, with identification holding only up to an explicit stratum-specific sensitivity parameter. This is a standard sensitivity analysis setup rather than a reduction of the result to its inputs by construction. Monotonicity feasibility is assessed via restrictions on counterfactual marginal distributions from empirical data, and the functional mediator definition plus restricted outcome avoids post-treatment conditioning by design. No self-definitional steps, fitted inputs renamed as predictions, or load-bearing self-citations appear in the derivation chain; the central result remains independent of the observed data distributions once the assumption and parameter are stated.
Axiom & Free-Parameter Ledger
free parameters (1)
- stratum-specific sensitivity parameter
axioms (1)
- domain assumption Monotonicity: the intervention always affects the treatment-dependent confounder only in one direction.
discussion (0)
Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.