Causal inference for calibrated scaling interventions on time-to-event processes
By: Helene Charlotte Wiese Rytgaard, Mark van der Laan
Potential Business Impact:
Changes how fast things happen to reach goals.
This work studies stochastic interventions in continuous-time event-history settings formulated as multiplicative scalings of the observed intensity governing an intermediate event process. This gives rise to a family of causal estimands indexed by a scalar parameter {\alpha}, which changes the event rate while preserving the temporal and covariate structure of the data-generating process. We introduce calibrated interventions, where \(\alpha\) is chosen to achieve a pre-specified goal, such as a desired level of cumulative risk of the intermediate event, and define corresponding composite target parameters capturing the resulting effects on the outcome process. Our proposal enables practical yet statistically principled intervention analysis in survival and longitudinal settings, which offers a flexible alternative to deterministic or static interventions that are often ill-defined. The framework applies broadly to causal questions involving time-to-event treatments or mediators, and offers a pragmatic analogue to indirect/direct effect decompositions. We present the efficient influence curves for various versions of target parameters under a nonparametric statistical model, discuss their double robustness properties, and propose an estimation procedure based on targeted maximum likelihood estimation (TMLE). The proposed estimands are illustrated through examples of event-history scenarios addressing distinct causal questions.
Similar Papers
Computationally and statistically efficient estimation of time-smoothed counterfactual curves
Methodology
Finds how treatments affect health over time.
Interpolated stochastic interventions based on propensity scores, target policies and treatment-specific costs
Methodology
Helps plan experiments with limited money.
On robust Bayesian causal inference
Methodology
Finds true causes even with missing information.