Target-Driven Stochastic Interventions under Treatment-Specific Costs

We introduce a new class of stochastic interventions derived from cost‐penalized Kullback-Leibler projections of the independent product between the organic propensity score and a prespecified target distribution. The induced marginals generalize incremental propensity score interventions (IPIs) accommodating arbitrary cost structures and targets. We show that these interventions arise as limiting cases of relaxed optimal transport problems and derive the semiparametric efficient influence functions for the associated expected outcomes. We present von Mises one‐step estimators that remain robust to model misspecification. Simulations and a clinical application demonstrate their practical benefits.