Admissible Diffusion for Multimodal Interventional Trajectories

2026-09-28

AI Summary

This study addresses the lack of causal interventional prediction capability in clinical trajectory generation by proposing the ADMIT framework. This framework integrates irregular multimodal representations with a treatment-conditioned latent diffusion model, innovatively translating medical prior knowledge into explicit physiological constraints. Through state projection and action gating mechanisms, it distinguishes causal hypotheses from constraint satisfaction to generate intervention trajectories aligned with clinical logic. Experimental results demonstrate that the proposed method significantly improves hidden state recovery quality and reduces treatment contrast error. Furthermore, simulation experiments show its effectiveness in predicting tumor volume changes, thereby enhancing the credibility of counterfactual intervention predictions.

Abstract

Generating a plausible clinical trajectory does not establish what would happen under a different treatment. We present ADMIT, a framework combining irregular multimodal representations, treatment-conditioned latent diffusion and explicit constraints on generated states or actions. We formulate its interventional target through sequential g-computation and distinguish causal assumptions from constraint satisfaction. Its admissibility mechanism translates physiological prior knowledge into explicit constraints on generated states and proposed actions. Treatment-exposure dynamics condition latent transitions, while state projection or action gating applies the constraints during rollout so that they influence subsequent trajectory generation. In our preliminary experiments, multimodal inputs improved supervised hidden-state recovery and reduced treatment-contrast error. In a simulated dosing-schedule experiment with leak-free history encoding, ADMIT predicted most of the tumor-volume change caused by redistributing a fixed total dose. An exposure input improved these predictions around a temporary dose reduction whether or not the assumed clearance rate was correct, but reduced the predicted size of a dose effect, and a deterministic recurrent baseline matched ADMIT's average predictions. Exposure projection reduced constraint violations, although enforcement remained incomplete. Semi-synthetic experiments using eICU context illustrated treatment-response generation under fixed and adaptive policies. Observational examples further characterize model treatment sensitivity. ADMIT provides a framework for testing whether complementary observations and physiological restrictions improve intervention trajectories, with representation recovery, effect accuracy and rule enforcement assessed separately.

Problem

• interventional trajectories
• multimodal clinical data
• causal inference
• treatment effect prediction
• physiological constraints

Innovation

• latent diffusion
• causal inference
• multimodal trajectories
• physiological constraints
• g-computation