Causal Diffusion Models for Counterfactual Outcome Distributions in Longitudinal Data

Farbod Alinezhad, Jianfei Cao, Gary J. Young, Brady Post|April 14, 2026arXiv

Key Takeaway

CDM achieves 15-30% better accuracy at capturing outcome distributions in sequential treatment scenarios by using diffusion models instead of traditional causal inference methods, making it practical for medical decision support.

Summary

This paper introduces Causal Diffusion Models (CDM), a new method for predicting what would happen under different treatment sequences in medical data over time.

reasoning evaluation safety

Key Terms

counterfactual-reasoning time-dependent-confounding diffusion-models wasserstein-distance inverse-probability-weighting