Time After Time: Scalable Effect Estimation for Interventions on When and What to do

ICLR 2025 Conference Submission12267 Authors

Keywords: effect estimation, treatment times, irregular times, sequential decision making

TL;DR: Two scalable effect estimation methods for intervention on time and treatment

Abstract:

Decision support in fields such as healthcare and finance requires reasoning about treatment timing. Artificial Intelligence holds great potential for supporting such decisions by estimating the causal effect of policies such as medication regimens, or resource allocation schedules. However, existing methods for effect estimation are limited in their ability to handle \emph{irregular time}. While treatments and observations in data are often irregularly spaced across the timeline, existing techniques either discretize time, do not scale gracefully to large models, or disregard the effect of treatment time.

We present a solution for effect estimation of sequential treatment times called Earliest Disagreement Q-Evaluation (EDQ). The method is based on Dynamic Programming and is compatible with flexible sequence models, such as transformers. It provides accurate estimates under the assumptions of ignorability, overlap, and no-instantaneous effects. We validate the approach through experiments on a survival time prediction task.

Primary Area: causal reasoning

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2025/AuthorGuide.

Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12267