Covariate-informed continuous-time gray-box modeling to identify responsiveness of post-surgical pain to opioid therapy

Ran Liu, Rodrigo Gutierrez, Rory Vu Mather, Edward A Bittner, Patrick L. Purdon

Keywords: state space model, gray box, hybrid model, time series, treatment effects

TL;DR: We developed a continuous-time state space method combining mechanistic PK/PD with black box prediction to identify surgical patients' responsiveness to opioids, given preoperative covariates, pain scores, and opioid administration over time.

Abstract:

Quantifying responsiveness of pain to opioid administration is a clinically important, yet technically challenging problem.
Pain is a subjective phenomenon that is difficult to assess by means other than infrequent and low-resolution patient self-reporting.
We tackle this problem using a continuous-time state space modeling approach that incorporates mechanistic models of opioid effect site concentration as well as information from covariates using black-box models iteratively trained to predict the distributions of partially observed variables.
We evaluated our method in simulation, and applied it in a real-world observational study of 21,652 surgical cases, where our method is able to recapitulate the known potencies of different opioids, and stratify patients by pain and opioid use related outcomes.

Primary Area: learning on time series and dynamical systems

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.

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: 3994