Personalized Two-sided Dose Interval

Abstract: In fields such as medicine and social sciences, the goal of treatment is often to maintain the outcome of interest within a desirable range rather than to optimize its value. To achieve this, it may be more practical to recommend a treatment dose interval rather than a single fixed level for a study unit. Since individuals may respond differently to the same treatment level, the recommended dose interval should be personalized based on their unique characteristics. Existing methods for one-sided dose intervals and iteratively constructed two-sided intervals provide useful foundations, but their theory does not directly address simultaneous estimation over unrestricted product function spaces. To address this gap, we propose a direct method for learning personalized two-sided dose intervals based on empirical risk minimization with a doubly-robust loss function that is well-defined over a tensor product function space. This formulation enables simultaneous estimation of the lower and upper bounds without constrained alternating updates. We establish statistical properties of the estimated dose interval in terms of excess risk by leveraging reproducing kernel Hilbert space theory. Our simulation study and a real-world application in warfarin dosing show that the proposed direct method compares favorably with competing indirect regression-based methods.

Submission Type: Long submission (more than 12 pages of main content)

Previous TMLR Submission Url: https://openreview.net/forum?id=N39gNutM3s

Changes Since Last Submission: Per the Editor-in-Chief’s comments, we have revised the format and resubmitted the paper.

Assigned Action Editor: ~Kejun_Huang1

Submission Number: 8607