arXiv · 2611.11847
Severity-Controlled Prediction Sets for Medication Recommendation
Abstract
Medication recommendation is a multi-label prediction task. However, existing prediction-set evaluation and calibration methods rely mainly on exact-overlap metrics and therefore treat all missed medications as equally severe, overlooking how closely the prediction set matches them within the medication hierarchy. To address this, we introduce Conformal Severity Control (CSC), a post-hoc framework for calibrating medication prediction sets under hierarchy-aware mismatch severity. CSC uses the Anatomical Therapeutic Chemical (ATC) hierarchy to quantify mismatch severity and controls patient-level severity without retraining the predictor, supporting both average severity and upper-quantile severity for the highest-severity mismatches. Experiments on MIMIC-IV show that CSC tracks user-specified severity tolerances and shifts prediction sets away from distant mismatches toward exact or close matches.
Explore related subjects
Keep this discovery
Explore connections, maps & timelines
Yu Gu, Zijun Yu, Chi-Kuang Yeh, Xinyu Wang, Ziyang Song. 2026-11-09. Severity-Controlled Prediction Sets for Medication Recommendation. https://arxiv.science/abs/2611.11847
Citations retain the original authors and DOI and include this site’s record URL. If this discovery service contributed to your research, you can also cite arXiv Science by Indiciae: BibTeX · RIS.