OverdoseMoE: Multi-Expert Framework for Opioid Overdose Risk
October 1, 2026
The OverdoseMoE framework uses a multi-expert architecture to predict 180-day opioid overdose risk from longitudinal ICD histories. By integrating different model scales via expert-weighting, the system achieved an AUROC of 69.49 and an AUPRC of 25.17, outperforming general-purpose language model baselines.
HOW THIS AFFECTS YOU
●
researcherThe multi-expert weighting strategy for longitudinal medical sequences is a viable architecture for clinical tasks.
●
healthThis provides a more specialized clinical decision support tool for high-risk patient identification.