CIDGMed: Causal Inference-Driven Medication Recommendation with Enhanced Dual-Granularity Learning

Shunpan Liang, Xiang Li, Shi Mu, Chen Li, Yu Lei, Yulei Hou, Tengfei Ma

Abstract: Highlights•Causal inference reveals true links between diseases and medications.•Integrating entity-level and molecular-level data enhances model learning.•A post-processing bias correction module refines recommendations, enhancing predictive performance.•CIDGMed improves accuracy by 2.54%, reduces side effects by 3.65%, and boosts efficiency by 39.42%.•A detailed case study using real medical records demonstrates the model’s effectiveness.