CafeMed: Causal Attention Fusion Enhanced Medication Recommendation

TL;DR

CafeMed tackles the challenge of personalized, safe medication recommendation by integrating dynamic causal reasoning with cross-modal attention. It introduces the Causal Weight Generator (CWG) to adapt causal influences to patient context and the Channel Harmonized Attention Refinement Module (CHARM) to capture interdependencies across diagnoses, procedures, and medications. Empirical results on MIMIC-III and MIMIC-IV demonstrate superior accuracy and reduced drug–drug interactions compared with state-of-the-art baselines, with ablations confirming the complementary value of CWG and CHARM. The approach offers a clinically relevant, scalable pathway to safer, more personalized pharmacotherapy using electronic health records.

Abstract

Medication recommendation systems play a crucial role in assisting clinicians with personalized treatment decisions. While existing approaches have made significant progress in learning medication representations, they suffer from two fundamental limitations: (i) treating medical entities as independent features without modeling their synergistic effects on medication selection; (ii) employing static causal relationships that fail to adapt to patient-specific contexts and health states. To address these challenges, we propose CafeMed, a framework that integrates dynamic causal reasoning with cross-modal attention for safe and accurate medication recommendation. CafeMed introduces two key components: the Causal Weight Generator (CWG) that transforms static causal effects into dynamic modulation weights based on individual patient states, and the Channel Harmonized Attention Refinement Module (CHARM) that captures complex interdependencies between diagnoses and procedures. This design enables CafeMed to model how different medical conditions jointly influence treatment decisions while maintaining medication safety constraints. Extensive experiments on MIMIC-III and MIMIC-IV datasets demonstrate that CafeMed significantly outperforms state-of-the-art baselines, achieving superior accuracy in medication prediction while maintaining the lower drug--drug interaction rates. Our results indicate that incorporating dynamic causal relationships and cross-modal synergies leads to more clinically-aligned and personalized medication recommendations. Our code is released publicly at https://github.com/rkl71/CafeMed.

Figures (2)

• Figure 1: Comparison of causal modeling approaches in medication recommendation. Prior methods (a) use independent processing pipelines where diagnosis and procedure information are encoded separately and merged only at the final prediction stage, potentially ignoring critical cross-modal interactions. Our proposed interactive processing framework (b) introduces early-stage feature interaction between diagnosis and procedure representations, enabling the model to capture synergistic effects and medical contraindications, resulting in more accurate medication recommendations.
• Figure 2: The overall architecture of the proposed CafeMed, comprising two key innovative components: (i) CWG that dynamically transforms causal effects into adaptive modulation weights for entity representations. (ii) The CHARM module that enables efficient cross-modal feature fusion through channel attention, channel shuffle, and spatial attention mechanisms.