Precise and safe drug recommendation remains a critical challenge in clinical decision support, particularly in polypharmacy, where dru-drug interactions (DDIs) can lead to severe adverse outcomes. Existing deep learning approaches primarily rely on structural or sequence-based drug representations, which often fail to capture the physicochemical mechanisms underlying DDIs. In this study, we propose QUantum-informed Analysis for Recommendation of Kombinations (QUARK), a multimodal framework that integrates quantum-derived molecular features with longitudinal patient data for DDI-aware drug recommendation. QUARK leverages three-dimensional electron density and electrostatic potential maps, obtained from density functional theory (DFT) calculations, to capture chemical reactivity beyond atomic connectivity. These quantum-informed features are fused with patient embeddings derived from electronic health records (EHRs) through a cross-attention mechanism, enabling the model to jointly capture chemical reactivity and clinical context. Experimental results on the MIMIC-III dataset demonstrate that QUARK outperforms prior methods in both recommendation accuracy and DDI reduction. These findings highlight the potential of incorporating quantum-level molecular information to enhance the safety and reliability of clinical decision support systems.