DFNet: Dual-Decision Fusion Network for Drug Combination Prediction

In recent years, the application of deep learning methods in the health care sector has achieved significant success. In particular, drug recommendations based on electronic health records have enabled the generation of personalized drug prediction combinations tailored to patients' health conditions. However, existing drug recommendation methods still have shortcomings in effectively integrating drug knowledge using patient health representations. Therefore, we propose a dual-decision fusion network (DFNet) for drug combination prediction. Specifically, DFNet first extracts patient features by capturing the connections between medical events in patient medication records to obtain patient representations. Subsequently, we design a dual-decision fusion network that computes the similarity between patient representations and drug representations to achieve recommendations. Moreover, we incorporate drug relationship knowledge into the recommendation process. Finally, we fused the two recommendation results to propose a safe and effective drug recommendation combination. Extensive experiments on the MIMIC-III dataset validate the effectiveness of the proposed method.