Knowledge graph-enhanced deep learning for pharmaceutical demand forecasting

Accurate pharmaceutical demand forecasting is essential to ensure timely drug availability, reduce inventory costs, and improve operational efficiency in healthcare supply chains. However, existing statistical, machine learning, and deep learning approaches often struggle to capture the nonlinear and dynamic demand patterns arising from drug substitutions, comorbidity treatments, and seasonal disease fluctuations. To address this challenge, we propose KG-GCN-LSTM, a novel hybrid model that integrates a pharmaceutical knowledge graph (KG) with deep learning techniques. A clipped Graph Convolutional Network (GCN) is employed to extract feature representations from both the historical demand of the target drug and the related drugs encoded in the knowledge graph. The outputs of the GCN are subsequently processed by a Long Short-Term Memory (LSTM) network to capture temporal dynamics in drug demand. Experiments on real-world pharmacy sales data demonstrate that KG-GCN-LSTM consistently outperforms established benchmarks-including ARIMA, SVR, XGBoost, RNN, CNN-LSTM, TimeMixer and NBEATS, achieving a 3.62% reduction in Symmetric Mean Absolute Percentage Error (SMAPE) relative to NBEATS, while delivering performance comparable to the state-of-the-art TimeMixer. These results highlight the effectiveness of knowledge graph-enhanced deep learning in improving the accuracy and robustness of pharmaceutical demand forecasting, which can support data-driven decision-making in healthcare supply chain management.