Hybrid GCN-LSTM for Drug Demand Prediction Using Multi-Regional IQVIA Dataset

Abstract

Drug availability is a crucial aspect in the healthcare system that requires accurate demand management to avoid the risk of stock-outs and overstocks. This study proposes a Hybrid Graph Convolutional Network–Long Short-Term Memory (GCN-LSTM) approach to predict drug demand using the IQVIA Pharmaceutical Sales dataset consisting of 960 weekly observations from eight drug therapy categories during the period 2014–2019. The novelty of this study lies in the integration of spatial information based on graph correlation and temporal learning in a single prediction architecture to capture the relationship between drug categories and demand dynamics over time. The data is processed through cleaning, Min-Max Scaling normalization, and sliding window stages before being formed into a graph based on Pearson correlation between drug categories. The GCN-LSTM model is then compared with ARIMA, Random Forest, and a single LSTM using RMSE, MAE, and MAPE metrics. The experimental results show that the GCN-LSTM model produces the best performance with RMSE 3.821; MAE 2.741; and MAPE 7.86%, better than all comparison models. These results demonstrate that the integration of spatial and temporal information can significantly improve the accuracy of drug demand predictions. The proposed approach has the potential to support decision-making in pharmaceutical supply chain management, inventory planning, and drug distribution optimization.