Inaccurate pharmaceutical demand forecasting undermines hospital inventory management, causing drug shortages, overstocking, and avoidable logistics costs. Classical “World Healthcare Organization” methods, particularly the consumption method, rely solely on historical consumption data and cannot account for the operational context driving demand. This paper presents a machine learning-based causal forecasting framework applied to monthly medicine consumption data from five French hospitals. Using twelve operational features, including patient volumes, bed capacity, and staff levels, Random Forest and Artificial Neural Network models are trained and tested on ten medicines. Random Forest achieves a mean R² of 0.81 on the test set and substantially outperforms the WHO consumption method across all medicines. These results demonstrate the operational feasibility and accuracy gain of machine learning-based causal forecasting for hospital pharmaceutical supply chain management.