Effective drug inventory management is crucial for maintaining service quality and cost efficiency in hospitals. Inaccurate procurement planning can cause stockouts or overstock conditions, disrupting healthcare operations. This study presents a predictive model for outpatient drug consumption using a Multivariate Long Short-Term Memory (LSTM) network. The dataset comprises historical records from the general, pediatric, and maternity polyclinics at RSIA Fatimah Hospital, Probolinggo Regency, East Java, Indonesia, collected in January 2023. The variables include timestamp, polyclinic name, drug name, and quantity used. Data preprocessing involved cleaning, one-hot encoding for categorical features, min-max normalization, and time-based train-test splitting to avoid data leakage. The multivariate LSTM model was trained for 500 epochs under various configurations, evaluated using Mean Absolute Error (MAE) and Mean Absolute Percentage Error (MAPE). Three model groups (A, B, C) with distinct neuron counts and batch sizes were tested to assess performance variations. Model B1 achieved the best results, with the lowest MAE (10.239), MAPE (1.979%), and highest R² (0.199). Although the R² value indicates limited variance explanation, Nonetheless, the model remains useful for operational forecasting, the model effectively captures temporal patterns in drug consumption, demonstrating its potential as a decision-support tool for optimizing hospital pharmaceutical inventory management.