Predictive Process Monitoring aims at forecasting how ongoing process executions will unfold. Recent advancements have shown the advantages of leveraging contextual and inter-case features for increasing the accuracy of the prediction. However, existing methodologies are based on feature engineering. This study introduces the novel concept of a Status Graph, a comprehensive graph representation of the status of all ongoing executions, incorporating all active cases and their interdependencies. By leveraging graph neural networks in predictive process monitoring tasks, the proposed approach eliminates the need for feature engineering, enabling the model to learn directly from graph structures. A proof-of-concept implementation shows improvements for the next activity and remaining completion time prediction tasks.