Towards Scalable and Stable Machine Learning in Clinical Contexts: Addressing Computational Efficiency and Dataset Shift