Temporal adaptive transfer learning for clinical language models
Findings suggest that drift-triggered temporal and institutional adaptation can improve the robustness and stability of clinical language models under the distribution shifts represented in the evaluated datasets, but further prospective and independent external validation is required before the framework can be consid...