New publication in Healthcare Management Science: Generalisable machine learning models in healthcare

A new article has been published in Healthcare Management Science as a result of collaboration between the Chair of Industrial Management, the Chair of Intelligent Systems and Services (TUD), the University of Regensburg and Dresden University Hospital. With contributions from Lasse Bohlen, Daniel Zähringer and Patrick Zschech, the study uses data from European and US healthcare systems to investigate how well machine learning models for predicting patient outcomes in intensive care units (mortality, length of stay, readmission) perform when transferred between different hospitals.

The findings provide an important insight: simpler, inherently interpretable models can often be transferred more robustly to different healthcare facilities than complex ‘black-box’ approaches, whilst at the same time requiring less local data. For various prognostic tasks in the intensive care unit, increased model complexity did not necessarily lead to better performance across different hospitals.
Furthermore, an interpretability analysis provides hospital managers with practical tools to identify which clinical features may cause external models to fail in their specific hospital setting, thereby enabling more reliable model validation prior to implementation.

Reference:
Bohlen, Lasse, et al. "Toward generalizable and interpretable machine learning models in healthcare: Insights from ICU outcome predictions." Health Care Management Science 29.2 (2026): 23. https://doi.org/10.1007/s10729-026-09760-y