ClassShift-Clinical: Benchmarking Clinical Prediction Models under Class Imbalance and Distribution Shift
International Journal of Computational and Biological Sciences · 10.04.2026 · 3 (1) · 312-321
DOI: 10.66238/ijcbs74 ↗Kurz gefasst
Fachartikel in der Zeitschrift International Journal of Computational and Biological Sciences, online erschienen am 10.04.2026. Verfasst von Amelia Fischer und 2 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0).
Abstract (Originalsprache)
The deployment of machine learning in healthcare settings has revealed a critical vulnerability in clinical prediction models when they are transitioned from controlled training environments to real world clinical workflows. Two of the most pervasive challenges in this domain are severe class imbalance, where the pathology or outcome of interest is rare compared to the control group, and distribution shift, where the statistical properties of the target population change over time or across different clinical sites. This paper introduces a comprehensive framework for systematically evaluating these twin challenges. Through extensive theoretical analysis and empirical benchmarking, we investigate how various predictive architectures and mitigation strategies perform when subjected to simultaneous demographic, temporal, and site specific shifts while operating under extreme class imbalance ratios. Our analysis utilizes large scale electronic health record datasets to simulate realistic deployment scenarios, revealing that models which achieve high performance on static, balanced datasets often suffer catastrophic degradation under dynamic clinical conditions. We propose a standardized set of evaluation protocols that prioritize clinically relevant metrics over traditional accuracy measures. Ultimately, this research provides a vital foundation for the development of robust, reliable, and equitable clinical decision support systems that can maintain their predictive integrity in the face of inevitable environmental variations.
Zitieren
APA
Fischer, A., Romano, A., & Chamberlain, T. (2026). ClassShift-Clinical: Benchmarking Clinical Prediction Models under Class Imbalance and Distribution Shift. International Journal of Computational and Biological Sciences, 3(1), 312-321. https://doi.org/10.66238/ijcbs74
Vancouver
Fischer A, Romano A, Chamberlain T. ClassShift-Clinical: Benchmarking Clinical Prediction Models under Class Imbalance and Distribution Shift. International Journal of Computational and Biological Sciences. 2026;3(1):312-321. doi:10.66238/ijcbs74
BibTeX
@article{Fischer2026,
author = {Fischer, Amelia and Romano, Amelia and Chamberlain, Theodore},
title = {ClassShift-Clinical: Benchmarking Clinical Prediction Models under Class Imbalance and Distribution Shift},
journal = {International Journal of Computational and Biological Sciences},
year = {2026},
volume = {3},
number = {1},
pages = {312--321},
doi = {10.66238/ijcbs74}
}
EUVIMED