Test-Time Data Quality Degradation in Clinical ML: A Systematic Robustness Analysis on MIMIC-IV
Studies in Health Technology and Informatics · 21.05.2026
DOI: 10.3233/shti260137 ↗Kurz gefasst
Fachartikel in der Zeitschrift Studies in Health Technology and Informatics, erschienen am 21.05.2026. Verfasst von Nour Idris Pacha und 4 weiteren Autor:innen. Beteiligte Einrichtung: Angers University Hospital, Angers, France. Hauptthemen laut MEDLINE: Machine Learning, Data Accuracy, Hospital Mortality, Electronic Health Records. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).
Abstract (Originalsprache)
We conducted a controlled experimental study using the MIMIC-IV database to evaluate the robustness of five clinical classification models under data quality (DQ) degradation. Dimension-specific corruptions were applied exclusively at test time, while models were trained on clean data. Performance (AUROC and Brier score) declined with increasing degradation, with stronger effects observed for completeness, coherence, and validity than for precision. These findings indicate that DQ degradation can substantially compromise the reliability of ML-based predictions of in-hospital mortality.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Angers University Hospital, Angers, France.
- Enovacom, Orange Business, France.
- MINT, INSERM UMR 1066 / CNRS UMR 6021, University of Angers, Angers, France.
Zitieren
APA
Idris Pacha, N., Aloui, S., Rabaoui, A., Escudie, A., & Riou, J. (2026). Test-Time Data Quality Degradation in Clinical ML: A Systematic Robustness Analysis on MIMIC-IV. Studies in Health Technology and Informatics. https://doi.org/10.3233/shti260137
Vancouver
Idris Pacha N, Aloui S, Rabaoui A, Escudie A, Riou J. Test-Time Data Quality Degradation in Clinical ML: A Systematic Robustness Analysis on MIMIC-IV. Studies in Health Technology and Informatics. 2026. doi:10.3233/shti260137
BibTeX
@article{IdrisPacha2026,
author = {Idris Pacha, Nour and Aloui, Saber and Rabaoui, Asma and Escudie, Antony and Riou, Jérémie},
title = {Test-Time Data Quality Degradation in Clinical ML: A Systematic Robustness Analysis on MIMIC-IV},
journal = {Studies in Health Technology and Informatics},
year = {2026},
doi = {10.3233/shti260137}
}
EUVIMED