Research

Noel’s research interests focus on developing and applying machine learning methods that yield models with meaningful capacity for data-driven medical insight.

Fundamentally believing in the synergy of integrating interdisciplinary perspectives, he has already contributed to a range of both machine learning and medical research projects, including poster presentations, conference abstracts, and peer-reviewed journal publications. Notably, he also leads an acclaimed nonprofit initiative, empowering the next generation of researchers and establishing a culture of open, fair, and critical scientific practice.

Interests

Highlights¹

In medicine, routinely collected data are invaluable for real-world evidence studies. However, these data are burdened by data quality issues. In his bachelor’s thesis at the Charité Berlin and Technical University of Berlin (2024), Noel developed a proof-of-concept electronic health record (EHR) data processing pipeline. The objective was to streamline the use of quality-assessed EHR data for real-world evidence studies and to build the foundation for the Charité Outcomes Research Repository (CORR). As a demonstration, established clinical risk scores were derived and extensively validated. This work has been awarded the Rolf Niedermeier Faculty Prize for outstanding theses.

A primary line of his work involves exploring the broad potential of deep learning models in medicine.

Domain-specific EHR foundation models have shown promising results in predictive accuracy and generalization. However, their development is constrained by limited access to adequate datasets. In Hegselmann et al. (2025), general-purpose large language models (LLMs) were evaluated for encoding EHR data in clinical prediction tasks, demonstrating performance on par with or superior to specialized models. In a separate study (von Wedel et al., 2024), the team investigated affiliation bias in LLM-assisted peer review of medical abstracts.

A secondary line of work focuses on medical outcomes research in intensive care unit (ICU) patients.

Comorbidity risk scores are widely used in clinical and research settings. However, many of them are additive and fail to capture the complexity of multimorbidity (i.e., the presence of two or more chronic comorbidities) and its impact on ICU mortality. Noel co-led a study (Kronenberg et al., 2024) that identified multimorbidity clusters in ICU patients using latent class analysis (LCA). These clusters were independently associated with mortality and enhanced the predictive performance of a widely used clinical risk score. He presented the work as a poster at the Congress of the German Interdisciplinary Association for Intensive Care and Emergency Medicine (DIVI).

Noel has also contributed to a diversity of other works investigating medical outcomes in supporting roles, assisting with data processing, analysis, and visualization. For a complete list, please reference Google Scholar.

Supervisors

Noel has contributed to various research projects, working closely with: