On this page you can find a selection of exemplary projects of the individual research groups.

Research Group „Data-driven Healthcare“

1. National Quality Monitoring
   Across several large projects funded by the Swiss Innovation Agency (Innosuisse), we developed comprehensive quality indicator sets—including mortality, readmission, and complication rates—and an innovative quality monitoring software. This software is now used in all Swiss hospitals. It enables statistical provider comparisons, advanced filtering and drill-down functionalities, and a novel case-finding methodology, which compares machine learning–based patient outcome predictions with actual retrospective outcomes to identify quality-related discrepancies (see news item).
2. Wearables for Predicting Health Outcomes
   A newer research direction of ours explores the use of wearable technologies to predict health outcomes, risks, and trajectories. For instance, we generated an early warning system for hypoglycemia in people with diabetes, using only data from a commercially available smartwatch (see publication).

Research Group „Machine Learning in Healthcare“

1. Interpretable Histopathology Image Analysis
   In collaboration with the University of Tübingen and Radboud University Medical Center, we developed an explainable AI system for histopathological image analysis. While existing AI tools could highlight suspicious regions, they offered no insight into their reasoning. Our system addressed this by generating natural language explanations, such as identifying “necrotic areas” or “enlarged nuclei” as reasons for concern. This allowed pathologists to understand and question the AI’s decisions, enabling a more interactive and transparent diagnostic process (see publication #1 | publication #2).
2. Decision Deferral for Trustworthy AI in Medical Diagnosis
   In this project, we are developing strategies that allow ML systems in healthcare to decide when to make a prediction and when to defer to a human expert. This is essential for ensuring patient safety in uncertain or high-risk situations. Our research focuses on ophthalmology, where we study these deferral strategies using large datasets of eye images to support more effective human-AI collaboration in diagnosis.

Research Group „Functioning Epidemiology“

1. Functioning Indicator
   The Functioning Epidemiology group is leading the development of a functioning indicator to complement traditional health metrics such as mortality and morbidity. This indicator aims to capture individuals’ lived experience with health conditions by quantifying levels of functioning across physical, mental, and social domains. Drawing on large-scale real-world data, including electronic health records and rehabilitation datasets, we apply advanced statistical and machine learning methods to model functioning trajectories and transitions. The goal is to create a scalable, clinically meaningful, and policy-relevant indicator that can be used across health system levels, to support individualized care planning, inform clinical decision support tools, and monitor population-level trends in disability and recovery. This work supports a broader shift toward more inclusive, person-centered health systems.
2. Return to Play Prediction
   The Functioning Epidemiology group is developing machine learning models to predict return-to-play outcomes in athletic populations. By analyzing real-world data on injuries, rehabilitation, and performance, we aim to identify key prognostic factors influencing recovery duration and reinjury risk. These models support personalized return-to-play strategies, balancing performance goals with health and long-term functioning. Our work contributes to safer, data-driven decision-making in sports rehabilitation, with applications across youth, professional, and elite athletes.