Portfolio

Hospital Resource Optimization with Causal ML and Reinforcement Learning
Hospital Operations | Causal Machine Learning | Reinforcement Learning | Resource Allocation
Designed and implemented causal machine learning and offline reinforcement learning algorithms to optimize hospital decision making and patient flow. Projects included optimizing surgery scheduling, patient mobilization, normal care unit allocation, and intensive care discharge decisions. Leveraged high-dimensional patient-level and operational data from German and Swiss hospitals to model bottlenecks and predict demand surges. Developed decision support tools in Python and R, integrating advanced ML frameworks (TensorFlow, PyTorch) and collaborating closely with hospital IT and management. These solutions demonstrate the transferability of advanced analytics and optimization methods to real-world healthcare operations and other resource-constrained industries.
Skills: Python, R, TensorFlow, PyTorch, CUDA, SQL
Project partners: Fedaral Office of Statistics, Cantonal Hospital St. Gallen, PromQuality Consortium
Related papers:

• Causal Effects and Optimal Policy Learning for Intensive Care Unit Discharge Decisions to Solve Hospital Process Bottlenecks: Approach, Methods, and First Results (2024)
• Is rapid recovery always the best recovery? - Developing a machine learning approach for optimal assignment rules under capacity constraints for knee replacement patients (2023)

Unlabeled Data Learning for Improving Health Administrative Data Quality
Health Insurance | Positive-Unlabeled Learning | Data Quality | Claims Data
Developed and implemented positive-unlabeled learning and unsupervised learning methods to enhance the quality of health administrative data. Addressed the challenge of diagnoses and treatments being only positively labeled, leading to substantial missingness for certain conditions and procedures. Built scalable data pipelines in Python to identify and correct label gaps, improving downstream analytics and reporting for insurers and healthcare providers.
Skills: Python, PyTorch, scikit-learn, SQL
Project partners: Fedaral Office of Statistics
Related papers:

• Application of positive and unlabeled learning: A novel approach for identifying sepsis cases from hospital administrative data (2024)

Representation Learning of Claims Items
Health Economics | Deep Learning | Representation Learning | Claims Data | Feature Embedding
Engineered and trained deep learning models to generate dense vector representations of medical claims items, enabling improved feature extraction for predictive modeling and policy analysis. Leveraged large, heterogeneous claims datasets from Swiss insurers, applying PyTorch and TensorFlow for model development. Used NLP-inspired methods (word2vec, transformer-based models) to build representations for over 70,000 unique claims items, and evaluated embeddings with boosting methods (XGBoost, WarpGBM for CUDA, LSTM). This approach supports transferability to other domains requiring structured data representation and advanced analytics.
Skills: TensorFlow, XGBoost, WarpGBM, Keras, SQL
Project partners: Groupe Mutuel
Related papers:

• Entity embedding of high-dimensional claims data for hospitalized exacerbation prediction (2024)