Artificial Intelligence and Deep Learning
We develop and apply trustworthy AI to turn complex medical data into clinically useful evidence and decision support. Our work focuses on longitudinal electronic health records (EHRs) like that found in critical care, with an emphasis on robustness, external validation across hospitals, and privacy-preserving collaboration.
Foundation models for longitudinal EHR
A central direction is developing foundation models for EHRs that represent the patient record as a continuous stream of clinical events, including measurements, medications, procedures, documentation. This line of work is funded through an ERC Starting Grant (GPT-MEDIC). The goal is to learn reusable patient representations that support flexible prediction for many clinical outcomes, enable uncertainty-aware decision support, and improve generalisability through large-scale pre-training.
Sequential decision-making and digital twins
In close collaboration with Paul Elbers at Amsterdam UMC, we are working towards trustworthy digital twins in critical care. This includes reinforcement learning methods to optimise ventilator settings over time and predictive modelling to anticipate complications and improve decision-making.
Reproducible benchmarking
Progress in clinical AI is often slowed by incomparable cohorts, outcomes, and preprocessing. We contribute methods and infrastructure that make task definitions explicit and reusable, so models can be compared fairly and validated systematically across sites and time. To this end, we also contribute to the [Medical Event Data Standard (MEDS)](https://github.com/Medical-Event-Data-Standard/meds) to improve interoperability of event-based clinical data representations for machine learning.
Federated learning and privacy-preserving collaboration
Access to diverse hospital data often requires federated approaches, where analyses are brought to the data rather than moving patient data centrally. We contribute to the [INDICATE consortium](https://indicate-europe.eu/), which is building a federated infrastructure for intensive care data across Europe and using it to develop and evaluate trustworthy AI in real clinical environments.
Selected publications
The impact of evaluation strategy on sepsis prediction model performance metrics in intensive care data.() JMIR (in press). 2026.
Continuous Multimodal AI with Wearable Vital Signs Predicts Postoperative Complications in the General Ward.(https://doi.org/10.1101/2025.11.25.25340950) medRxiv. 2025.
External validation of AI-based scoring systems in the ICU: a systematic review and meta-analysis. (https://doi.org/10.1186/s12911-024-02830-7) BMC Medical Informatics and Decision Making. 2025.
MEDS Decentralized, Extensible Validation (MEDS-DEV) Benchmark: Establishing Reproducibility and Comparability in ML for Health. (https://openreview.net/pdf?id=DExp3tRRel) ML4H. 2024.
The impact of multi-institution datasets on the generalizability of machine learning prediction models in the ICU. (https://doi.org/10.1097/CCM.0000000000006359) Critical Care Medicine. 2024.
Yet Another ICU Benchmark: A flexible multi-center framework for clinical ML. (https://openreview.net/pdf?id=ox2ATRM90I) ICLR. 2024
Robust prediction under missingness shifts. (https://arxiv.org/abs/2406.16484) arXiv. 2024.
Last updated: Jan 2026