Publications

The Lancet Digital Health, 8(8), 101011, 2026 · Viewpoint

Beyond language: generative artificial intelligence as a general computing model for medicine

Arkadiusz Sitek and David W. Bates

A perspective on learning directly from tokenized medical data. Using ETHOS as an example, we discuss patient health timeline models as a foundation for simulating clinical trajectories and sharing models across institutions without sharing patient data.

DOI: 10.1016/j.landig.2026.101011

NEJM AI, 3(6), 2026 · Review Article

MEDS — An Emerging Data Standard and Ecosystem for Health AI Research

Matthew B. A. McDermott, Ethan Steinberg, Jason A. Fries, Robin P. van de Water, Chao Pang, Patrick Rockenschaub, Pawel Renc, Jungwoo Oh, Kamilė Stankevičiūtė, Justin Xu, Tom J. Pollard, Nassim Oufattole, Michael Wornow, Teya S. Bergamaschi, Hyewon Jeong, Simon A. Lee, Vincent Jeanselme, Kiril V. Klein, Mikkel Odgaard, Maria E. Montgomery, Arkadiusz Sitek, Mads Nielsen, Jeffrey N. Chiang, Noa Dagan, Isaac Kohane, Shalmali Joshi, Edward Choi, and Nigam H. Shah

The Medical Event Data Standard (MEDS) provides a shared, open-source foundation for health AI research. This review describes the standard, its growing ecosystem, and its role in reproducible workflows and foundation model development.

DOI: 10.1056/AIra2501253

NeurIPS 2025

Mamba Goes HoME: Hierarchical Soft Mixture-of-Experts for 3D Medical Image Segmentation

Szymon Płotka, Gizem Mert, Maciej Chrabaszcz, Ewa Szczurek, and Arkadiusz Sitek

A hierarchical mixture-of-experts architecture built on Mamba for 3D medical image segmentation. The model combines local and global context across CT, MRI, and ultrasound data.

GigaScience, Volume 14, 2025

Foundation model of electronic medical records for adaptive risk estimation

Pawel Renc, Michal K. Grzeszczyk, Nassim Oufattole, Deirdre Goode, Yugang Jia, Szymon Bieganski, Matthew B. A. McDermott, Jaroslaw Was, Anthony E. Samir, Jonathan W. Cunningham, David W. Bates, and Arkadiusz Sitek

ETHOS-ARES uses patient health timelines to estimate changing clinical risks without task-specific fine-tuning. The framework supports adaptive risk assessment from electronic health records using the MEDS data format.