Summary
A forum post introduces BERT-LER, a BERT-style transformer model for encoding electronic health record (EHR) timelines, presented in arXiv paper 2608.20315 by Jun Ni Du, Lukas Adamek, Maxim Kryukov, and Flavio Dormont. The model was pre-trained and fine-tuned on a de-identified EHR dataset covering 75 million patients. Its key innovation is encoding lab test results as discrete tokens while preserving graded information through percentile-based binning, combined with Integrated Gradients for token-level attribution relative to input medical events. BERT-LER was evaluated on the public EHRShot benchmark and on a real-world study of asthma severity progression. Results show competitive predictive performance, often exceeding publicly available baseline models on lab-related tasks, with attributions that align with clinically known risk factors. The architecture and interpretability approach are applicable across many therapeutic areas and prediction tasks, addressing a gap in combining quantitative lab information with explainability.
Paper Overview
Research Area: Machine Learning
Authors: Jun Ni Du, Lukas Adamek, Maxim Kryukov, Flavio Dormont
Published: 2026-08-22
arXiv: 2608.20315
Summary
Prediction models on structured electronic health records (EHR) remain central to medical machine learning, yet few approaches jointly emphasize quantitative laboratory information and interpretability relative to input medical events.
This paper proposes BERT-LER, a BERT-style model for encoding EHR timelines, pre-trained and fine-tuned on a de-identified EHR dataset covering 75 million patients. It encodes lab test results as discrete tokens while preserving graded information via percentile-based binning, and pairs this with Integrated Gradients to provide token-level attributions grounded in the input EHR sequence.
Evaluation
- Benchmarked on the public EHRShot benchmark
- Applied to a real-world study of asthma severity progression
Findings
- Competitive predictive performance, often outperforming publicly available baseline models on lab-related tasks
- Attribution outputs align with clinically known risk factors
The architecture and interpretability method can be applied across many therapeutic areas and prediction tasks.
---
*Auto-collected on 2026-08-22*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178633796