A Multi-Center Study on the Adaptability of a Shared Foundation Model for Electronic Health Records

Abstract

Foundation models hold promise for transforming AI in healthcare by providing modular components that are easily adaptable to downstream healthcare tasks, making AI development more scalable and cost-effective. Structured EHR foundation models, trained on coded medical records from millions of patients, demonstrated benefits including increased performance with fewer training labels, and improved robustness to distribution shifts. However, questions remain on the feasibility of sharing these models across different hospitals and their performance for local task adaptation. This multi-center study examined the adaptability of a recently released structured EHR foundation model ($F{M}_{SM}$), trained on longitudinal medical record data from 2.57M Stanford Medicine patients. Experiments were conducted using EHR data at The Hospital for Sick Children and MIMIC-IV. We assessed both adaptability via continued pretraining on local data, and task adaptability compared to baselines of training models from scratch at each site, including a local foundation model. We evaluated the performance of these models on 8 clinical prediction tasks. In both datasets, adapting the off-the-shelf $F{M}_{SM}$ matched the performance of GBM models locally trained on all data while providing a 13% improvement in settings with few task-specific training labels. With continued pretraining on local data, label efficiency substantially improved, such that $F{M}_{SM}$ required fewer than 1% of training examples to match the fully trained GBM's performance. Continued pretraining was also 60 to 90% more sample-efficient than training local foundation models from scratch. Our findings show that adapting shared EHR foundation models across hospitals provides improved prediction performance at less cost, underscoring the utility of base foundation models as modular components to streamline the development of healthcare AI.

46 pages, 5 figures, 3 tables, 14 appendices

References in corpus (9)

• Towards Expert-Level Medical Question Answering with Large Language Models
• Fine-Tuning can Distort Pretrained Features and Underperform Out-of-Distribution
• HyenaDNA: Long-Range Genomic Sequence Modeling at Single Nucleotide Resolution
• GenHPF: General Healthcare Predictive Framework with Multi-task Multi-source Learning
• EHRSHOT: An EHR Benchmark for Few-Shot Evaluation of Foundation Models
• MOTOR: A Time-To-Event Foundation Model For Structured Medical Records
• Skill-it! A Data-Driven Skills Framework for Understanding and Training Language Models
• No Fair Lunch: A Causal Perspective on Dataset Bias in Machine Learning for Medical Imaging
• Zero-Shot Robustification of Zero-Shot Models

Cited by in corpus (2)

• Emergency Department Decision Support using Clinical Pseudo-notes
• Generative AI in Health Economics and Outcomes Research: A Taxonomy of Key Definitions and Emerging Applications, an ISPOR Working Group Report