Lessons Learned on Information Retrieval in Electronic Health Records: A Comparison of Embedding Models and Pooling Strategies

TL;DR

This study systematically compares various embedding models and pooling strategies to optimize information retrieval from electronic health records, revealing that model choice and query formulation critically influence retrieval effectiveness in clinical NLP tasks.

Contribution

It provides an empirical ablation analysis of embedding models and pooling methods, highlighting the superior performance of a general-domain model (BGE) in clinical retrieval tasks.

Findings

BGE outperforms medical-specific models across datasets

Pooling strategies significantly affect retrieval performance

Model and query variability impact results and transferability

Abstract

Objective: Applying large language models (LLMs) to the clinical domain is challenging due to the context-heavy nature of processing medical records. Retrieval-augmented generation (RAG) offers a solution by facilitating reasoning over large text sources. However, there are many parameters to optimize in just the retrieval system alone. This paper presents an ablation study exploring how different embedding models and pooling methods affect information retrieval for the clinical domain. Methods: Evaluating on three retrieval tasks on two electronic health record (EHR) data sources, we compared seven models, including medical- and general-domain models, specialized encoder embedding models, and off-the-shelf decoder LLMs. We also examine the choice of embedding pooling strategy for each model, independently on the query and the text to retrieve. Results: We found that the choice of embedding model significantly impacts retrieval performance, with BGE, a comparatively small general-domain model, consistently outperforming all others, including medical-specific models. However, our findings also revealed substantial variability across datasets and query text phrasings. We also determined the best pooling methods for each of these models to guide future design of retrieval systems. Discussion: The choice of embedding model, pooling strategy, and query formulation can significantly impact retrieval performance and the performance of these models on other public benchmarks does not necessarily transfer to new domains. Further studies such as this one are vital for guiding empirically-grounded development of retrieval frameworks, such as in the context of RAG, for the clinical domain.