JEDA: Query-Free Clinical Order Search from Ambient Dialogues

2025-10-15

AI Summary

To address the high latency, instability, and poor interpretability in identifying implicit clinical orders (e.g., “chest X-ray required”) within clinical dialogues, this paper proposes a query-free clinical order retrieval method. Instead of relying on LLM-based explicit query rewriting, we design a dual-encoder joint embedding model initialized with PubMedBERT, which implicitly models clinical intent via a sliding dialogue window. We introduce a deduplicated contrastive learning objective and train the model using multi-form synthetic annotations generated by a constrained LLM. Our approach significantly outperforms baseline methods and state-of-the-art embedding models in real-world deployment, demonstrating strong generalization, robustness to noise, and real-time inference (millisecond-level latency). It has been integrated into clinical practice, enabling accurate, low-latency, and interpretable automatic matching of clinical orders.

Technology Category

• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Large language models for search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking

Abstract

Clinical conversations mix explicit directives (order a chest X-ray) with implicit reasoning (the cough worsened overnight, we should check for pneumonia). Many systems rely on LLM rewriting, adding latency, instability, and opacity that hinder real-time ordering. We present JEDA (Joint Embedding for Direct and Ambient clinical orders), a domain-initialized bi-encoder that retrieves canonical orders directly and, in a query-free mode, encodes a short rolling window of ambient dialogue to trigger retrieval. Initialized from PubMedBERT and fine-tuned with a duplicate-safe contrastive objective, JEDA aligns heterogeneous expressions of intent to shared order concepts. Training uses constrained LLM guidance to tie each signed order to complementary formulations (command only, context only, command+context, context+reasoning), producing clearer inter-order separation, tighter query extendash order coupling, and stronger generalization. The query-free mode is noise-resilient, reducing sensitivity to disfluencies and ASR errors by conditioning on a short window rather than a single utterance. Deployed in practice, JEDA yields large gains and substantially outperforms its base encoder and recent open embedders (Linq Embed Mistral, SFR Embedding, GTE Qwen, BGE large, Embedding Gemma). The result is a fast, interpretable, LLM-free retrieval layer that links ambient context to actionable clinical orders in real time.

Problem

• Retrieving clinical orders directly from ambient dialogues without queries
• Aligning heterogeneous medical expressions to shared order concepts
• Reducing sensitivity to speech disfluencies and ASR errors

Innovation

• Bi-encoder retrieves orders from ambient dialogue
• Contrastive training aligns intent to shared concepts
• Query-free mode uses rolling window for noise resilience